Li Kang 2009

This document is downloaded from DR‑NTU (https://dr.ntu.edu.
sg)
Nanyang Technological University, Singapore.
Smartglove : a multi‑finger sensing system based

on optical linear encoder
Li, Kang
2009
Li, K. (2009). Smartglove : a multi‑finger sensing system based on optical linear encoder.
Master’s thesis, Nanyang Technological University, Singapore.
https://hdl.handle.net/10356/19317
https://doi.org/10.32657/10356/19317
Downloaded on 24 Mar 2024 06:46:49 SGT

ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library
MASTER OF ENGINEERING
SMARTGLOVE – A MULTI-FINGER SENSING SYSTEM

BASED ON OPTICAL LINEAR ENCODER
LI KANG
LI KANG
SCHOOL OF MECHANICAL AND AEROSPACE
2009
ENGINEERING
2009
SMARTGLOVE – A MULTI-FINGER SENSING SYSTEM

BASED ON OPTICAL LINEAR ENCODER
LI KANG
School of Mechanical and Aerospace Engineering
A thesis submitted to the Nanyang Technological University

in partial fulfillment of the requirement for the degree of
Master of Engineering
2009
ACKNOWLEDGEMENT
The author would like to express his heartfelt thanks and gratitude to the following people
and organization for their efforts and support to the successful completion of the project.
Associate Professors Chen I-Ming and Yeo Song Huat for their ever encouraging support,
guidance, patience and valuable advices. Their expertise in technical background of this
research have immensely contributed to the completion of this thesis.
Mr. Nguyen Kim Doang, Mr. Luo Zhiqiang, Mr. Lim Kwang Yong, Mr. Dong Wei, Mr.
Goh Young Koon, and Mr. Lim Wenbin, for their beneficial studies and research work on
the optical liner encoder and programming. Their roles are important in guiding me in the
fulfillment of this research project.
Staff of Robotics Research Centre: Mr Lim Eng Cheng, Technician-in-charge, for his
coordination of the available resources. Ms Agnes Tan for her guidance, support and
valuable advice. Mr You Kim San and Miss Toh Yen Mei for their valuable advices and
assistance in the test process.
My parents and my girlfriend, for their unwavering support, encouragement, and

understanding in all my works. They are always there for me.
Finally, to friends who have contributed in one way or another, thank you for your
continuous support and endless encouragement during this time.
.
TABLE OF CONTENTS
Page
ABSTRACT ................................................................................................... 1
LIST OF FIGURES ...................................................................................... 2
LIST OF TABLES ........................................................................................ 5
LIST OF ABBREVIATIONS ...................................................................... 6
CHAPTER 1. INTRODUCTION ................................................................ 7
1.1. Motivation ..................................................................................... 7
1.2. Objectives .................................................................................... 10
1.3. Scope ............................................................................................ 12
1.4. Report Organization .................................................................. 12
CHAPTER 2. LITERATURE REVIEW.................................................. 15
2.1. Glove-based Hand Mocap Technologies .................................. 17

2.1.1. Fiber-optic Sensing Technology............................................. 17
2.1.2. Strain Gauge Sensing Technology ......................................... 19
2.1.3. Hall-Effect Sensing Technology ............................................. 21
2.1.4. Other Sensing Technologies ................................................... 23
2.2. Applications ................................................................................ 25

2.2.1. Natural Human Computer Interface .................................... 25
2.2.2. Interpreting Sign Language ................................................... 26
2.2.3. Tele-operation and Control .................................................... 27
2.2.4. Rehabilitation and Training................................................... 27
2.2.5. Computer Animation .............................................................. 28
2.3. Discussion .................................................................................... 30
CHAPTER 3. BIOMECHANICS OF HUMAN HAND ......................... 31
3.1. Skeleton Model ........................................................................... 31

3.2. Constraints of Human Hand Motion ....................................... 33
3.3. Kinematic Model ........................................................................ 36

3.3.1. Coordinate System Setup ....................................................... 36
3.3.2. Link and Joint Parameters..................................................... 36
3.3.3. Transformation of the Coordinate Frames .......................... 37
3.3.4. Kinematic Model of the Hand ................................................ 39
3.4. Discussion .................................................................................... 50
CHAPTER 4. SENSING PRINCIPLE ..................................................... 51
4.1. Inverted OLE Sensing Principle ............................................... 51
4.2. Multi-point Sensing Principle ................................................... 53

4.2.1. Finger Model ........................................................................... 55
4.2.2. Thumb Model .......................................................................... 57
4.3. Discussion .................................................................................... 58
CHAPTER 5. SMARTGLOVE DEVELOPMENT ................................ 59
5.1. Hardware Design........................................................................ 60

5.1.1. OLE Module Design................................................................ 60
5.1.2. Microcontroller ....................................................................... 64
5.1.3. Glove Design ............................................................................ 66
5.2. Firmware Design ........................................................................ 68

5.2.1. Serial Communication ............................................................ 69
5.2.2. Communication Protocol ........................................................ 69
5.2.3. Flow Chart ............................................................................... 71
5.3. Application software Design ..................................................... 73
5.4. Discussion .................................................................................... 76
CHAPTER 6. CALIBRATION OF SMARTGLOVE ............................ 78
6.1. Notations ..................................................................................... 78
6.2. Data Mapping ............................................................................. 79
6.3. Calibration Model ...................................................................... 80

6.3.1. One-to-One Mapping .............................................................. 80

6.3.2. One-to-Two Mapping ............................................................. 81
6.4. Calibration Methods .................................................................. 81
6.5. Validation of Calibration Model............................................... 83
6.6. Discussion .................................................................................... 86
CHAPTER 7. EXPERIMENTAL VERIFICATION .............................. 88
7.1. OLE Characterization Test ....................................................... 88

7.1.1. Linearity Test .......................................................................... 88
7.1.2. OLE Bending Test................................................................... 90
7.1.3. Human Finger Test ................................................................. 92
7.2. Glove Testing .............................................................................. 94

7.2.1. Material and Subjects ............................................................. 95
7.2.2. Experiment Protocols and Procedure ................................... 95
7.2.3. Statistical Analysis .................................................................. 97
7.2.4. Results ...................................................................................... 98
7.3. Discussion .................................................................................. 102
CHAPTER 8. CONCLUSION & FUTURE WORK............................. 104
8.1. Contribution ............................................................................. 104
8.2. Future Work ............................................................................. 106
REFERENCES.......................................................................................... 108
APPENDIX………………………………………………………………………………115
ABSTRACT
Objective measures of human hands as individuals participate in everyday activities are

needed in order to expand the dexterous use of the hand or to evaluate the hand functions
in rehabilitation or skill training. Data gloves for measurements of finger movements are a
promising tool for this purpose. The requirements for the data glove include easy and
comfortable to wear and remove, durability, cost-effectiveness and measurement
repeatability and reliability. This thesis presents the design of a wearable glove-based
multi-finger motion capture device (SmartGlove) with a specific focus on the development
of a new optical linear encoder (OLE) with novel sensing technology. Modelling of the
full hand kinematics and constraints are introduced, working principles of the OLE and the
multi-point sensing method are illustrated. The OLE development and the SmartGlove
construction are also presented. The OLE specially designed for this project has a compact
size, light weight and low power consumption. The characterization tests also show that
the OLE’s digital output has good linearity and accuracy. The first prototype of
SmartGlove which uses ten OLEs to capture the flexion/extension motion of the 14 finger
joints is constructed based on the multi-point sensing method. A case study for the
evaluation of SmartGlove using a standard protocol shows high repeatability and
reliability in both the gripped and flat hand positions compared with another four
evaluated data gloves using the same protocol. Conclusively, measuring outcomes in a
portable manner can provide important information for the utilization and evaluation of
the hand’s motion data. Results demonstrated that SmartGlove is an important
improvement in this direction as both a research and an evaluation tool for widespread use
of hand motion capture.
Keywords:
Low cost, Wearable, Glove Based Hand Motion Capture, Hand Kinematic Model, Optical
Linear Encoder, Calibration.
-1-
LIST OF FIGURES
Figure 2-1 Optical Tracking [3] ........................................................................................ 15

Figure 2-2 Hand Mocap System Using Magnetic Sensor [4] ........................................... 16
Figure 2-3 Power Glove [5] .............................................................................................. 16
Figure 2-4 5DT Data Glove [6] ........................................................................................ 18
Figure 2-5 Wrist, Finger and Thumb Motion Capture by ShapeHand [7] ....................... 18
Figure 2-6 Measurand ShapeHand [7] .............................................................................. 19
Figure 2-7 CyberGlove II [8] ............................................................................................ 20
Figure 2-8 DG5 Vhand Data Glove [9] ............................................................................ 20
Figure 2-9 Humanglove [10] ............................................................................................ 21
Figure 2-10 Air Tube Glove [11]...................................................................................... 23
Figure 2-11 AcceleGlove [13] .......................................................................................... 24
Figure 2-12 Textile Integrated Sensing Glove [14] .......................................................... 24
Figure 2-13 Textile Integrated Sensing Glove [15] .......................................................... 25
Figure 2-14 VirtualHand for V5 (CATIA) [8] ................................................................. 26
Figure 2-15 Interpreting Sign Language Using StrainGlove [18] .................................... 26
Figure 2-16 Robotic Hand Manipulation Using CyberGlove II [19] ............................... 27
Figure 2-17 The Poared Hand Exoskeleton for Hand Rehabilitation [20] ....................... 28
Figure 2-18 Motion Editing with P5 Glove [22] .............................................................. 29
Figure 3-1 Diagram for the Human Hand Skeleton [24] .................................................. 31
Figure 3-2 Finger Flexion after Constraint Application [28] ........................................... 34
Figure 3-3 Denotation of FE & AA motion...................................................................... 35
Figure 3-4 Coordinate System and Joint Parameters ........................................................ 37
Figure 3-5 Hand’s Kinematic Model (Left Hand) [32] .................................................... 39
Figure 3-6 Thumb kinematic Model ................................................................................. 41
Figure 3-7 Index & Middle finger’s Kinematic Model .................................................... 44
Figure 3-8 Ring & Little finger’s Kinematic Model ........................................................ 46
Figure 3-9 Palm motion (Left Hand) ................................................................................ 48
Figure 3-10 Simplified Hand’s Kinematic Model (Left Hand) ........................................ 49
Figure 4-1 OLE for SmartSuit ......................................................................................... 51
Figure 4-2 OLE and Inverted OLE’s Sensing Principles ................................................. 52
Figure 4-3 Multi-point Sensing Principle ......................................................................... 54
-2-
Figure 4-4 Arches of the Hand [35] .................................................................................. 55

Figure 4-5 Finger Model ................................................................................................... 56
Figure 4-6 Thumb model .................................................................................................. 57
Figure 5-1 System Block Diagram ................................................................................... 59
Figure 5-2 OLE & Its Cross Sectional View .................................................................... 60
Figure 5-3 Avago ADNS-3530 [36] ................................................................................. 61
Figure 5-4 Avago lens ADNS-3150-001 [36] .................................................................. 61
Figure 5-5 Lens and Illumination System [36] ................................................................. 61
Figure 5-6 OLE Circuit ..................................................................................................... 62
Figure 5-7 Mean Resolution vs. z-height [36] .................................................................. 63
Figure 5-8 Housing and Strip ........................................................................................... 63
Figure 5-9 Arduino Diecimila [37] ................................................................................... 64
Figure 5-10 Arduino with Interface Board ....................................................................... 65
Figure 5-11 Glove Layout ................................................................................................ 66
Figure 5-12 Glove Prototype ............................................................................................ 67
Figure 5-13 Firmware Architecture .................................................................................. 68
Figure 5-14 SPI Bus: One Master with Three Slaves ....................................................... 70
Figure 5-15 Firmware Flow Chart .................................................................................... 72
Figure 5-16 Application Software Structure..................................................................... 73
Figure 5-17 Hand Avatar .................................................................................................. 74
Figure 5-18 Graphical User Interface ............................................................................... 74
Figure 5-19 Real-time Data Display and Hand Simulation .............................................. 75
Figure 5-20 System Reset ................................................................................................. 76
Figure 5-21 Recorded Data Checking .............................................................................. 76
Figure 6-1 Simplified Hand Skeleton Notational Diagram (Left Hand) .......................... 79
Figure 6-2 Calibration Postures [43] ................................................................................ 82
Figure 6-3 Calibration Block ............................................................................................ 82
Figure 6-4 Calibration Block User Guide ......................................................................... 83
Figure 6-5 Measure Joint Angle by Protractor ................................................................. 83
Figure 6-6 Posture1, Posture3 and Posture4 ..................................................................... 84
Figure 6-7 Calibration Test Result (MCP Joint)............................................................... 86
Figure 6-8 Calibration Test Result (PIP Joint) ................................................................. 86
Figure 7-1 Linearity Test .................................................................................................. 88
Figure 7-2 Linearity Test Result ....................................................................................... 89
-3-
Figure 7-3 Linearity Test Result (Average)...................................................................... 89

Figure 7-4 OLE Bending Test .......................................................................................... 90
Figure 7-5 OLE Bending Test Result ............................................................................... 91
Figure 7-6 Bending Test Result (Average) ....................................................................... 91
Figure 7-7 Human Finger Test ......................................................................................... 93
Figure 7-8 Human Finger Test Result .............................................................................. 94
Figure 7-9 Cylindrical Grip Measurement ....................................................................... 96
Figure 7-10 Flat Hand Measurement ................................................................................ 96
Figure 7-11 Averages of Ten Data Block for a Single Subject for the Grip Test ............ 98
Figure 7-12 Sample Data Block (Grip Test)..................................................................... 99
Figure 7-13 Histogram of Averaged Rang and SD for Each Subject and Each Test ..... 100
Figure 7-14 Histogram of Averaged Range and SD for Each OLE ............................... 101
-4-
LIST OF TABLES
Table 2-1 Products’ Comparison Chart ............................................................................ 22

Table 3-1 Thumb’s D-H Model Parameters ..................................................................... 41
Table 3-2 Index & Middle Finger’s D-H Model Parameters ........................................... 44
Table 3-3 Index & Middle Finger’s D-H Model Parameters ........................................... 47
Table 5-1 Data Block Structure ........................................................................................ 71
Table 5-2 Technical Specifications of OLE ..................................................................... 77
Table 6-1 Joint/Encoder Relationships ............................................................................. 79
Table 6-2 Calibration Verification Data ........................................................................... 84
Table 6-3 Calibration Methods Comparison .................................................................... 85
Table 7-1 Averaged Range & SD for Each Subject and Each Test................................ 100
Table 7-2 Averaged Range & SD for Each OLE ........................................................... 101
Table 7-3 Intraclass Correlation Coefficient of Reliability ............................................ 102
Table 7-4 Comparison of Repeatability Results ............................................................. 103
-5-
LIST OF ABBREVIATIONS
Mocap Motion Capture

DOF Degree Of Freedom
ROM Range Of Motion
DIP Distal Interphalangeal
PIP Proximal Interphalangeal
MCP Metacarpophalangeal
TM Trapeziometacarpal
HM Humate-Metacarpal
CMC Carpometacarpal
IP Interphalangeal
FE Flexion/Extension motion
AA Abduction/Adduction motion
SD Standard Deviation
ICC Intraclass Correlation Coefficient
OLE Optical Linear Encoder
SCLK Serial Clock
MOSI Master Output, Slave Input
MISO Master Input, Slave Output
CS Chip Select
-6-
Chapter 1. Introduction
CHAPTER 1. INTRODUCTION
1.1. Motivation
As the intricate and prehensile parts in the body, human hands are our primary physical
interaction with the world. We use our hands in varies of aspects to perform our
everyday activities, such as grasping, manipulating, drawing, writing and sculpting.
Sometimes, we even use our hands as a communication tool in sign language or finger
spelling.
With the rapid development of computer technology, we gradually tend to use

computers to work, learn and entertain. However, we always found it unnatural when
we work with a computer or use computer-based applications because we are
constrained by clumsy input devices such as keyboards and mice. In order to be free
from the limitations of the input devices, researchers have been studying and
developing methods of getting computers able to “read” and “understand” user’s hand
gestures or motions directly. The development of hand motion capture device has been
a significant move in this direction that expands the natural use of our hands from
physical world to the virtual world.
Motion capture or motion tracking technology started in the 1970s as a

photogrammetric analysis method in biomechanics research. Later, with the matureness
of the technology, it was expanded into other areas like rehabilitation, training and
entertainment. Motion capture is the process of capturing the posture, location and
motion information of a subject over a certain period of time. The subject can varies
from a whole human body to a single finger joint based on different applications. From
the motion capture it is able to obtain the posture and motion data of the subject. Other
parameters of the motion (such as displacement, angle, velocity, acceleration) can be
calculated from these captured raw motion data later. The captured and calculated data
can be used in certain theoretical analyses of the human’s motion. Also, these data can
be utilized in either virtual or physical environment to control or drive some objects.
-7-
For different applications, there are different requirements for the motion capture
device and the captured motion data. In case the applications are animation, gesture
recognition, rehabilitation and interactive games which do not require high accuracy in
motion capture, the captured data should be stable and the device should be low-cost,
comfortable to wear, and easy to use. In case the applications are motion analysis,
biomechanics and biodynamics analysis, virtual training which have higher
requirements for data in order to get accurate and precise results, the data should be high
in both accuracy and resolution, and is better to have other sensors working together to
get extra information such as forces or orientations. Also the device should be small and
light so that the movements of the subject will not be affected. In case the applications
are tele-operation, motion feedback control, besides the stability and accuracy, the data
should also have high data rate in order to be real-time. The device should also contain
force feedback to increase the intuition of use.
For the last few decades, several motion capture methods have been developed with
different characteristics. [1]
1) Mechanical sensing
In the typical mechanical sensing, several rigid mechanical pieces and
electromechanical transducers (potentiometers or shaft encoders) are placed on the
object. When the object moves, the mechanical pieces change shape and the
electromechanical transducers move accordingly. Using the previous knowledge about
the rigid mechanical pieces and the measurements of the transducers, the object’s
position can be estimated with respect to the environment. Mechanical sensing
approach can provide very precise posture estimation for a single object, but only over a
relatively small range of motion due to the restriction of the mechanical structure.
Mechanical motion capture systems are usually real-time, relatively low-cost and
non-occlusion.
2) Inertial sensing
Inertial motion capture technology is an effective way for full-body motion capture
which is based on the combination of inertial sensing unit, kinematical models and
sensor fusion algorithms. The motion data of the inertial sensors is usually transmitted
wirelessly to a computer. Specially designed software in the computer can save these
-8-
data and replay the motion in a 3D virtual environment. Inertial motion capture systems
are capable of capturing the full six degree of freedom (DOF) in real-time without
external cameras or markers. Inertial sensors are completely self-contained and have
low latency and high rates. They are also not sensitive to the electromagnetic fields or
ambient noise interferences. The weakness that prevents inertial trackers from being an
ideal choice is drift.
3) Acoustic sensing
The working principle of acoustic sensing device is using high-frequency sound to
triangulate a source within the work area. There cannot be hard obstacles or other
acoustically reflective surface between the source and the surrounding microphones
because the acoustic sensing systems can suffer from acoustic reflections. Acoustic
sensing systems work well with large motions, but tend to have difficulties with fingers,
faces, wrist rotations or small motions because of the system’s resolution.
4) Magnetic sensing
Magnetic sensing systems measure the local magnetic field vector at the sensor. When a
changing magnetic field passes through a coil, the system can measure the current
induced in the electromagnetic coil. In a single inertial sensing unit there are three
orthogonally oriented magnetic sensors. The 3D vector indication of the unit’s
orientation can be determined by the three sensors with respect to the excitation source.
The magnetic sensing has the advantages of compact size, penetrable magnetic fields,
and only a single source needed for multiple sensors. It also has the disadvantages of the
distortion and strength reduction of the magnetic field.
5) Optical sensing
Optical sensing systems are usually comprised with two components: light source and
optical sensors. The light sources can either be the active devices that emit internally
generated light (LED) or the passive objects that reflect ambient light (marker). The
optical sensors can detect the light from the source and determine the position and
motion of the light source. The main disadvantage of optical systems is that there
should not be other light source or reflective obstacles between the source and the
sensor. However, the optical systems, especially the marker-based camera systems can
produce the most realistic motions with enough markers on the object.
-9-
The five sensing technologies mentioned above are mainly focus on the whole body
motion capture. However, when it comes to the hand motion capture, it is another
different scenario. In the field of hand motion capture, there are several long standing
problems and difficulties need to be addressed.
The primary difficulty is that the hand is relatively smaller but more articulated than the
whole body. There are totally 27 bones and 23 DOFs (excluding the wrist) within one
hand which offers dexterity for the hand to perform most of our everyday tasks. In order
to cope with the multi-DOF structure, kinematic modeling becomes an important factor,
especially the modeling of the palm. The way of modeling determines the placement of
sensors, and eventually determines the naturalness and integrity of the whole hand
motion capture.
Another problem is the method for hand motion capture. Because the hand’s bones and
joints are relatively smaller than other parts of the body (such as arm, leg and trunk),
several body motion capture technologies may not be suitable for the hand. The sensor
for hand motion capture should be small and light so that it can be placed on the palm
and fingers easily. Meanwhile, it should not occlude the natural movement of the hand.
Given the importance and variety potential application areas of hand motion capture, as
well as the existing problems and difficulties, the purpose of this research is to develop
a low cost, high performance, glove-based, un-tethered wearable human hand motion
capture device with novel sensing technology and ergonomics, as well as the processing
system for real-time sensing and processing of the hand’s anatomical motion data. The
captured motion data can be further processed to be used in numerous areas such as
rehabilitation, virtual input, interactive media and games, real-time tele-control and
manipulation.
1.2. Objectives
In order to fulfill the requirements of multi-finger motion capture, several objectives

need to be accomplished in the thesis project.
- 10 -
(1) Kinematic modeling of realistic human hand anatomy.

The kinematic model should be a realistic model of the hand, including the fingers
and the palm. Anatomical studies on human hand show that the hand is a highly
articulated part with 27 bones and up to 23 DOFs. Meanwhile, it is also highly
constrained with constraints of the joint’s motion range and coupling between joint
motions. Based on the structure and constraints, mathematical formulations can be
developed using a systematic method for representing constrained large DOF
models of human hand. With the kinematic model, the layout of OLEs and the
construction of a realistic 3D hand model can be obtained.
(2) Development of a new encoder with novel sensing technology suitable for hand
motion capture.
Several sensing technologies used for human body motion capture are mentioned
above. However, because of the small size of hand joints, some of the sensing
technologies may not be able to implement in the hand motion capture.
Consequently, a new encoder with novel sensing technology needs to be developed
in this project. The encoder should meet the requirements of hand motion capture
with small size, light weight, high resolution, high rate and low cost.
(3) Modeling the encoders on each finger based on the hand’s kinematic model.
With the hand’s kinematic model and the working principle of the encoder, it is
possible to arrange the encoders on each finger to capture each joint’s motion. The
layout of encoders should be able to capture every joints’ movements efficiently,
and do not affect the natural movement of the hand at the same time. This model can
be used to match the encoder’s digital reading with the real joint’s angle.
(4) Calibration methods and verification.

Calibration is critical to the hand motion capture system because different people
have different hand biometric data. In order to get accurate motion data, calibration
process is necessary before capture the hand motion on different subjects. Based on
the tradeoff between the accuracy and time, two calibration protocols will be
adopted (one is gesture-based protocol for rough calibration with less time; the
- 11 -
other one is tool-based protocol for accurate calibration with more time, a
calibration tool will be designed to assist the accurate calibration).
(5) Experimental tests and analysis.

A set of experimental tests will be carried out based on the standard protocols to
characterize the encoder and evaluate the performance of and the whole hand
motion capture system. The results will be analyzed with comparison to those from
other hand motion capture devices using the same test protocols.
1.3. Scope
The scope the research project will cover the following:

- Analyses of the human hand bio-mechatronics.
- Development of a new linear encoder using novel multi-point sensing method to
detect the motions of finger joints.
- Modeling of the sensors on each finger and thumb on the human hand.
- System development, including hardware, software and application design.
- Calibration methods design.
- Experimental verification and validation of the characteristics of the encoders, as
well as the performance of the SmartGlove.
1.4. Report Organization
This thesis presents the setup of the full hand kinematic model, the flow of the
development of a multi-finger sensing device (both hardware and software), and also
some experimental tests and results.
Chapter 2 is the literature review of existing hand motion capture devices and their
applications. Because of the different requirements between the body motion capture
and the hand motion capture, the sensing technologies are also different. In the literature
review, three major sensing technologies for hand motion capture are introduced and
illustrated with several commercialized hand motion capture devices. Also, some other
experimental prototypes with different novel sensing technologies are also presented. In
- 12 -
the application part, five typical applications of the hand motion capture device are
introduced. Some applications are applied in the physical world such as rehabilitation,
training and tele-operation, and others are applied in the virtual world such as computer
animation and interpreting sign language. All of the applications show the benefits that
hand motion capture device brings.
Chapter 3 introduces the skeleton, constraints, and kinematic modeling of the human
hand. Twenty three major DOFs are derived from the skeleton model, including five
DOFs in the thumb, four DOFs for each of the four fingers and another two DOFs in the
palm. An effective full hand kinematic model based on these twenty three DOFs is
defined using Denavit-Hartenberg (D-H) transformation to illustrate the realistic
motions of the hand. The proposed kinematic model is able to represent the transverse
deformation of the palm arch, which brings a significant improvement of hand posture
and motion representation with comparison to the normal kinematic models without
palm arch. Based on the different structures, three different kinematic models are setup
respectively for the thumb, the index and middle finger, and the ring and little finger, in
which the joints and links are defined by D-H notation. Three kinds of constraints are
also studied in this chapter which is able to reduce the 23 DOFs to 17 DOFs in order to
reduce the number of OLEs used.
Chapter 4 describes the sensing principle of the OLE. Different from any of the sensing
technologies introduced in the literature review, a new sensing principle using optical
mouse sensor to measure a finger joint’s flexion is illustrated. It is an improved OLE
sensing method which is able to use multiple OLEs to capture the displacement of
different points on the same strip. The advantage of multi-point sensing method is that it
can be used in the situation where the bend angles of several joints in a line need to be
captured (such as fingers or even the spine), and the number of joints is not limited as
long as they are located in the same straight line.
Chapter 5 illustrates the development of the SmartGlove. Ten OLEs are used to capture
the FE motion of ten joints based on the multi-sensing method. The OLE has a very
compact design and use Velcro to achieve easily attachable to or de-attachable from the
glove. Two OLEs are chained by a strip with rough surface to measure one finger’s
flexion/extension motion using the multi-sensing method and five such multi-OLE
- 13 -
chains are placed on the hand’s five natural longitudinal arches. The OLEs
communicate with the microcontroller (Arduino Diecimila) using Serial Peripheral
Interface (SPI) protocol and the digital data from the ten OLEs are arranged in a certain
defined format. In order to be intuitive, a software application is designed in the
OpenGL environment where a 3D virtual hand is displayed and controlled by the
SmartGlove directly. A Graphical User Interface (GUI) with basic control functions, as
well as real-time data display and store, is also designed to provide easy-of-use for the
users.
Chapter 6 focuses on the calibration methods of the SmartGlove. Two calibration

methods are proposed based on this demand. For fast calibration, a protocol with four
standard postures is able to calibrate all of the ten OLEs roughly. For precise calibration,
a special calibration block is designed to calibrate each OLE one by one. The linear
regression method is adopted to calibrate the OLE’s reading to joint angle. The results
of the calibration tests proved that both methods can improve the accuracy, especially
the second method.
Chapter 7 presents the experimental tests and results. Three tests are carried out to
characterize the OLE, including the linearity of the OLE’s reading in both flat and
bending conditions, and the accuracy of the OLE comparing to an accelerometer. Two
tests based on the standard protocols are also conducted to evaluate the performance of
the SmartGlove, including the repeatability, reliability. The evaluation is done by
comparing the result of the SmartGlove with the results from the previous four studies
using the same standard protocols.
Chapter 8 states the conclusion and recommendation for further improvement. Four
major contributions of this project are concluded, including the full hand kinematic
model, the development of a new OLE, the design of the novel multi-sensing method,
and the tool-based effective calibration method. However, the works done so far are
only the preliminary in this project, more efforts need to be done in several aspects in
the future work in order to develop a full hand motion capture device.
Additionally, the dimensions of mechanical design, the specifications of sensors, and

the layout of PCB boards are attracted as Appendix at the end of the report.
- 14 -
Chapter 2. Literature Review
CHAPTER 2. LITERATURE REVIEW
Based on the different methods, hand motion capture systems fall into two categories:
Outside-In method and Inside-In method. [2]
Outside-In method employs external sensors that sense sources or markers on the object
(such as a video camera based system that tracks the markers on the body). Based on the
different sources, Outside-In systems can be classified as the optical tracking system, the
magnetic tracking system, and the acoustic tracking system.
Optical tracking, as the Vicon 6-camero passive motion capture system [3] in Figure 2-1, is
to attach small markers on the hand. A series of six cameras are setup surrounding the hand
and pick out the markers in their visual field. Software correlates the marker’s position in
the multiple viewpoints and uses the data from different cameras to calculate the 3D
coordinate for each marker.
Figure 2-1 Optical Tracking [3]
Magnetic tracking method uses a source element radiating a magnetic field and a series of
small sensors that report their positions and orientations with respect to the source. One
example is the hand motion capture system using magnetic 3D position sensor to capture
the hand movements when playing a piano, as shown in Figure 2-2. [4]
- 15 -
Figure 2-2 Hand Mocap System Using Magnetic Sensor [4]
Similarly, acoustic tracking method uses high-frequency sound to triangulate a source

within the work area. Most systems, like the one used in the Mattel Power Glove (see
Figure 2-3), send out high-frequency sound from the source (mounted on the hand, for
instance) and received by a series of surrounding microphones. [5]
Figure 2-3 Power Glove [5]
These systems can provide 3D information about the hand and finger’s gesture and motion,
but they suffer from some general disadvantages:
 Because of the small and complex structure of the hand, the resolution is quite low.
 User’s activities are limited within the preset work area.
 Fingers are difficult to track, as they occlude each other and are occluded by hand.
It is also difficult to put enough markers on hand and fingers appropriately.
 The cost for a whole system is too high to be commonly used.
 The equipment setup, data collection and analysis demands users with high
standards.
- 16 -
For these reasons, researchers have turned to the Inside-In method, which place both the
sensors and the sources on the object for practical monitoring of the hand motion. The
sensors are usually small and easy to wear. Also, it will not be restrained in the preset work
area and relatively cheaper than the above mentioned Outside-In systems. Because the
Inside-In method always attaches sensors and sources on gloves, systems based on this
method are also known as glove-based hand mocap systems.
2.1. Glove-based Hand Mocap Technologies
Glove-based hand Mocap devices measure the shape of the hand as the fingers bend.
Over the past few decades, many researchers have built hand movement and gesture
measuring devices for computer input. Based on the sensing technology, these existing
products can be mainly classified into four categories: fiber-optic sensing glove, strain
gauge sensing glove, Hall-Effect sensing glove and other sensing glove.
2.1.1. Fiber-optic Sensing Technology
A light source (such as the LED) and a light detector are required for the fiber-optic
sensing. The amount of bending angle is proportional to the attenuation of the
detected light in specially treated sections of glass fiber that passes over the top of the
knuckle. The advantage of fiber-optic sensing is that the number of the sensing
component can be minimized because a bunch of specially designed optical fibers are
able to tracking the whole finger’s motion. However, complexity of the optical fiber’s
sectional design and glove construction, as well as the high price becomes its
disadvantages.
There are several popular products based on fibre-optic sensing. For example, Fifth
Dimension Technologies developed a 5-sensor and a 16-sensor version wireless data
glove which transmits data to a nearby computer [6]. As shown in Figure 2-4, two
joints per finger (PIP and MCP) are captured by fibre optical sensors. Optical fibers
and wires are sewed into the glove which makes the glove looks not so bulky;
however, it also adds difficulty to replace the sensor. Besides, the glove is not easy to
wear because of the tightness.
- 17 -
Figure 2-4 5DT Data Glove [6]
Figure 2-5 Wrist, Finger and Thumb Motion Capture by ShapeHand [7]
Another popular product, Shape Hand, from Measurand (uses five flex optical fiber
ribbons to capture whole hand motion (Figure 2-5). [7]
Different from the 5DT Data Glove’s embedded design, the unique design of the
easily attachable/de-attachable flexible sensor makes the Shape Hand very easy to use
(Figure 2-6). One set of ribbon sensors can fit to gloves with different sizes for both
hands. However, the cost is too high to be commonly used.
- 18 -
Figure 2-6 Measurand ShapeHand [7]
2.1.2. Strain Gauge Sensing Technology
Strain gauge is also known as resistive bend sensor. Because of its light weight and
low cost, strain gauge becomes another choice for hand motion capture. Bend sensors
are thin flexible foils that changes resistance when bent. The change of the bend angle
is generally associated with the change of measured resistance of the strain gauge.
Resistive bend sensor is popular for measuring small changes in shape which makes it
suitable for accurate measurement of finger posture. However, it is easily to have a
time-varying creep behavior when held in a fixed bend position which could reduce
the accuracy of measurement. Also, it is very sensitive to the change of temperature.
- 19 -
Figure 2-7 CyberGlove II [8]
Figure 2-8 DG5 Vhand Data Glove [9]
One famous example using strain gauge sensing method is CyberGlove from
Immersion Corporation (Figure 2-7) which contains 5 to 22 strain gauges to measure
human finger movements [8]. This glove can interface with computer via Bluetooth.
CyberGlove is one of the most successful glove products on market because of its
good design as well as satisfactory performance. However, because the resistive
sensor is extremely thin, it is also quite delicate and irreplaceable. What’s more, the
high cost will also be a disadvantage for more widely usage.
Similar to CyberGlove, the DG5 Vhand data glove from DGTech (Figure 2-8) uses
five embedded bend sensors to measure the finger movements, with additional 3-axis
accelerometer senses both the hand movement and the hand orientation (roll and
pitch). The glove communicates with external devices via RS-232 and wireless [9].
Because of the combination use of bend sensors and accelerometers, the DG5 Vhand
data glove can be used in different applications like robotics, motion capture, virtual
reality, innovative games, rehabilitation, and also as an innovative aid for disabled
- 20 -
people. However, the resolution and accuracy is not satisfactory because it uses only
five bend sensors to capture five fingers’ motion.
2.1.3. Hall-Effect Sensing Technology
Hall-Effect sensors can be configured as proximity sensors to produce a linear output

by detecting the magnetic fields. The output is proportional to the distance from a
magnetic source. Several sensors are placed on the back of a glove in a predefined
pattern. The finger joint angle can be computed from the changing field strengths
measured by these sensors. Though the Hall-Effect sensors are small, the whole glove
system can be bulky because an additional magnetic source is needed. Another
disadvantage is the interference from other electromagnetic sources.
Figure 2-9 Humanglove [10]
The HumangloveTM (Humanware S.R.L., Pisa, Italy) is a flexible glove with 20

Hall-Effect sensors to measure finger joints’ bend angles (Figure 2-9). [10] Hall
Effect sensors can provide linear response and an excellent degree of reliability and
robustness. However, as mentioned in the working principle of the Hall-Effect sensor,
the additional external source and the electromagnetic interference will be the major
disadvantages.
Table 2-1 is a comparison of the specifications of the above introduced five popular
glove-based hand Mocap products:
- 21 -
Table 2-1 Products’ Comparison Chart
5DT Data Measurand CyberGlove

DG Tech Humanware
Comparisons Glove 14 ShapeHand II 22 sensor
Vhand Humanglove
Ultra III model
Price(US$) 9,000 11,000 10,000 5,000 7,000
Interface USB USB USB USB USB
Number of
14 5 22 6 22
sensors
Cross
platform
SDK for
Operating Windows,
Windows Windows Windows Windows
system Linux and
UNIX
operating
systems
Sensor data
75 n/a 90 25 100
rate(Hz)
Sensor
0.11 0.5 0.5 0.5 0.2
resolution ( °)
Strain Gauge
Type Fibre Strain Hall Effect
Fibre optic &
of sensor optic Gauge Sensor
Accelerometer
Sensor
No Yes No No No
removable
Wireless Ethernet Ethernet Bluetooth Bluetooth Bluetooth
Operation
20 20 9 9 9
range(m)
Alias
Alias
MOCAP,
MOCAP, Motion
XSI,
XSI, Builder,
Motion 3D Studio
Software Motion CATIA V5, n/a
Builder, MAX, MAYA
Builder, VirtualHand
Character
Character SDK
Studio,
Studio, etc
etc
- 22 -
2.1.4. Other Sensing Technologies
Beside the glove products above, there are also several research glove prototypes
which are reported with no detailed experimental results.
Pressure sensing electronics are used to track finger flexion by measuring the pressure
changes in airtight PVC tubes that are mounted along the fingers (as in Figure 2-10)
[11]. Similar trial has been done by using the air tube and pressure sensor which are
tiny, light, and easy to attach to glove [12], however, the resolution and accuracy is
not satisfactory.
Figure 2-10 Air Tube Glove [11]
Another example adopts six dual-axis accelerometers to report their positions with
respect to the gravitational vector [13]. As shown in Figure 2-11, accelerometers are
placed on the back of the middle phalanges of the four fingers, on the back of the
distal phalange of the thumb, and on the back of the palm. The aim of this research is
to use the AcceleGlove as a virtual ASL keyboard because the performance, both
accuracy and response, is not good enough for hand motion tracking. However,
adopting accelerometers is a good idea for 3D-space hand position and gesture
tracking.
- 23 -
Figure 2-11 AcceleGlove [13]
Figure 2-12 Textile Integrated Sensing Glove [14]
The Textile Integrated Sensing Glove is realized by directly printing Electrically

Conductive Elastomer (CE) material on a Lycra/cotton fabric glove (Figure 2-12) [14].
The CE composites showed piezoresistive properties when a deformation is applied.
This material represented the possibility of integration in textiles which could
minimize the size of the glove, although disadvantages include extreme sensitivity to
small changes in temperature, and a moderately slow response time.
Variable resistors sensors are also used to design and implement the DHM Glove [15].
Different from Strain Gauge Sensing, the DHM Glove had a mechanical exoskeleton
structure which made it robust, but not easy to wear and remove (Figure 2-13).
- 24 -
Figure 2-13 Textile Integrated Sensing Glove [15]
2.2. Applications
With the availability of commercialized glove-based hand mocap devices and various
research prototypes described above, researches using the hand for computer input, as
well as other relevant applications, have blossomed. These projects can be categorized
into five different application areas as 1) natural interfaces, 2) systems for
understanding signed languages, 3) tele-operation and robotic control, 4) rehabilitation,
and 5) computer animation.
2.2.1. Natural Human Computer Interface
Since human interact with the physical world mostly and naturally by hands, there is a
great desire to transfer the skills, dexterity, and intuition of the hand directly to the
human-computer interface. [16] Many research projects have dealt with this subject
and most of the work has been done in the area of virtual reality. In the following
application, the hand Mocap glove is used as a master device for a 3D graphical hand
in the virtual environment. The user could grab and move objects with the 3D virtual
hand (Figure 2-14 [8]) in the virtual environment. Also, the users can use finger
postures and motions to select from the on-screen menus. The advantage of this kind
of interaction over the interaction with a 3D joystick is the intuition – user’s actions
correlate with those that might be performed on physical objects.
- 25 -
Figure 2-14 VirtualHand for V5 (CATIA) [8]
2.2.2. Interpreting Sign Language
Another interesting application of glove-based devices is the interpretation of sigh

languages for human communication, as well as computer input and control. Many
researchers have investigated different way of recognizing hand signs from simple
finger spelling to analysis of the sign language. The basic concept of hand sign
language is to analyze the hand-space-DOF vector for each posture or gesture and to
match it with a landmark hand-space vector representing the target posture or gesture
within error tolerances weighted by the significance of each degree of freedom. [17]
One example called StrainGlove is shown in Figure 2-15. [18]
Figure 2-15 Interpreting Sign Language Using StrainGlove [18]
- 26 -
2.2.3. Tele-operation and Control
Glove-based devices used in tele-operation and robotic control are important for
dexterous control of the remote end device. Researchers constructed algebraic
transformation matrices to map the human hand poses to the robot hand poses. The
kinematic differences between the human hand (as measured by the data glove) and
the robot hand are compensated by the transformation matrices. The user can control
the robot hand by mimicking the desired poses. Figure 2-16 is an example of using
CyberGlove II as a master device to manipulate the slave robotic hand [19].
Figure 2-16 Robotic Hand Manipulation Using CyberGlove II [19]
2.2.4. Rehabilitation and Training
Virtual Reality technology showed its potential benefit as a therapeutic tool for
rehabilitation in the past few decades. This technology provides the capability to
create a virtual environment where the rehabilitation and training can be
systematically monitored, manipulated and enhanced in order to create the most
appropriate, individualized motor learning paradigm. Similar to the computer games,
VR rehabilitation exercises can be designed to be intuitive and engaging which are
important to the patient motivation. The systems can also be used to monitor, record
and fully quantify the progress made by the patient, especially in terms of motor
learning improvement. [16] The Poared Hand Exoskeleton, as shown in Figure 2-17,
is a practical device for hand rehabilitation purpose [20]. The actuation is performed
- 27 -
by motors which transmit forces to the joints through cables. A real-time

microcontroller calculates the appropriate control signals for the motor controllers
based on the measured forces and joint angles. An interface computer allows the
therapist to monitor the progress, change the control modes, define new exercises, and
supervise the rehabilitation.
Figure 2-17 The Poared Hand Exoskeleton for Hand Rehabilitation [20]
2.2.5. Computer Animation
Two techniques are mostly used in computer animation. One is the key-frame
technique which is similar to the conventional hand animation. The animation is
produced by generating the frames between keys positions and postures and usually
looks very unnatural. The other one Programmed animation yields motion that is
occasionally life-like, but often too regular to be a product of life itself. To inject life
into computer animation, and as a way to overcome the trade-off between animation
time and motion quality, production companies have turned to body motion capture
for computer animation of characters. Putting a performer in direct interactive control
of a character, as in puppetry, or capturing body motion for later processing, translates
the nuances of natural motion to computer characters, making them seem very much
alive. [21] One interesting application that edits captured hand motion data by using
the data glove is illustrated as follows: The animator first wears a glove and mimics
the human body motion using his hand. Then, a mapping function that converts the
- 28 -
motion of the hand to that of the whole body will be generated (Figure 2-18). Finally,
by moving the hand in a slightly different way, a new motion with different taste will
be generated [22].
Figure 2-18 Motion Editing with P5 Glove [22]
- 29 -
2.3. Discussion
From the literature review of various glove-based hand motion capture devices with
different sensing technologies, it is obvious that the existing glove systems do have
shortcomings. First, most glove systems are very expensive. From the survey of the
commercialized products it can be found that the prices of such devices all cost over
$1,000, which enormously limits the usage and its application. Second, custom sizing
of different users is required to reduce tracking errors because measurement errors are
decreased when gloves fit snugly according to [23]. Additionally, actions such as wrist
movement in full fabric gloves may cause the glove to move over the skin. The friction
between the glove and the skin may prevent the material from returning to the original
position exactly, leaving the sensors located at different positions over the joints and
causing measurement drift. Third, the environment proposed for the glove system to use
is also very restrictive. Due to the disadvantage of the sensor, some glove systems
cannot produce accurate result under certain circumstances. For example,
electromagnetic interference can affect the Hall-Effect sensor; results from strain gauge
sensing vary with the change of temperature. Lastly, in most glove systems, sensing
units and wires are integrated in the glove and that they are always delicate.
Consequently, it is difficult or even impossible to replace the sensor for maintenance.
- 30 -
Chapter 3. Biomechanics of Human Hand
CHAPTER 3. BIOMECHANICS OF HUMAN HAND
Before starting to capture hand motion, it is necessary to understand the biomechanics of

the hand. In this chapter, the modeling of human hand’s skeleton, constraints, and
kinematics will be discussed.
3.1. Skeleton Model
Human hand is the most articulated part in the human body. As shown in the Figure 3-1,
the hand consists of a palm (metacarpus), four fingers and one thumb. It is attached to
the forearm by the wrist (carpus).
Figure 3-1 Diagram for the Human Hand Skeleton [24]
There are a total of 27 bones in a hand: the carpus (or wrist) includes eight bones which
are arranged in two rows; the metacarpus (or palm) contains five bones; the remaining
- 31 -
14 are phalanx bones with three in each of the four fingers (index finger, middle finger,
ring finger, and little finger) which are the distal phalanx, the intermediate (or middle)
phalanx, and the proximal phalanx, and two in the thumb (no middle phalanx). [24]
Totally 23 internal degrees of freedoms (DOF) are located in the above hand skeleton
model based on anatomical and medical analysis of the hand by previous studies and
researches. [25][26]
Each of the four fingers has four DOFs. The Distal Interphalangeal (DIP) joint and the
Proximal Interphalangeal (PIP) joint both have one DOF and the remaining two DOFs
are located at the Metacarpophalangeal (MCP) joint as shown in Figure 3-1.
Different from the four fingers, the thumb has five DOFs. There are two DOFs at the
Trapeziometacarpal (TM) joint (also referred as Carpometacarpal (CMC) joint), and
two DOFs at the Metacarpophalangeal (MP) joint. The remaining one DOF of the
thumb is located at the Interphalangeal (IP) joint as shown in Figure 3-1.
The basic flexion/extension and abduction/adduction motions of the thumb and fingers
are performed by the articulation of the above mentioned 21 DOFs. The flexion and
extension motions are used to describe rotations toward and away from the palm which
occurred at every joint within the hand. The abduction motion is the movement of
separation (e.g., spreading fingers apart) and the adduction motion is the movement of
approximation (e.g., folding fingers together). The abduction/adduction motions only
occur at each finger’s MCP joint as well as thumb’s MCP and TM joints.
Another two internal DOFs are located at the base of the 4th and 5th (ring and little
finger’s) metacarpals (as HM joint in Figure 3-1) which performs the curve or fold
actions of the palm.
In addition to the 23 internal DOFs in the hand, the wrist also has another three DOFs
which determine the overall orientation and rotation in space of the entire hand. [27]
- 32 -
3.2. Constraints of Human Hand Motion
As mentioned above, the human hand is highly articulated with up to 23 internal DOFs.
Meanwhile, it is also a highly constrained body part with couplings among fingers and
joints. By applying those constraints, the number of DOFs in the hand can be reduced
from 23 to 17, making the human hand motion capture more cost-efficient. Besides, the
application of the hand motion constraints is able to synthesize natural hand motion in
order to produce realistic hand animation.
Although applying human hand motion constraints could greatly reduce the number of
DOFs in a human hand, too many or too complicated constraints would also add to
computational complexity. Hence, it becomes an important issue of determining the
appropriate constraints to be implemented in the hand kinematics. In this project, three
common constraints are applied: the constraints of joints within the same finger (Type I
constraints), constraints of joints between different fingers (Type II constraints), and the
maximum range of the joint motions (Type III constraints). The major reason to adopt
these three types of constraints is that all these have been studied and used in many
previous works [28,29]. Mathematically these constraints can be presented as either
equalities or inequalities of joint angles.
1) Type I constraints [28]
This type of constraints refers to the couplings of joints within the same finger, also
called interfinger constraints. A commonly used one based on the hand anatomy states
that for the index, middle, ring and little fingers, in order to bend the DIP joints, the
corresponding PIP joints must also be bent (as shown in Figure 3-2).
The relations can be approximately presented as follows:

2
 DIP   PIP
3 (3-1)
Where  DIP refers to the flexion angle of the DIP joint and  PIP refers to the flexion
angle of the PIP joint (Figure 3-2).
By using this relationship, four DOFs are reduced in a human hand.
- 33 -
 PIP
 DIP
 MCP
Figure 3-2 Finger Flexion after Constraint Application [28]
2) Type II constraints [28]
This type of constraints refers to the couplings of joints between different fingers, also
called intrafinger constraints. For instance, when an index finger’s MCP joint is bent,
the middle finger’s MCP joint is forced to bend as well. Lee and Kunii [28] have
performed measurements on several different subjects and obtained a set of inequalities
that approximates the couplings of adjacent finger’s MCP joints:
d max ( MCP ( I ))  min( MCP (M )  25, S max ( MCP ( I )))

d min ( MCP ( I ))  max( MCP (M )  54, S min ( MCP ( I )))
d max ( MCP ( M ))  min( MCP ( I )  54, MCP ( R )  20, S max ( MCP (M )))
d min ( MCP ( M ))  max( MCP ( I )  25, MCP ( R )  45, S min ( MCP (M )))
d max ( MCP ( R))  min( MCP ( M )  45,  MCP ( L)  48, S max ( MCP ( R)))
d min ( MCP ( R))  max( MCP ( M )  20,  MCP ( L)  44, S min ( MCP ( R)))
d max ( MCP ( L))  min( MCP ( R)  45, S max ( MCP ( L)))
d min ( MCP ( L))  max( MCP ( R)  48, S min ( MCP ( L))) (3-2)
In the above inequalities, I, M, R, and L refer to the index finger, middle finger, ring
finger, and little finger; d max and d min refer to the maximum and minimum dynamic
angles of a specific joint’s movement; and Smax and Smin refer to the maximum and
minimum static angle of a specific joint [29].
- 34 -
3) Type III constraints [29]
This type of constraints refers to the couplings of the range of the joint’s motion based
on the hand anatomy. Here the only precondition is that the range of motion of each
joint that can be achieved without applying external forces. Some common Type III
constraints can be presented by the following inequalities:
15   MCP _ AA  15

0   MCP _ FE  90
(3-3)
0   PIP  110
0   DIP  90
Extension
Adduction
Flexion Abduction
Figure 3-3 Denotation of FE & AA motion
where the subscript FE denotes the flexion/extension motion of the MCP joint and AA
denotes abduction/adduction motion of the MCP joint (as illustrated in Figure 3-3).
Another commonly adopted constraint states that the middle finger displays little
abduction/adduction motion, as presented:
MCP _ AA  0
(3-4)
This will reduce one DOF from the 23-DOF model. Similarly, the TM joint also
displays limited abduction motion and will be approximated by zero as well:
TM _ AA  0
(3-5)
This condition reduces another one DOF.
By applying the above three types of constraints, the 23-DOF model can be
approximately reduced to a 17-DOF model. Previous researches have shown that the
- 35 -
natural hand postures can be approximately estimated using these constraints without
severe degradation in performance [30]. Also, there are other constraints imposed by
the naturalness of hand motions and which are difficult to detect or quantify. This type
of constraints will not be considered in this project because it is not possible to be
explicitly represented by equations or inequalities.
3.3. Kinematic Model
A kinematic model of the hand is formulated. As described in the skeleton model, four
DOFs are used for each finger, and the thumb is modelled by five DOFs. Another two
DOFs are for the palm motions. This hand kinematics can be illustrated by forward
kinematics using Denavit-Hartenberg transformation (D-H) [31].
3.3.1. Coordinate System Setup
Suppose there are n joints in a model, the joints numbered from 1 to n starting with the
base and ending with the end-effecter. A right-handed orthonormal coordinate system
( X 0 , Y0 , Z0 ) is established at the supporting base with Z 0 axis lying along the axis of
motion of the ith joint. Align the Z i with the axis of motion (rotary or sliding) of the
(i+1)th joint. Locate the origin of the ith joint’s coordinate at the intersection of the Z i
and Z i 1 axes or at the intersection of the common normal between the Z i and Z i 1
axes and the Z i axis. Establish X i  (Zi 1  Zi )/ || Zi 1  Zi || or along the common
normal between the Z i and Z i 1 axes (when the Z i and Z i 1 axes are parallel) and
assign Yi  (Zi  X i )/ || Zi  X i || to complete the right-handed coordinate system as

shown in Figure 3-4.
3.3.2. Link and Joint Parameters
The joint parameters illustrated in Figure 3-4 are defined as follows:
- 36 -
i th joint i
zi
zi 1
yi
xi
yi 1 ai Oi
i
zi  2 (i  1)th joint
'
O i 1
yi  2
di
yi 1
Oi' 2 xi 1
Oi 1
Figure 3-4 Coordinate System and Joint Parameters
- Joint angle  i : the angle of rotation from the X i 1 -axis to the X i -axis about the
Zi 1 -axis. It is the joint variable if the ith joint is rotary.
- Joint distance d i : the distance from the origin of the (i-1)th coordinate system to the
intersection of the Z i 1 -axis and the X i -axis along the Z i 1 -axis. It is the joint
variable of the ith joint is prismatic.
- Link length ai : the distance from the intersection of the Z i 1 -axis to the origin of the
ith coordinate system along the X i -axis.
- Link twist angle  i : the angle of rotation from the Z i 1 -axis to the Z i -axis about
the X i -axis.
3.3.3. Transformation of the Coordinate Frames
Four successive elementary transformations are required to relate the ith coordinate
frame to the (i-1)th coordinate frame:
- 37 -
 Rotate about the Z i 1 axis an angle of  i to align the X i 1 axis with the X i axis.
 Translate along the Z i 1 axis a distance of d i to bring X i 1 axis to the X i axis

into coincidence.
 Translate along the X i axis a distance of ai to bring the two origins Oi 1 and Oi
as well as the X axis into coincidence.
 Rotate about the X i axis with the angle of  i (in the right-handed sense), to bring
the two coordinates into coincidence.
The position and orientation of the ith frame coordinate can be expressed in the (i-1)th
frame by the following homogeneous D-H transformation matrix:
Aii1  R( zi 1 , i )T ( zi 1 , di )T ( xi , ai ) R ( xi ,  i )
cos i  sin i 0 0  1 0 0 0  1 0 0 ai  1 0 0 0
 sin  cos i 0 0  0 1    
0 0  0 1 0 0  0 cos  i  sin  i 0 
 i
 0 0 1 0  0 0 1 di  0 0 1 0  0 sin  i cos  i 0
    
 0 0 0 1  0 0 0 1  0 0 0 1  0 0 0 1
cos i  cos  i sin i sin  i sin i ai cos i 
 sin  cos  i cos i  sin  i cos i ai sin i 
 i
 0 sin  i cos  i di 
 
 0 0 0 1 
(3-6)
The location of the ith coordinate frame is specified with reference to the base
i
coordinate system as A0 , which is the chain product of successive coordinate
i
transformation matrices of Ai 1 :
i
A  A A ... A   Ajj1
i
0
1 2
0 1
i
i 1
j 1
R i
P   ni
i
si ai P0i 
 0

0

0 0  0 0 0 1 (3-7)
where R0   ni si ai  is the 3×3 orientation matrix of the ith coordinate frame and
i
P0i is the position vector of the ith coordinate frame related to the base coordinate
system.
- 38 -
3.3.4. Kinematic Model of the Hand
Figure 3-5 Hand’s Kinematic Model (Left Hand) [32]
Human hand is modelled with a hierarchical tree structure that consists of rigid links
and joints. Each joint consists of one or two degrees of freedom. This hierarchical
structure is represented in Figure 3-5 and each joint’s position is described using D-H
- 39 -
representation with reference to the heel of the hand (the world coordinate system
( x0, , y0, , z0, ) ). The posture of each finger ray (labelled as 1 to 5 from the thumb to the
little finger as shown in Figure 3-5) is represented by a local coordinate system. Using
D-H representation, the position of each joint can be transformed from the local
coordinates to the world coordinates sequentially.
As shown in Figure 3-5, the five finger rays can be divided into three groups based on
the different kinematic structures (thumb ray with five DOFs, index and middle finger
rays with four DOFs, ring and little finger rays with five DOFs). The kinematic model
for each group is described respectively as follows [32,33].
1) Thumb kinematic model
The model of the thumb is shown in Figure 3-6. There are totally five DOFs within
the three joints (one DOF at the IP joint, two DOFs at the MP joint and another two
DOFs at the TM joint). The coordination system is set up according to D-H principles.
The flexion/extension motion of the joint can be treated as the rotation about z1 j -axis
(j=1,3,4) and the abduction/adduction motion of the joint can be treated as the
rotation about z1 j -axis (j=0,2), where 1 is the notation of the thumb and j is the
notation of a specific rotation axis as illustrated in Figure 3-6.
Comparing with the other four fingers, the major difference of the thumb is the initial
position of the coordinate frame of the TM (MCP in the four fingers) joint with
respect to the palm’s coordinate system. There is an initial angle (approximately 90°)
between the flexion/extension motion plane of the thumb’s TM joint and the
flexion/extension motion planes of the other four fingers’ MCP joints which will
cause the difference when linking the local coordinate system to the world coordinate
system in the following descriptions.
- 40 -
Figure 3-6 Thumb kinematic Model
The D-H parameters of the thumb can be defined in Table 3-1:
Table 3-1 Thumb’s D-H Model Parameters
ith joint i di i ai Joint range cos  sin 
1 10 0 -90° 0 0°~ 60° 0 -1
2 11 0 90° l11 -25°~ 35° 0 1
3 12 0 -90° 0 0°~ 60° 0 -1
4 13 0 0° l12 0°~ 55° 1 0
5 14 0 0° l13 0°~ 80° 1 0
- 41 -
Two DOFs at the TM joint can be divided into two motions: One is the
abduction/adduction motion about the z10 -axis represented by 10 . Putting the
parameters in Table 3-1 into the D-H transformation matrix, we have
cos i  cos  i sin i sin  i sin i ai cos i  cos 10 0  sin 10 0
 sin  cos  i cos i  sin  i cos i ai sin i   sin 10 0 cos 10 0 
0 1 
1
A（）  i

 0 sin  i cos  i di   0 1 0 0
   
 0 0 0 1   0 0 0 1
(3-8)
The other is the flexion/extension motion about the z11 -axis represented by 11 .
Putting the parameters in Table 3-1 into the D-H transformation matrix, we have
cos 11 0 sin 11 l11 cos 11 
 sin  0  cos 11 l11 sin 11 
2
A（） 1   11
1
 0 1 0 0 
 
 0 0 0 1  (3-9)
Also we can divide the two DOFs at the MP joint into two motions: One is the
flexion/extension motion about the z12 -axis represented by 12 . Putting the
parameters in Table 3-1 into the D-H transformation matrix, we obtain
cos 12 0  sin 12 0
 sin  0 cos 12 0 
3
A（）1   12
2
 0 1 0 0
 
 0 0 0 1 (3-10)
The other is the abduction/adduction motion about the z13 -axis represented by 13 .
Putting the parameters in Table 3-1 into the D-H transformation matrix, we obtain
cos 13  sin 13 0 l12 cos 13 
 sin  cos 13 0 l12 sin 13 
3 1 
4
A（）  13
 0 0 1 0 
 
 0 0 0 1  (3-11)
The IP joint only has one DOF: the flexion/extension motion about the z14 -axis
represented by 14 . Putting the parameters in Table 3-1 into the D-H transformation
matrix, we have
- 42 -
cos 14  sin 14 0 l13 cos 14 

 sin  cos 14 0 l13 sin 14 
A45 (1)   14
 0 0 1 0 
 
 0 0 0 1  (3-12)
Consequently, the D-H transformation matrix of the thumb tip’s coordinate frame
5
with reference to the local coordinate system is A05 (1)   Ajj1 (1) . With respect to the
j 1
world coordinate system (the heel of hand as shown in Figure 3-5), an additional
transformation matrix is needed to represent the position vector which is defined as
cos  01' 0  sin  01' l01
'
cos  01' 
 
 sin  01' 0 cos  01' '
l01 sin  01' 
0' 1 
0
H（）
 0 1 0 0 
 
 0 0 0 1  (3-13)
where  01 is the angle between the thumb and the world y0 -axis and l01 is the
' ' '
length between the root of the thumb’s local coordinate system and the world
coordinate system, as shown in Figure 3-5.
Therefore, the thumb tip’s coordinate frame with respect to the world coordinate
system is defined by A0' (1)  A0 (1) H 0' (1)

5 5 0
2) Index & Middle finger’s kinematic model
There are totally four DOFs within the three joints (one DOF at the DIP joint, one
DOF at the PIP joint and another two DOFs at the MCP joint). The coordination
system is set up according to D-H principles similar to the thumb’s model. The
flexion/extension motion of the joint can be treated as the rotation about zik -axis
(i=2,3 and k=1,2,3) and the abduction/adduction motion of the joint can be treated as
the rotation about zi 0 -axis (i=2,3), where i is the notation of index finger (i=2) and
middle finger (i=3) and k is the notation of a specific rotation axis. Take the index
finger as an example as illustrated in Figure 3-7.
- 43 -
Figure 3-7 Index & Middle finger’s Kinematic Model
The D-H parameters of the thumb are shown in Table 3-2:
Table 3-2 Index & Middle Finger’s D-H Model Parameters

ith
joint
i di i ai Joint range cos  sin 
-15°~ 45°
1  20 0 -90° 0 (-10°~ 45° 0 -1
for the middle finger)
2  21 0 0° l21 0°~ 80° 1 0
3  22 0 0° l22 0°~ 100° 1 0
4  23 0 0° l23 0°~ 90° 1 0
- 44 -
With these parameters in Table 3-2 and the method used in analysing the thumb, the
D-H transformation matrix of the index finger can be obtained:
cos  20 0  sin  20 0
 sin  0 cos  20 0 
1
A（
0 2） 20
 0 1 0 0
 
 0 0 0 1 (3-14)
cos  21  sin  21 0 l21 cos  21 
 sin  cos  21 0 l21 sin  21 
A（2
1 2） 21
 0 0 1 0 
 
 0 0 0 1  (3-15)
cos  22  sin  22 0 l22 cos  22 
 sin  cos  22 0 l22 sin  22 
3
A（ 2）  22
2
 0 0 1 0 
 
 0 0 0 1  (3-16)
cos  23  sin  23 0 l23 cos  23 
 sin  cos  23 0 l23 sin  23 
4
A（
3 2） 23
 0 0 1 0 
 
 0 0 0 1  (3-17)
Consequently, the D-H transformation matrix of the index finger’s coordinate frame
4
with reference to the local coordinate system is A（2）  Ajj（
1 2）
4
. With respect to the
0
j 1
world coordinate system (the heel of hand as shown in Figure 3-5), an additional
cos  02'  sin  02' '
0 l02 cos  02' 
 
 sin  02 cos  02' sin  02' 
' '
0 l02
0
H（ 2）
0'
 0 0 1 0 
 
 0 0 0 1  (3-18)
where  02 is the angle between the index finger and the world y0 -axis and l02 is
' ' '
the length between the root of the index finger’s local coordinate system and the
world coordinate system, as shown in Figure 3-5.
Therefore, the index finger tip’s coordinate frame with respect to the world coordinate

4 4 0
- 45 -
3) Ring & little finger’s kinematic model
Figure 3-8 Ring & Little finger’s Kinematic Model
Different from the index and middle finger, there are totally four DOFs within the
four joints (one DOF at the DIP joint, one DOF at the PIP joint, two DOFs at the MCP
joint, and another one DOF at the HM joint). The coordination system is set up
according to D-H principles similar to the previous two models. The
flexion/extension motion of the joint can be treated as the rotation about zin -axis
(i=4,5 and n=0,2,3,4) and the abduction/adduction motion of the joint can be treated
as the rotation about zi1 -axis (i=4,5), where i is the notation of ring finger (i=4) and
little finger (i=5) and n is the notation of a specific rotation axis. Take the ring finger
as an example as illustrated in Figure 3-8.
- 46 -
The D-H parameters of the thumb are shown in Table 3-3:
Table 3-3 Index & Middle Finger’s D-H Model Parameters

th
i
joint
i di i ai Joint range cos  sin 
0°~ 10°
1  40 0 90° l41 (0°~ 20° 0 1
for the little finger)
-15°~ 20°
2  41 0 -90° 0 (-20°~ 35° 0 -1
For the little finger)
3  42 0 0° l42 0°~ 80° 1 0
4  43 0 0° l43 0°~ 100° 1 0
5  44 0 0° l44 0°~ 90° 1 0
With these parameters in Table 3-3 and the same method, the D-H transformation
matrix of the ring finger can be obtained:
cos  40 0 sin  40 0
 sin  0  cos  40 0 
1
A（
0 4） 40
 0 1 0 0
 
 0 0 0 1 (3-19)
cos  41 0  sin  41 l41 cos  41 
 sin  0 cos  41 l41 sin  41 
A（2
1 4） 41
 0 1 0 0 
 
 0 0 0 1  (3-20)
cos  42  sin  42 0 l42 cos  42 
 sin  cos  42 0 l42 sin  42 
3
A（ 4）  42
2
 0 0 1 0 
 
 0 0 0 1  (3-21)
cos  43  sin  43 0 l43 cos  43 
 sin  cos  43 0 l43 sin  43 
4
A（
3 4） 43
 0 0 1 0 
 
 0 0 0 1  (3-22)
cos  44  sin  44 0 l44 cos  44 
 sin  cos  44 0 l44 sin  44 
5
A（
4 4） 44
 0 0 1 0 
 
 0 0 0 1  (3-23)
- 47 -
Consequently, the D-H transformation matrix of the ring finger’s coordinate frame
5
with reference to the local coordinate system is A05 (4)   Ajj1 (4) . With respect to
j 1
the world coordinate system (the heel of hand as shown in Figure 3-5), an additional
cos  04' 0  sin  04' '
l04 cos  04' 
 
 sin  04 0 cos  04' sin  04' 
' '
l04
0
H（ 4）
0'
 0 0 1 0 
 
 0 0 0 1  (3-24)
where  04 is the angle between the ring finger and the world y0 -axis and l04 is the
' ' '
length between the root of the ring finger’s local coordinate system and the world
coordinate system, as shown in Figure 3-5.
Therefore, the ring finger tip’s coordinate frame with respect to the world coordinate

5 5 0
4) Palm motion
Figure 3-9 Palm motion (Left Hand)
- 48 -
As described above, the palm can be divided into three parts as shown in Figure 3-9.
The first part (part 1 in Figure 3-9) is fixed which includes the metacarpal bones of the
index and middle finger. The rest two parts (part 2 and part 3 in Figure 3-9) are
movable due to the rotation of the two HM joints located at the end of the ring and
little finger rays. The transverse deformation of the palm arch is formed by the
relative movements of these three parts. The D-H transformation matrixes of the HM
joints are described in Equation 3-19 and 3-24.
5) Simplified hand model
At the first stage of this project, the main focus is on the simple finger motion capture
(basically the FE motion of the joint). Thus, a simplified kinematic model, which does
not include the palm arch motion and the MCP joint’s AA motion, is proposed based
on the full kinematic model (Figure 3-10). The rest part of this thesis project is based
on this simplified model.
Figure 3-10 Simplified Hand’s Kinematic Model (Left Hand)
- 49 -
3.4. Discussion
In this chapter, the biomechanics of human hand are studied, including the skeleton
model, constraints of joint motions and the kinematic model. Twenty three major DOFs
are derived from the skeleton model, including five DOFs in the thumb, four DOFs in
each of the four fingers and another two DOFs in the palm. A full hand kinematic model
based on these twenty three DOFs is defined using D-H representation to illustrate the
realistic motions of the hand. The proposed kinematic model is able to represent the
transverse deformation of the palm arch, which brings a significant improvement of
hand posture and motion representation with comparison to the normal kinematic
models [32,33] without palm arch. Based on the different structure, three different
kinematic models are setup respectively for the thumb, the index & middle finger, and
the ring & little finger, in which the joints and links are defined by D-H notation.
In the constraints part, three kinds of hand constraints are introduced. By applying these
constraints, it can to reduce seven DOFs from the full 23-DOF model, which directly
results in using fewer sensors to capture the whole hand movement. Meanwhile, the
motion range of each joint is also defined by the constraints.
A simplified model with the joint’s flexion/extension motion only is proposed to fulfil
the requirements for the motion capture of the basic and simple finger motions.
- 50 -
Chapter 4. Sensing Principle
CHAPTER 4. SENSING PRINCIPLE
This chapter describes the multi-point sensing principle for SmartGlove which is based on
the inverted OLE sensing modules. In the SmartSuit project we have carried out, a new
OLE module (as shown in Figure 4-1) is developed by using the optical sensor to convert
the joint’s linear displacement to the joint’s angle [34]. However, this OLE is too large to
be used in hand motion capture. Thus, an inverted OLE sensing principle is formulated to
reduce the size. Based on this, a multi-point sensing principle which uses multiple OLEs to
detect multiple joints’ movement along the same strip is presented.
Figure 4-1 OLE for SmartSuit
4.1. Inverted OLE Sensing Principle
In the SmartSuit project, joint angle of a human body joint (such as elbow) is obtained
through a new linear encoder sensor. The linear encoder is attached to a wire that is
fixed on one forearm while the linear encoder assembly is fixed on upper arm and the
wire will be displaced along the circumference of the joint (as shown in Figure 4-2(a)).
The linear encoder is free to slide along its longitudinal axis. When the elbow bends, the
bending causes the skin to stretch. This stretch is translated into a linear displacement
and can be captured by the linear encoder assembly.
However, the OLE developed for the SmartSuit project is aimed for limb motion
capture. The size of the sensor assembly is too large for the finger joint motion capture.
Thus, an inverted sensing principle for OLE is formulated for the SmartGlove project.
As shown in Figure 4-2(b), in the inverted OLE, the encoder is fixed on the finger
segment to capture the displacement of the moving strip. Without the moving
- 51 -
mechanism, the size of the OLE for SmartGlove can be reduced significantly to fulfil
the dimensional requirement.
R0 R0
L0 L0
L0 L0
0 0
le) ed)
vab fix
mo E(
LE( OL
O
(a) OLE Sensing Principle (b) Inverted OLE Sensing Principle
Figure 4-2 OLE and Inverted OLE’s Sensing Principles
From the studies of human hand anatomy it is obvious that the joint axis is not fixed as
the finger bends. However, the displacement of the change of the finger joint centre is
very small compared to the dimension of the fingers. Hence, in this study we assume
that the finger joint centre is a fixed point instead of a moving point. Based on this
assumption, it is possible to use a circular disk to model the finger joint rotation. The
strip is placed on the surface of the finger with one end fixed (Figure 4-2(a)). As the
finger joint bends, it causes the skin the stretch. This stretch is translated into a linear
displacement and is captured by the OLE. As shown in Figure 4-2(b), when the joint
bends, the movement of the two segments can be approximated by the rotation of a
circular disk when the centre of the joint becomes the centre of the disk. The radius of
the circle is based on the biometric data on the subject under consideration and can be
approximately obtained via physical measurement. Assume the radius of the joint is R0
and the bend angle of the joint is 0 , by measuring the displacement of the strip ( L0 )
with the linear encoder, 0 can be estimated by
- 52 -
L0
0   360 (4-1)
2 R0
In principle, the calculation of the joint angle (Equation (4-1)) is identical to the OLE
sensor [34]. However, in next section, we will describe using the same principle to
detect multiple joints’ movement along the same strip.
4.2. Multi-point Sensing Principle
The above discussed inverted sensing principle is valid for single joint condition. As
there are more than three joints in one finger, multiple OLEs are required to get all the
three finger joints’ angle information. Therefore, an improved version is improvised for
this situation and is termed multi-point sensing. The basic working principle of
multi-point sensing is to place multiple OLEs in series on different finger segments to
capture the displacements of different detecting points using the same strip. As shown
in Figure 4-3, three disks (from left to right) represent three in-line joints with radius of
R1 , R2 and R3 respectively. Denote their bend angles as 1 ,  2 and 3 respectively.
Three OLEs are to be placed and fixed at positions (A), (B) and (C) as shown in Figure
4-3. Assume the displacement readings obtained by these three OLEs are D1 , D2 and
D3 . Because of the accumulated displacement at the distal joints, we have

2 R11
D1  L1  (4-2)
360
2 R11 2 R22
D2  L1  L2   (4-3)
360 360
2 R11 2 R22 2 R33
D3  L1  L2  L3    (4-4)
360 360 360
From Equations (4-2), (4-3) and (4-4), we can obtain
360 D1
1  (4-5)
2 R1
360( D2  D1 )
2  (4-6)
2 R2
360( D3  D2  D1 )
3  (4-7)
2 R3
- 53 -
(C)
R3 (B) R2 (A) R1
L3 L2 L1
Finger Palm
Tip (a) OLEs at Initial Position
L1
) 1
(A
L2
Palm
2
(B)
L3
3
(C)
Finger
Tip
(b) Conversion from Displacements to Angles
Figure 4-3 Multi-point Sensing Principle
Due to the natural arches of human hand, the multi-point sensing can be adopted in
finger motion capture. As shown in Figure 4-4, the hand has five longitudinal arches,
one for each of the five digital rays. Each arch is composed of a metacarpal and its
phalanges, linked by the MCP, PIP, and DIP joints. (The longitudinal arch for thumb is
linked by the MCP and IP joints) [35] As introduced in the hand kinematics, there are at
least 14 joints’ FE motions need to be captured in order to perform basic multi-finger
sensing, and all these 14 joints are all within the five longitudinal arches. Hence, by
introducing one strip for each longitudinal finger arch, it is able to use the multi-point
sensing method to capture the finger’s movement. In other words, multi-point sensing
- 54 -
only a single encoding strip is needed for sensing the multiple DOF movement of an
articulated object.
Figure 4-4 Arches of the Hand [35]
Based on the difference in the structures of the thumb and fingers, the detailed
multi-point sensing method and encoder placement for the thumb and fingers are
introduced separately as follows.
4.2.1. Finger Model
As introduced in the last chapter, there are three joints in one finger, and at least two
OLEs are required to obtain the three angle information. (The constraint of
2
 DIP   PIP is applied to reduce one sensor.) In order to obtain the joint angles of a
3
finger joint, the OLEs is fixed on the dorsal side of the hand. A strip goes through the
OLE’s plastic casing with one end attached to the back of the hand. The strip and the
two OLEs form a multiple encoder chain on the finger’s longitudinal arch. When the
- 55 -
finger starts to bend, the strip will move along the circumference of the joint angle, as
OLE Strip
RMCP RPIP RDIP
One end fixed

OLEs move and get
DOLE1 reading
 MCP
 PIP
DOLE 2
 DIP
Figure 4-5 Finger Model
In Figure 4-3, three disks (from left to right) are used to represent MCP, PIP and DIP
joints in a finger with radius of RMCP , RPIP and RDIP . Assume that the bend angles of
MCP, PIP and DIP are  MCP ,  PIP and  DIP respectively and the two OLEs’ readings
are DF 1 and DF 2 . The bend angles of MCP and PIP joints can be calculated using
Equations (4-5) and (4-6).
360 DF 1
 MCP  (4-8)
2 RMCP
360( DF 2  DF 1 )
 PIP  (4-9)
2 RPIP
- 56 -
The bend angle of the DIP joint can be obtained using the following constraint
equation:
2 240( DF 2  DF 1 )
CIP   PIP  (4-10)
3 2 RDIP
Using Equations (4-8), (4-9) and (4-10) the posture of the finger’s first three joints can
be calculated by converting linear encoder’s displacement reading to joint’s angle
information.
4.2.2. Thumb Model
Strip OLE
R IP
RMP
One end
fixed LT 1
OLEs move and get
 MP reading
 IP
LT 2
Figure 4-6 Thumb model
The kinematic modelling of fingers and thumb in Chapter 3 show that the structure of
thumb is similar to that of the other four fingers. Thus, the same multi-point sensing
method can be applied to capture the thumb’s motion. The only difference is that only
- 57 -
two joints in the thumb need to be measured compared to three joints in other fingers.
The setup of the OLEs on the thumb is as shown in Figure 4-6. Two OLEs are fixed
on the back of the thumb’s first and second segments to capture the rotation angles of
the MP and IP joint.
The way to obtain the rotation of finger’s MCP and PIP joints can also be
implemented in obtaining the rotation angle of thumb’s MP and IP joints:
180  LT 1
 MP  (4-11)
 RMP
180  ( LT 2  LT 1 )
 IP  (4-12)
 RIP
4.3. Discussion
In this chapter, a new sensing principle of using multiple OLEs to measure the joint
angle of a finger is illustrated. It is an improved OLE sensing method using multiple
OLEs to capture the displacement of different points along one strip. The advantage of
the multi-point sensing method is in its application for sensing multiple DOF movement
in an articulated object such as fingers or even the spine with unlimited number of joints
and unlimited number of sensors. The multiple DOF may be in the form of joint angles,
or curvature of the spine.
However, at this moment, the multi-point sensing method can only capture the planar
flexion/extension motions of the fingers (14 joints), but not the out-of-plane motion due
to the construction of the OLE. Although the flexion/extension motion is sufficient for
most finger actions, the abduction/adduction of the TM and MCP joints are also
important for the hand motion. Thus, further improvements on the multi-point sensing
method to include the detection of out-of-plane finger movement are needed in order to
capture the full hand motion in high fidelity.
- 58 -
Chapter 5. SmartGlove Development
CHAPTER 5. SMARTGLOVE DEVELOPMENT
This chapter focuses on the development of the multi-finger sensing system – SmartGlove
based on the sensors and method described in Chapter 4. As shown in the system block
diagram (Figure 5-1), the SmartGlove system includes three main parts: 1) hardware (OLE
sensors, microcontroller and glove); 2) firmware in the microcontroller for sensor data
handling; 3) application software in the computer for data management.
Glove
Hardware
OLEs
Micro-controller Firmware
PC Software
Figure 5-1 System Block Diagram
In order to capture the finger’s motion effectively, the proposed system will fit the
following requirements:
- High accuracy (±1°)
- High resolution (0.2°)
- Low cost (<S$1000 per glove)
- Comfortable to wear, easy to use
- Replaceable sensor
- Highly usable (un-obtrusive to user, wireless)
- Wearable over long periods
- Wireless
- 59 -
5.1. Hardware Design
Based on different specific tasks, the hardware of SmartGlove can be divided into three
modules: the OLE module, the microcontroller module, and the glove module. The
designs for each module are introduced respectively as follows.
5.1.1. OLE Module Design
The OLE module is the sensing module in the system which includes three basic units:
the sensing unit (sensor and lens), the interface unit (the customized PCB board), and
the housing unit (the customized base plate & strip), as shown in Figure 5-2. The
sensing unit is fixed in the housing unit to obtain the displacement of strip and to
communicate with the microcontroller through the interface unit.
Interface Unit
Sensing Unit
Housing Unit
Figure 5-2 OLE & Its Cross Sectional View
1) Sensing Unit (Sensor and lens)
The core sensor used in OLE is Avago’s optical mouse sensor product ADNS-3530
[36] as shown in Figure 5-3, which is based on Optical Navigation Technology that
measures changes in position by optically acquiring sequential surface images
(frames) and mathematically determining the direction and magnitude of movement.
- 60 -
Figure 5-3 Avago ADNS-3530 [36] Figure 5-4 Avago lens ADNS-3150-001 [36]
The ANDS-3530 is a compact optical navigation sensor with low-power architecture

and automatic power management modes. It is capable of high-speed motion
detection – up to 20 inch per second (ips) in velocity and 80m/s2 in acceleration. All
these features make it ideal for the SmartGlove. In addition, it consists of an on-chip
oscillator and integrated LED in order to minimize external components. The
ADNS-3530, along with the ADNS-3150-001 lens (Figure 5-4) [36], forms a
complete and compact linear tracking system. There is no moving part which means
high reliability and less maintenance.
The ANDS-3530 contains an Image Acquisition System (IAS), a Digital Signal

Processor (DSP), and a four-wire serial port (SPI communication). The working
principle of the ANDS-3530 is: the IAS acquires microscopic surface images via the
illumination system (Figure 5-5). The acquired images are processed by the internal
DSP to determine the direction and distance of motion. The DSP calculates the
relative displacements and send them out to the external microcontroller via the
sensor serial port. The microcontroller then translates the date into USB or Bluetooth
before sending them to the host PC.
Figure 5-5 Lens and Illumination System [36]
- 61 -
2) Interface Part (PCB)
In order to make the size of the OLE compact, the ADNS-3530 sensor is designed for
surface mounting on a PCB board. Based on the recommendation of the official data
sheet, the PCB board is designed as shown in Figure 5-6. The latest version has a
compact size which is about the same size as ADNS-3530.
Connector for SPI

and power
Sensor soldered on
the back side
13mm
12mm
Figure 5-6 OLE Circuit
There are seven output pins in the connector: two pins for 3.3V power supplier and the
other five pins for the Serial Peripheral Interface Bus (SPI) which is used for
communicating with the microcontroller. The Schematics and PCB layout, and
several considerations of PCB and sensor assembly are listed in Appendix A.
3) Housing Unit (Base plate and strip)
The housing unit is the holder for the optical navigation sensor and the moving strip.
According to [36], the distance between the lens and the moving strip (z-height)
determines the resolution of the result. As shown in Figure 5-7, in order to get high
resolution of the sensor, the distance should be within the range of 0.77mm to 0.97mm.
Furthermore, the surface material of the strip also affects the sensor’s resolution. To
make sure the strip sliding smoothly in the housing, there must be a gap between the
strip and the base plate. Consequently, for the stable readout, white Formica is the
- 62 -
ideal choice for surface material of the strip because the mean resolution is very stable
within the pre-defined range (0.77mm to 0.97mm).
Figure 5-7 Mean Resolution vs. z-height [36]
Taking all the above into design consideration, the strip and the housing are designed
as shown in Figure 5-8 (The dimensions of the housing refer to Appendix B.). The
bottom slot of the housing is for the sliding strip with a depth of 0.9mm. The material
of the housing is Delrin – a high quality engineering plastic used for reel parts and
even side plates.
Housing
Strip (White Formica Surface)
Figure 5-8 Housing and Strip
The detailed assembly drawing with several important dimensions is shown in

Appendix C.
- 63 -
5.1.2. Microcontroller
This project uses the Arduino Diecimila [37] (Figure 5-9) as the microcontroller to
communicate with the encoder. Arduino is an open-source physical computing
platform based on a simple I/O board. The programming language for the Arduino
microcontroller is an implementation of Wiring/Processing language.
Figure 5-9 Arduino Diecimila [37]
Arduino Diecimila is a microcontroller board based on the ATmega168. The on-board

ATmega168 microcontroller has 16K bytes of program memory pre-loaded with boot
loader which allows user to download program without additional hardware. The
ATmega168 supports SPI communication which can be used to communicate with the
mouse sensors.
The Arduino Diecimila has several ways to communicate with a computer. The
ATmega168 provides five-volt Universal Asynchronous Receiver/Transmitter
(UART) Transistor-Transistor Logic (TTL) serial communication, which is available
on the digital receive pin 0 (RX), and transmit pin 1 (TX). An FTDITM FT232RL chip
on the board channels this serial communication over USB and the FTDI drivers
provide a virtual COM port for the software on the computer to communicate.
- 64 -
The Arduino Diecimila can be powered either by the USB or by an external power
supply of 6 to 20 volts. The on-board USB chip can generate a 3.3 volt supply with a
maximum current of 50mA.
In order to connect ten OLEs in an easy way, an interface board (Figure 5-10) is
designed to work with the Arduino Diecimila. Because the on-board USB chip can
only generate a maximum current of 50mA, a voltage regulator is designed in the
interface board to draw 500mA current directly from the USB port to make sure that
the microcontroller is able to power up the ten OLEs. For safe operation,
open-collector buffers are added to the MOTION pin [36] which is used to get motion
signal from the encoder, and voltage translators are added to the four SPI pins (SCLK,
MOSI, MISO, and NCS) to avoid the violation of the different voltage level between
the microcontroller and the encoder. The detailed schematic drawing and PCB layout
of the interface board are in Appendix D.
Interface Board
Arduino Board
Figure 5-10 Arduino with Interface Board
- 65 -
5.1.3. Glove Design
The layout of OLEs, strips, and the microcontroller is as shown in Figure 5-11. On the
back of each finger, there is one strip placed along the finger’s midline. One end of
the strip is fixed to the dorsoulnar side of the glove (the five dots in Figure 5-11) as
mentioned in the multi-point sensing principle. Two encoders are chained by the strip
to measure the joint motion on each finger. The microcontroller is placed on the back
of the hand.
Finger's
Midline
Strip
Joints
OLE
Strip
Fixed
Microcontroller
Figure 5-11 Glove Layout
- 66 -
In order to make the glove type OLEs sensitive, the glove should fit nicely on the
human hand. At the same time, the glove should not hinder free motions of the hand.
Therefore, soft and stretchable fabric is used for the SmartGlove. In this project, two
different fabrics are used: the semi-stretching fabric, which can be stretched only in a
single direction, and the stretching fabric, which stretches in all directions. The glove
uses stretching fabric for backside of the MCP joints and semi-stretching fabric for the
palm side to avoid stretching along the finger direction. Thus, the glove has good
elasticity to fit the hand of users.
For ease of the replacement or maintenance of the sensors, the OLEs are mounted onto
the glove using Velcro and the microcontroller connects OLEs by ribbon wires. Thus,
the glove can be separated from the OLEs and the hardware for cleaning. This feature
is a significant step towards using data gloves in common daily living.
The photo type of the multi finger sensing device is shown in Figure 5-12.
Stretchable
Cloth
Full View Velcro
Index Finger Thumb
Figure 5-12 Glove Prototype
- 67 -
5.2. Firmware Design
Firmware is a computer program that is embedded in the microcontroller (ATmega168).

It is somewhere between hardware and software. Similar to software, it is a computer
program which is executed by a microcontroller. It is also tightly linked to a piece of
hardware, and has little meaning outside of it. The function of the firmware here is to
Figure 5-13 Firmware Architecture
- 68 -
communicate with the optical mouse sensor through SPI, including setting up
communication, sending enquiry data, receiving displacement data, and sending out
received data to PC’s COM port. The architecture of the firmware is illustrated in
Figure 5-13.
5.2.1. Serial Communication
A serial port is a computer interface that transmits data one bit at a time. In the
computer, applications access most serial ports as COM ports. Some USB devices
function as virtual COM ports so that application can interface with computer in the
same way as physical serial ports. Microcontrollers in embedded systems can use
serial ports to communicate with other embedded systems and PCs.
There are several advantages about the serial communication:

 Serial ports can exchange almost any type of information, suitable for
applications like reading sensors or other inputs.
 The hardware is inexpensive and readily available. Almost every PC and
microcontroller family includes built-in serial ports.
 Other than the Start, Stop, and optional parity bits added to each transmitted
byte, serial interfaces assume nothing about the content of the data being
transmitted.
 For devices that connect to PC, Windows and other operation systems provide
drivers for accessing COM ports.
 Wireless technologies (like Bluetooth) enable transmitting serial data without
cables.
Based on these advantages, serial communication is chosen to transmit data from

encoders to PC, in wired way (USB) or wireless way (Bluetooth).
5.2.2. Communication Protocol
A communication protocol defines how the bits travel. Based on the specification of
the optical mouse sensor, the Serial Peripheral Interface (SPI) is chosen to be the
- 69 -
communication protocol. The SPI bus contains four logical signals: SCLK (Serial
Clock, output from master), MOSI (Master Output, Slave Input), MISO (Master Input,
Slave Output), NCS (Chip Select, active low, output from master) [38]. Another
MOTION pin of the optical sensor is a level-sensitive output that signals the
microcontroller when motion has occurred. The SPI bus can operate with a single
master device (microcontroller) and multiple slave devices (OLE). An independent
NCS signal is required from the microcontroller to identify and select each sensor. As
the example of one master and three slaves shown in Figure 5-14, three slaves share
the same SCLK, MOSI, MISO and MOTION pin. NCS is identical in order to select
the specific slave device when the voltage level is active low. In this case, only one
slave can be selected at one time. Thus, the slave devices can only be operated by the
master device one by one in sequence. Additionally, the control (SCLK, MOSI, and
NCS) and data (MISO, MOTION) signal in each pin can only be transferred in
one-way direction (the arrows in Figure 5-14 show the direction of data transfer).
Sensor #1
(Slave #1)
SCLK
MOSI
MISO
MOTION
NCS
Microcontroller Sensor #2
(Master) (Slave #2)
SCLK SCLK
MOSI MOSI
MISO MISO
MOTION MOTION
NCS1 NCS
NCS2
NCS3
...
Sensor #3
(Slave #3)
SCLK
MOSI
MISO
MOTION
NCS
Figure 5-14 SPI Bus: One Master with Three Slaves
- 70 -
5.2.3. Flow Chart
According to the specification of ADNS-3530, the baud rate of the serial

communication is 9600bps for USB and 15200bps for Bluetooth, which is defined at
the beginning of the firmware.
In the system, the microcontroller needs an acknowledgement after an inquiry is sent

from PC, and the PC also needs an acknowledgement after the required data is
received. When sending to a PC that has no input buffer or a very small buffer, the PC
uses a full handshake to ensure that it is ready to receive a block of data (e.g. sending
“I” as an enquiry to the microcontroller). The microcontroller begins by sending a
block of data in a certain format (start with a capital letter and a three-digit number) as
a reply. The PC receives and sorts the data block based on the first capital letter – “A”
refers to the data from the first OLE; “B” refers to the data from the second OLE, etc.
(Table 5-1)
Table 5-1 Data Block Structure

Data
A--- B--- C--- D--- E--- F--- G--- H--- I--- J---
Form
Referred MCP IP MCP PIP MCP PIP MCP PIP MCP PIP
Joint (thumb) (thumb) (index) (index) (middle) (middle) (ring) (ring) (little) (little)
D-H
Angle
13 14  21  22 31 32  42  43 52 53
The microcontroller will scan ten OLEs as a loop, and the communication with each
OLE is identical. Based on the above information, the firmware flow chart for the
microcontroller to communicate with one encoder is indicated in Figure 5-15.
- 71 -
Start
Initialize
the internal
system
Initialize the
transmitter
software module
Initialize
external
devices
Enter deep
sleep mode
Generate new
ID
Prepare for
deep sleep
Enable global mode
interrupts
No
No Yes
Active Any message to Continue to
mode? send? send messages
Yes
Enable sensor
No Wait until
Transmission
transmission
completed?
complete
Yes
Enable motion
interrupts
Read motion
burst
Yes
Any motion Process
interrupts? interrupts
No
Disable
motion
interrupt
Transmit
motion data
Figure 5-15 Firmware Flow Chart
- 72 -
5.3. Application software Design
In order to be intuitive, real time response is one of the critical issues in the system
design. As a consequence, a software application is designed in the OpenGL
environment where a 3D virtual hand is displayed and controlled by the SmartGlove
directly.
GUI (Glui)
Environment
(OpenGL)
Renderer(Glut)
Inner Data
Manager
Adapter
Sensor
Data
Figure 5-16 Application Software Structure
As shown in Figure 5-16, the application software consists of three components: the
Adapter, the 3D Environment and the Graphical User Interface (GUI). The OpenGL
environment is the primary part which contains the hand avatar with a defined structure
and specific movement constraints [39]. The hand avatar is a 3D object in the OpenGL
environment with three spatial coordinates for the position and three angles for the
orientation, the setup of the hand avatar is based on the simplified kinematic model
(including the FE motion of 14 finger joints as shown in Figure 3-10). There are two
subcomponents of the OpenGL environment: the Inner Data Manager and the Renderer.
As mentioned in the firmware design section, the microcontroller gathers data from ten
OLEs and converts them into a packet form as in Table 5-1. The Adapter pushes the
data packages into the Inner Data Manager. The Inner Data Manager converts the data
package and maps the individual OLE’s data to the specific joint angle. Meanwhile, the
Renderer, which is based on the OpenGL Utility Toolkit (GLUT) library, automatically
redraws the 3D scene and generates the hand avatar (Figure 5-17).
- 73 -
Figure 5-17 Hand Avatar
Function Buttons
Real-time Joint Angle Display
Location of Captured Joints
Figure 5-18 Graphical User Interface
Due to the needs for debugging the application software and testing the individual OLE,
a control GUI (Figure 5-18) is developed with the GLUT-based C++ User Interface
(GLUI) open source library. This GUI has three main function areas which are 1)
- 74 -
Function Buttons; 2) Location of Captured Joints; 3) Real-time Joint Angle Display.

There are several control buttons in the function button area, such as starting or
stopping the hand avatar simulation, resetting system, checking data and quitting
software. After pressing the Start Simulation button, the software will acquire data from
ten OLEs. Each OLE’s data will display in the Real-time Joint Angle Display area and
the 3D hand avatar will move accordingly (Figure 5-19). The notation of each OLE is
illustrated in the Location of Captured Joints area with different colors and numbers.
The system can be reset by pressing the Reset button in the Function Buttons area. This
will clear all the displayed number in the Real-time Joint Angle Display area and
initialize the hand gesture (Figure 5-20).
The software also has several additional functions for certain applications. For example,
there is a red dot beside the displayed data of every joint which will be turned on if the
joint angle of a certain joint exceeds the preset limitation (as shown in Figure 5-19).
This function will be useful in training or rehabilitation to monitor the object’s motion
and give intuitive visual feedback of the abnormal motion. Besides, during one process,
the real-time data from each OLE is saved into a text file based on the structure as
shown in Table 5-1 and can be opened and checked by pressing the Data Check button
(Figure 5-21). The saved data can be used afterwards in hand motion analysis or motion
playback.
Hand Avatar
Over Limit
Indication
Figure 5-19 Real-time Data Display and Hand Simulation
- 75 -
Reset
Figure 5-20 System Reset
Recorded Data
Checking
OLE1 OLE2 OLE3 OLE4 OLE5 OLE6 OLE7 OLE8 OLE9 OLE10
Figure 5-21 Recorded Data Checking
5.4. Discussion
In this chapter, the SmartGlove is described, including hardware design, firmware

design, and the application software design.
The hardware design consists of three main parts: OLE module design, microcontroller
design, and glove design. In the OLE module design, Avago’s optical mouse sensor
- 76 -
product ADNS-3530 is chosen to be the main sensing unit of OLE. This is because of its
advantages such as low power, high resolution, fast speed and tiny size. Based on the
specifications of the sensor, a customized PCB, a housing unit, and a sliding strip are
designed to match the sensor. The final specifications of the OLE module are listed in
Table 5-2. From the specifications, it is obvious that the OLE is capable of capturing the
normal motion of the finger. As for the microcontroller, Arduino Diecimila, an
open-source electronics prototyping platform, is selected to obtain motion data from the
OLEs. In order to make it able to communicate with ten OLEs at the same time, an
additional interface board with voltage regulator and data buffer units are designed.
Lastly, the glove is designed based on the layout of the OLEs and strips. However, the
first prototype can only capture the FE motion of finger joints. An improved design is
needed for the capture of the MCP joint’s AA motion.
Table 5-2 Technical Specifications of OLE

Linear Linear
Power
Size Weight Data Rate Resolution Speed of Acceleration of
Consumption
the Strip the Strip
13mm(L)
12mm(W) 10 g >100Hz 0.06mm/s <0.51 m/s <80 m/s2 3.6mA
4mm(H)
The firmware is programmed based on the SPI protocol and ADNS-3530’s preloaded
operation commands. The SPI protocol is used in the communication between the
microcontroller and the OLEs. The data structure of the final data string which contains
displacement information from ten OLEs is also defined.
A presentation software with a 3D hand avatar and GUI is programmed in the OpenGL
environment which could provide real-time animation of hand movement and display
of joint angle data. There are also several basic functions integrated into the software,
such as over-limit indication and data store. However, it is only an early form of the
software which needs further improvement based on application requirements. For
example, the hand avatar is based on the 14-joint simplified kinematic model that can
only show the FE motion of the finger joints. In order to animate the full hand motion, a
complete hand avatar based on the 23-joint kinematic model is required. Additionally,
other useful functions are not integrated: calibration and playback of the recorded
motion, for instance.
- 77 -
Chapter 6. Calibration of SmartGlove
CHAPTER 6. CALIBRATION OF SMARTGLOVE
Calibration is a process to find a relationship between raw sensor data captured by the data
glove and the actual joint angles of the finger joint, which is necessary for the linear
encoder to obtain accurate measurement on different hand sizes. A standard way of doing
the calibration is to use external sensors (e.g. vision systems or hand masters) to provide
the ground-truth of the measurements. Fischer et al. [40] used a stereo vision system to
measure the real 3D positions of the fingertips, while also storing the joint sensor readings
of the data glove. The drawback of this approach is that any extra hardware required for
calibration will prevent the method from achieving wide-spread usage. Chou et al. [41]
also developed a method for CyberGlove calibration, which used linear regression to
establish the mappings between joint angle values and raw sensor readings of the
CyberGlove. The advantage of this method is the elimination of external devices. However,
lacking of enough calibration data for the linear regression, the calibration results are not
satisfactorily accurate. Based on the calibration method proposed by Chou et al., a flexible
and easy calibration process which can work without complex external sensory devices is
designed based on the sensing principle of the linear encoder.
6.1. Notations
The skeleton model has been introduced in Chapter 3. But for now, based on the sensing
principle of the multi-point sensing, only FE motion of 14 joints can be measured in one
hand. A simplified hand skeleton based on these 14 joints is shown in Figure 6-1.
A notation for the representation of the joint angles in the simplified skeleton model and
the sensor data from the OLEs is developed as follows. The FE rotational position of
each joint of a finger is represented by  mi , where m represents the fingers (m=1: thumb;
m=2: index finger; m=3: middle finger; m=4: ring finger; m=5: little finger.) and i
represents the number of the joint on the finger m as shown in Figure 6-1 (i=1: MCP
joint; i=2: PIP joint; i=3: DIP joint). For example, 11 refers to a rotation around the
x-axis of the thumb’s MP joint.
- 78 -
Middle
Index Ring
Z
Little
Y
 mi DIP(3)
X [6]
[4] [8]
PIP(2)
[10]
Thumb
[5] [7] MCP(1)

[3] [9]
IP(2) [2]
[1]
MP(1)
Figure 6-1 Simplified Hand Skeleton Notational Diagram (Left Hand)
Next, notation  j (where 1  j  10 , to denote the 10 OLEs as shown in Figure 6-1) is
used to describe the data from a specific linear encoder. For example,  6 refers to the
reading from OLE 6.
6.2. Data Mapping
As mentioned in Chapter 3, one joint angle may depend on the values of two OLEs (e.g.
the FE motion of each finger’s MCP joint depends on the values of both two OLEs on
that finger.) [41]. These relationships are summarized in Table 6-1 as followed.
Table 6-1 Joint/Encoder Relationships
Joint Relation Joint Relation Joint Relation

11 1 12 1  2 13 2/3 12
 21 3  22 3 4  23 2/3  22
31 5 32 5  6 33 2/3 32
 41 7  42  7 8  43 2/3  42
51 9 52  9 10
- 79 -
6.3. Calibration Model
Based on the two different relationships as described in Table 6-1, calibration of the
human finger is divided into two related models, one-to-one mapping and one-to-two
mapping. As described in Chapter4, there is a linear relationship between the joint angle
and the corresponding OLE’s reading. Thus, the linear regression model is adopted to
obtain this functional linear relationship as the calibration model.
6.3.1. One-to-One Mapping
In one-to-one mapping, the joint angles of PIP joints and IP joints (as 11 ,  21 , 31 ,  41
and 51 mentioned in the data mapping section) which have a one-to-one relationship
with an OLE is considered. A simple linear regression model with least squares
analysis is used. The general simple linear regression equation can be written as:
 A  GA ( A   A  A ) (6-1)
where  A  (11 ,21 ,31 , 41,51 ) is the joint angle in radians,  A  ( 1 , 3 , 5 , 7 , 9 )
is the value from the related OLE,  A and  A are the regression coefficients, and the
GA is the conversion factor [42].
In the calculation of the two regression coefficients, we define  A   A   A and
A   A   A (where  A and  A as the average value of  A and  A ). Then the
equations for GA ,  A and  A can be obtained as:
A  
 A A
(6-2)
 A
2
A  A  A A (6-3)
A
GA  (6-4)
A  A A
.
- 80 -
6.3.2. One-to-Two Mapping
In one-to-two mapping, the FE joint angles of MCP joints (as 12 ,  22 , 32 ,  42 and 52
mentioned in the data mapping section) which have a one-to-two relationship with the
OLEs are considered (as shown in Table 6-1). A multiple linear regression analysis
method with least squares analysis is used. The general multiple linear regression
equation is defined as:
 B  GB ( B   B  A   B  B ) (6-5)
where  B  (12 ,22 ,32 ,42 ,52 ) is the joint angle in radians,
(  A ,  B )  (( 1, 2 ),( 3 , 4 ),( 5 , 6 ),( 7 , 8 ),( 9 , 10 )) are the values from two
related OLEs,  B ,  B and  B are the regression coefficients, and GB is the

conversion factor [42].
In the calculation of the two regression coefficients, we define  A   A   A .
B   B  B and B   B   B (where  A ,  B and  B as the average value of  A ,  B
and  B ). Then the equations for GB ,  B ,  B and  B can be obtained as follows.
B 
      
B
2
A B A B B B
(6-6)
     (   )
A
2
B
2
B A
2
B        
2
A B B A B A B
(6-7)
     (   )
A
2
B
2
B A
2
 B  B  B  A   B B (6-8)
B
GB  (6-9)
B  B  A   B B
6.4. Calibration Methods
As mentioned in the calibration model, in order to calculate the coefficients in

Equations (6-1) and (6-5), at least two different angles are needed for each linear
encoder. Four simple calibration postures which are easy to perform (Figure 6-2) are
- 81 -
proposed by Rachid Kadouche in order to get two approximately standard angles for
each of the 10 linear encoders [43].
- Posture 1 corresponds to an angle of 0°for all the ten measured joints, also the
homing position for the ten linear encoders.
- Posture 2 defines the angles for the thumb’s MP and IP joints. ( 11 =45°;12 =90°)
- Posture 3 defines a 90° angle for all the fingers’ MCP joints. (  21 , 31 ,  41 and 51
=90°)
- Posture 4 defines a 90° angle for all the fingers’ PIP joints. (  22 , 32 ,  42 and 52 =90°)
Figure 6-2 Calibration Postures [43]
These four calibration postures are simple and easy to perform, however, the accuracy
is not good because there are only two angles (0°and 90°) for each joint to calibrate and
also the joints can only approximately reach the desired degree without external tools.
Thus, a single-joint calibration, which calibrates each OLE with a specially design
calibration tool, is proposed in this project for more precise calibration.
In order to obtain more than two different angles for a single joint, a calibration block
with five known angles (0°, 30°, 45°, 60°and 90°) is designed as shown in Figure 6-3.
Mechanical drawing is in Appendix E.
Figure 6-3 Calibration Block
- 82 -
By attaching the finger joint on different edges of the calibration block (Figure 6-4),
four standard angles can be obtained. Based on these, a precise calibration for one
single joint can be done by resolve Equations (6-2) and (6-6).
30°
45°
60° 90°
Figure 6-4 Calibration Block User Guide
Protractor
MCP PIP
Figure 6-5 Measure Joint Angle by Protractor
6.5. Validation of Calibration Model
The purpose of the calibration validation is to compare the result without any
- 83 -
calibration and the results after the above mentioned two calibration methods. Without
loss of generality, only the MCP and PIP joint on the index finger are calibrated. The
procedures for calibration verification could be described as follows.
Figure 6-6 Posture1, Posture3 and Posture4
Table 6-2 Calibration Verification Data

MCP PIP
Ref. (ο) 0 15 30 45 60 75 90 0 15 30 45 60 75 90
Radius
A 26.7 23.8
(mm)
B Reading 0 22.8 0 40.6
C (mm) 0 3.7 7.7 11.1 15.3 18.6 22.7 0 6.8 13.8 20 27.3 33.5 40.7
A. Use a vernier caliber to measure the radius of the MCP and PIP joint (Table 6-2).
Wear the SmartGlove and use the protractor to obtain seven readings from each joint
at 0°, 15°, 30°, 45°, 60°, 75°and 90°respectively (Figure 6-5). The radius and
readings are then substituted in Equation (4-5) and (4-6) to generate two sets of
angles without calibration (labeled as SetA1 for the MCP joint and SetA2 for the PIP
joint) as shown in Table 6-3.
B. As shown in Figure 6-6, perform Posture 1, Posture 3 and Posture 4 to obtain two
readings from each joint at 0°and 90°respectively (Table 6-2). Substitute the two
readings from the MCP joint (  A ) in Equation (6-1) to obtain the one-to-one
- 84 -
mapping linear regression equation for the MCP joint.

 A  3.95 A (6-10)
Similarly, substitute four readings from both the PIP (  B ) and MCP (  A ) joint in
Equation (6-5) to obtain the one-to-two mapping multiple linear regression equation
for the PIP joint.
 B  5.06(  B   A ) (6-11)
Finally, use the protractor to obtain seven readings from each joint at 0°, 15°, 30°,
45°, 60°, 75°and 90°respectively and substitute these readings in Equations (6-10)
and (6-11) to generate two sets of angles (labeled as SetB1 for the MCP joint and
SetB2 for the PIP joint) as shown in Table 6-3.
C. As shown in Figure 6-4, use the calibration block to obtain five readings from each
joint at 0°, 30°, 45°, 60° and 90° respectively (Table 6-2). Substitute the five
readings from the MCP joint (  A ) in Equation (6-1) to obtain the one-to-one
mapping linear regression equation for the MCP joint.
 A  3.98(  A  0.01) (6-12)
Similarly, substitute ten readings from both the PIP (  B ) and MCP (  A ) joint in
Equation (6-5) to obtain the one-to-two mapping multiple linear regression equation
for the PIP joint.
 B  5.03(0.05   A   B ) (6-13)
Finally, use the protractor to obtain seven readings from each joint at 0°, 15°, 30°,
45°, 60°, 75°and 90°respectively and substitute these readings in Equations (6-12)
and (6-13) to generate two sets of angles (labeled as SetC1 for the MCP joint and
SetC2 for the PIP joint) as shown in Table 6-3.
Table 6-3 Calibration Methods Comparison

MCP PIP
Ref. 0 15 30 45 60 75 90 0 15 30 45 60 75 90
A 0 14.1 28.4 42.5 56.6 70.1 83.9 0 15.5 31.2 46.6 62.3 77.7 93.4
(ο)
B 0 15.7 31.6 46.8 62.1 76.3 91.2 0 14.7 29.3 44 58.8 73.7 89.1
C 0 14.7 30.8 44.3 61.1 74.5 90.8 0 15.2 30.5 44.7 60.1 74.2 89.8
- 85 -
The deviations of the three sets of angles for the MCP joint (SetA1, SetB1 and SetC1)
from the reference angles are plotted as shown in Figure 6-7. Similarly, the deviations
of the three sets of angles for the PIP joint (SetA2, SetB2 and SetC2) from the reference
angle are plotted as shown in Figure 6-8. It is obvious that both calibration methods
(SetB and SetC) can improve the measurement results. Between the two calibration
methods, the second one (SetC) by using the calibration block can provide better result.
4 MCP Joint
2
0
Deviation(°)
0 20 40 60 80 100
-2
Reference1
SetA1
-4
SetB1
-6 SetC1
-8
Angle (°)
Figure 6-7 Calibration Test Result (MCP Joint)
4
PIP Joint
3
2 Reference2
Deviation(°)
SetA2
1 SetB2
SetC2
0
0 20 40 60 80 100
-1
-2
Angle (°)
Figure 6-8 Calibration Test Result (PIP Joint)
6.6. Discussion
The general demand for calibration of SmartGlove is that the methods are simple and
easy to carry out and can possibly work without complex external sensory hardware.
- 86 -
Two calibration methods are proposed based on this demand. For fast calibration, a
protocol with four standard hand postures is able to calibrate all of the ten OLEs roughly.
For precise calibration, a special calibration block is designed to calibrate each OLE
one by one. The linear regression method is adopted to model the OLE’s readings to
joint angles. The results of the calibration experiment showed that both methods can
improve the accuracy, especially the second tool-based method.
However, both two calibration methods are for the calibration of the joint’s FE motion.
When the joint’s AA motion is able to be captured by the SmartGlove, similar
calibration methods for the joint’s AA motion will also be our research consideration in
the future work.
- 87 -
Chapter 7. Experimental Verification
CHAPTER 7. EXPERIMENTAL VERIFICATION
Based on the completed hardware, firmware and software, it is necessary to conduct a

series of experiments to elaborate the performance of the OLE, and verify the design of the
SmartGlove. Two kinds of experiments are carried out: OLE characterization tests, and the
glove performance test.
7.1. OLE Characterization Test
In order to verify that our innovative OLE is suitable to be used in sensing human finger
motion, a series of tests are carried out to characterize the OLE as follows.
7.1.1. Linearity Test
The test on linearity of the OLE signal is carried out first. The purpose of this test is to
verify the linearity of the OLE data. The setup of the experiment is shown in Figure
7-1. When the jaw is moving along the scale of the vernier caliper, the reading from
the OLE and the displacement of the strip can be recorded. By comparing the two sets
of data, the linearity of the OLE reading can be determined.
Arduino
OLE
Vernier Caliper
Figure 7-1 Linearity Test
- 88 -
Experimental procedure of the linearity test can be described as follows.

1) Stick the OLE to the fixed jaw of a vernier caliper and stick the other end of the
strip to the other jaw of the vernier caliper, calibrate both to zero, wire the OLE to
Arduino Diecimila, and connect the Arduino Diecimila to computer through USB.
2) Move the strip from 0 to 20mm at an interval of 1mm. Read data from the OLE at
every interval.
3) Move the strip from 20 to 0mm at an interval of 1mm. Read data from the OLE at
every interval.
4) Repeat step 3) and 4) three times to obtain three sets of data and draw in Figure 7-2.
5) Calculate the average reading at each step upwards from 0 to 20mm, as well as
downwards from 20 to 0mm.
6) Plot the average encoder reading data against the displacement of the strip (read
from vernier caliber) in graph (Figure 7-3) and analyze the results.
340
290
OLE Reading(count)
240
190
140
90
40
-10
0 20 0 20 0 20 0
Displacement (mm)
Figure 7-2 Linearity Test Result
400
350
300
OLE reading(count)
250
200
0-20mm(Average)
150 20-0mm(Average)
100
50
0
-50 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Displacement (mm)
Figure 7-3 Linearity Test Result (Average)
- 89 -
Figure 7-3 indicates that the encoder has good repeatability performance in the
linearity test. The above two figures show that both the forward and backward data
sets are almost linear in the range of 0 to 20mm, which is the normal range of finger
skin deformation when a finger bends.
The slope (slope1) of the encoder’s average linear test results in Figure 7-3
approximately equals to 18, so the theoretical line can be drawn. Based on this result,
the linearity of the OLE reading can be obtained as follows:
average difference
linearity  1  100%  99.44% (7-1)
range
7.1.2. OLE Bending Test
The purpose of the OLE bending test is to exam the bending performance of the OLE
after the linearity of the OLE data is ensured. In order to eliminate the effect of finger
skin deformation, a wooden finger is used in the test. The experiment is carried as
indicated in Figure 7-4. The OLE is fastened to the wooden finger with the strip
wrapped over the knuckle. When the wooden finger bends, the OLE can read the
displacement of the strip, and based on the working principle of the OLE (Figure 4-1),
the bending angle could be calculated by Equation (4-1). By comparing the angle
captured from the OLE with the angle measured directly from the protractor sticked to
the wooden finger, the bending performance of the OLE can be examined.
OLE
Protractor
Wooden Finger
Figure 7-4 OLE Bending Test
- 90 -
The procedure of the OLE bending test is as follows:

1) Fasten the OLE to the back of one rod, wrap the strip over the knuckle and stick the
other end of the strip to the neighboring rod.
2) Connect the OLE to Arduino Diecimila.
3) Bend the wooden finger from 0°to 90°at an interval of 5°. Read data from the OLE
at each interval.
4) Bend the wooden finger from 90°to 0°at an interval of 5°. Read data from the OLE
at each interval.
5) Repeat step 4) and 5) three times to obtain three sets of data and plot.
6) Calculate the average readout at each step from 0°to 90°, as well as from 90°to 0°.
7) Plot the encoder readout data against the angle obtained from the protractor (Figure
7-6).
180
160
140
OLE Reading(count)
120
100
80
60
40
20
0
-20 0 90 0 90 0 90 0
Angle(°)
Figure 7-5 OLE Bending Test Result
155
135
OLE Reading(count)
115
95
75
0-90°(Average)
55 90-0°(Average)
35
15
-5
0 10 20 30 40 50 60 70 80 90
Angle(°)
Figure 7-6 Bending Test Result (Average)
- 91 -
Figure 7-6 shows good repeatability, as well as linearity, in the OLE reading in the
bending test in between 0°and 90°, which is the normal motion range of finger joints.
The slope (slop2) of the average bending test result in Figure 7-6 is approximately 1.9.
Also, the linearity of the OLE reading under bending condition can be calculated as
average difference
linearity  1  100%  99.42% (7-2)
range
The radius of the rotation joint can be obtained from the values of slope1 and slope2
by using the equation:
360  360 slope2
R   6.05mm (7-3)
2 l 2 slope1
(where slope1 means the value of OLE readout when the strip moves 1mm; slope2
means the value of OLE readout when the joint bends 1°.). The radius of the rotation
joint measured by vernier caliper is 6mm, which is very close to the value calculated
above.
7.1.3. Human Finger Test
The OLE has been demonstrated working in the bending test on a wooden finger.
However, the human finger is different from the wooden finger because of the skin
deformation which may affect the measure result. Hence, a further test on the real
human finger is necessary to test the accuracy of the OLE in deployment status.
The setup of the experiment is as shown in Figure 7-7. The OLE is attached to the first
knuckle of the index finger with an accelerometer attached to the second knuckle of
the index finger to measure the bending angle of the PIP joint. The palm is placed on
a flat metal plate as a stable reference during the test. In the human finger test, the PIP
joint of the index finger is bended back and forth three times. In the measurement,
data from the OLE and the accelerometer are recorded. By comparing these two sets
of data, the linearity of the OLE for actual use can be evaluated.
The model of the accelerometer is LIS3LV02DQ [44], a MEMS inertial sensor from
ST microelectronics.
- 92 -
OLE Accelerometer
Flat Metal Plate

Figure 7-7 Human Finger Test
The procedure for real human finger test is as follows:

1) Attach the OLE to the first knuckle of the index finger; attach the accelerometer to
the second knuckle of the index finger.
2) Tape the palm to a metal plate to keep it steady.
3) Connect the encoder and the accelerometer to Arduino Diecimila.
4) Rotate the index finger’s PIP joint back and forth three times and obtain two sets
of data from both the OLE and the accelerometer (ACC).
5) Convert the obtained two sets of data to angular form and plot them together (as
shown in Figure 7-8).
In order to map the output of accelerometer with that of the OLE, both two sensors are
connected to the Arduino board. While the finger bends, the microcontroller collects
data from the accelerometer and OLE one by one. Thus, the two output plots can be
superimposed based on the sensor counts.
Comparing the angular data from the OLE with the angular data from accelerometer
(the tilt angle calculated from the three orthogonal acceleration components [45]), it is
obvious that the results are very close and the difference between the linear encoder
and the accelerometer is within 1°, which indicate that the OLE is suitable for human
finger motion capture and can also produce good results.
- 93 -
Figure 7-8 Human Finger Test Result
7.2. Glove Testing
The previous three experiments showed that the individual OLE is capable of acting as
a sensing unit to capture human finger movements due to good linearity, repeatability
and accuracy. The next step is to conduct the glove performance evaluation tests in
order to verify the repeatability and reliability of the SmartGlove when ten OLEs work
together.
The SmartGlove evaluation procedure adopted in this project is based on the

standardized evaluation protocols for sensor glove devices proposed by Wise et al. for
the evaluation of Data Glove [46]. Similar tests are also adopted by Williams et al. for
SIGMA Glove evaluation [47], by Dipietro et al. for Humanglove evaluation [23], by
Lisa et al. for Shadow Monitor evaluation [48], and by Reinhard et al. for a sensor glove
evaluation [49].
- 94 -
7.2.1. Material and Subjects
To ensure that every time the finger joints can keep the same posture when gripping
an object, a cylindrical metal reference bar is prepared for the grip test so that the
relatively same grip characteristics are obtained.
Data is collected from five healthy male students aged 22-27 years with comparable
hand size and no hand movement disorders. All subjects are right-handed and the
glove is placed on the right hand. Calibration using the calibration block (the second
method introduced in Chapter 6) is performed on each subject before the test.
7.2.2. Experiment Protocols and Procedure
The standardized experiment protocols include four tests. However, focusing on

repeatability and reliability of multiple measurements over a single data collection
session, two tests (Grip Test and Flat Test) are adopted. The Grip Test (uses a gripped
hand position) and the Flat Test (uses a flat hand position) are carried out to analyse
the repeatability and reliability. Five sets of measurement are performed in each test
on each subject and each set of measurement includes ten grip/release actions.
Grip Test
The subject grips the prepared cylindrical reference metal bar (with the radius of
45mm) for six seconds and then releases for six seconds (Figure 7-9). During the
release, the subject’s hand is placed flat on the table. This grip/release cycle is
repeated 10 times. Repeating measurements are taken from each OLE during the grip
phase. A single data block is composed of data from ten grip/release actions on one
OLE (for instance, Figure 7-12 includes ten data blocks for ten OLEs). The test is
repeated five times without removing the glove between successive sets and a total of
50 grip/release cycles are done.
- 95 -
Reference
Metal Bar
Grip Flat
Figure 7-9 Cylindrical Grip Measurement
Flat Test
The subject places the hand on a table and alternately raises the hand and lightly
flexed the fingers, and then returns the hand back to the table top (Figure 7-10). Each
hand position lasts for six seconds and the flat/flex cycle is repeated for ten times. The
repeatability of the flat hand position is explored in this test. In order to keep the hand
and fingers in the same position during the flat period, an outline of the subject’s hand
profile is drawn on a paper and placed on the table. At the flat position, the subject is
asked to place the hand and fingers inside this drawn profile as shown in Figure 7-10.
The same as the Grip Test, this test is repeated five times without removing the glove
between consecutive measurements and a total of 50 flex/flat cycles is done.
Hand Profile
Flat Flex
Figure 7-10 Flat Hand Measurement
- 96 -
7.2.3. Statistical Analysis
In each measurement, an array { ijk } is obtained to specify the ith trial in the jth data
block which related to the kth encoder, where i=1,…,10, j=1,…,5, k=1,…,10. Two
variables are defined:
1 10
(1) Average OLE value  jk   ijk ;
10 i 1
(2) Range Rk  (max j ( jk )  min j ( jk )) ;
(3) Standard deviation (SD) of the  jk values;

(4) Average of the SD across the encoders.
For each OLE, the Range and SD are found to be correlated which give an
approximate evaluation of repeatability [46].
In order to evaluate the reliability, another statistical procedure is adopted which

based on the intraclass correlation coefficient (ICC) [23]. In statistics, the ICC is a
descriptive statistic which can be adopted when quantitative measurements are made
on objects that are organized into groups. It indicates how strongly objects in the same
group resemble each other. A prominent application of ICC is the assessment of
consistency or reproducibility of quantitative measurements made by different
observers measuring the same quantity. With removing the glove during successive
measurements, the data block obtained from each measurement can be treated as
different subject samples in the reliability analysis. [50] When the ICC approaches its
minimum value of zero, reliability decreases to its minimum value. Contrarily, the
reliability increases to its maximum value when the ICC approaches its maximum
value of one.
- 97 -
7.2.4. Results
1. Repeatability
A data block is a set of data that includes ten grip/release cycles for each OLE. In the
repeatability test, for each subject in each test, ten data blocks are processed to
calculate ten averaged angular values for each OLE, as shown in Figure 7-11.
120
MCP(little)
100
PIP(ring)
IP(thumb)
80 PIP(middle)
PIP(index)
Angle(°)
60 MCP(ring)
PIP(little)
MCP(middle)
40
MCP(thumb)
MCP(index)
20
0
0 2 4 6 8 10 12
Data Block
Figure 7-11 Averages of Ten Data Block for a Single Subject for the Grip Test
A sample data block from the Grip Test is shown in Figure 7-12.
- 98 -
Figure 7-12 Sample Data Block (Grip Test)
Repeatability is generally indicated by the range and standard deviation (SD),

consequently, the average range and SD are obtained from each subject in each test as
- 99 -
4
SD(Flat Test)
Subject
Range(Flat Test)
3
SD(Grip Test)
Range(Grip Test)
2
0 1 2 3 4 5
Angle(°)
Figure 7-13 Histogram of Averaged Rang and SD for Each Subject and Each Test
Table 7-1 Averaged Range & SD for Each Subject and Each Test
Grip Test Flat Test All Tests
Subject
Range SD Range SD Range SD
1 4.83 1.63 2.04 0.69 3.44 1.16
2 4.45 1.51 2.02 0.75 3.23 1.13
3 4.14 1.45 1.83 0.64 2.99 1.05
4 4.48 1.57 2.07 0.73 3.27 1.15
5 4.90 1.70 2.15 0.76 3.53 1.23
Average 4.56 1.57 2.02 0.71 3.29 1.14
From Figure 7-13, it is obvious that in all cases, the average flat hand repeatability
(from the Flat Test) for all subjects is better than the grip repeatability (from the Grip
Test). Average repeatability test for all the five subjects in the Grip Test is   1.57
and in the Flat Test is   0.71 , for an total average of 1.14 . Similarly, averaged
range result for the Grip Test is 4.56 , and for the Flat Test is 2.02 , for an total
average of 3.29 .
Looking into each OLE across Subjects 1 to 5 for each test, the histogram of Figure
7-14 summarizes the performance.
- 100 -
10
7 SD(Flat Test)
OLE
6 Range(Flat Test)
5 SD(Grip Test)
Range(Grip Test)
4
0.00 1.00 2.00 3.00 4.00 5.00

Angle(°)
Figure 7-14 Histogram of Averaged Range and SD for Each OLE
The above histogram indicates that the performances of OLEs on different joints are
similar, with maximum range= 3.76 , minimum range= 2.82 , overall average
range= 3.29 and maximum SD= 1.28 , minimum SD= 0.94 , overall average SD=
1.15 . Detailed data is shown in Table 7-2.
Table 7-2 Averaged Range & SD for Each OLE
Grip Test Flat Test All Tests

OLE
Range SD Range SD Range SD
1.MCP(thumb) 4.68 1.62 2.16 0.83 3.42 1.23
2.IP(thumb) 4.90 1.65 1.80 0.63 3.35 1.14
3.MCP(index) 4.50 1.55 2.16 0.79 3.33 1.17
4.PIP(index) 4.80 1.67 1.68 0.62 3.24 1.15
5.MCP(middle) 4.61 1.58 2.16 0.74 3.38 1.16
6.PIP(middle) 3.72 1.25 1.92 0.63 2.82 0.94
7.MCP(ring) 4.23 1.56 1.89 0.65 3.06 1.11
8.PIP(ring) 4.92 1.71 2.16 0.77 3.54 1.24
9.MCP(little) 5.28 1.78 2.24 0.77 3.76 1.28
10.PIP(little) 3.96 1.38 2.04 0.71 3.00 1.05
Average 4.56 1.58 2.02 0.72 3.29 1.15
- 101 -
2. Reliability
As stated before, reliability is indicated by intraclass correlation coefficient (ICC).

ICC analysis is performed for each test and for each OLE individually (ICC is
calculated using Excel). As shown in Table 7-3, for the Grip Test, the average ICC for
ten OLEs ranged from 0.882 to 0.987 with an overall average of 0.947. For the Flat
Test, the average ICC for each OLE ranged from 0.893 to 0.979 with an overall
average of 0.953. The combined average of ICC for the Grip Test and the Flat Test is
0.95. The ICC values in Table 7-3 show consistency from one data block to another
with no particular OLE showing significant lower reliability than the overall mean.
Table 7-3 Intraclass Correlation Coefficient of Reliability
Thumb Index Middle Ring Little

Average
MCP IP MCP PIP MCP PIP MCP PIP MCP PIP
Grip Test 0.937 0.954 0.882 0.963 0.913 0.987 0.948 0.957 0.969 0.964 0.947
Flat Test 0.955 0.968 0.893 0.966 0.908 0.976 0.955 0.968 0.958 0.979 0.953
Combined 0.950
7.3. Discussion
In this chapter, several tests are carried out to evaluate the characteristics of the OLE as
well as its performance when attached on the glove.
The linear relationship between the final digital signal and the real bend angle is a
desirable characteristic for the OLE. Unlike strain gauges or fibre optic sensors which
need a signal conditioning circuit that proportionally converts the sensor resistance into
voltage signals and an A/D converter which digitize the voltage signal, the optical
mouse sensor used in this project has an integrated DSP which can directly send out
digital signal of the relative displacement value. In this case, the OLE designed for this
project can provide better linearity. The results from the linearity test and OLE bending
test also show that the OLE’s linearity is high under both flat condition (99.44%) and
bending condition (99.42%).
- 102 -
Another desirable characteristic is the OLE’s accuracy, which is mandatory for the final
measurement. It is evaluated in the human finger test. Measuring the same joint’s
bending angle, the OLE shows satisfactory accuracy against the accelerometer (within
1°). This result also shows that the OLE is able to measure the human hand motion
effectively.
While many commercial and non-commercial sensor gloves exist, only a few studies
systematically address their repeatability and reliability follow the same standard
protocols proposed by Wise et al. As in Table 7-4, compared to the previous four
studies, the SmartGlove shows relatively good results in both repeatability and
reliability and also lies within the measurement reliability of manual goniometry [46].
Table 7-4 Comparison of Repeatability Results
Grip Test Flat Test Total

Glove Tested ICC
Range(°) SD Range(°) SD Range(°) SD
Wise et al.
(1990) data
6.5 2.6 4.5 1.6 5.5 2.1 0.94
glove (VPL
Research) [46]
Dipietro et al.
(2003) human
glove 7.47 2.44 3.84 1.23 5.66 1.84 0.7~1.0
(Humanware)
[23]
Simone et al.
(2007) shadow 5.22 1.61 1.49 0.5 3.36 1.05 0.95
monitor [48]
Reinhard et al.
(2008) WU 6.09 1.94 2.61 0.86 3.85 1.4 0.93
Glove [49]
SmartGlove 4.56 1.57 2.02 0.71 3.29 1.14 0.95
- 103 -
Chapter 8. Conclusion & Future Work
CHAPTER 8. CONCLUSION & FUTURE WORK
8.1. Contribution
This thesis documents the development of the SmartGlove based on a novel sensing
technology, including the design of OLE which uses the optical mouse sensor as the
basic sensing unit to measure joint’s flexion, the design of the firmware for
microcontroller to communicate with OLEs and gather displacement data, the glove
with Velcro that OLEs can be attached to, the calibration methods for both rough and
precise calibration, and lastly, the presentation software and GUI for 3D animation and
data display. The main contributions of this thesis are summarized as follows:
1. Modeling of the Full Hand Kinematics
Human hand is the most articulated, but also the most constrained part of the body. The
high dexterity of the hand also means the large numbers of DOFs need to be captured.
Within the hand’s relatively small space, there are 23 DOFs with five in the thumb, four
in each finger and another two in the palm. In this project, a full hand kinematic model
with these 23 DOFs is developed as a basic reference for the full hand motion capture.
The modelling of the palm’s two DOFs is the major improvement of this kinematic
model compared to the normal kinematic model for robotic hand or hand motion
capture. For now in the first stage of the project, because of the limitation of the OLE,
only a simplified model with 14 joints’ FE motion is adopted. In spite of this, the full
hand kinematic model is still important to produce realistic hand animation (the DOFs
cannot be captured now can be computed by the software). Also, the full hand
kinematic model guides the direction for further improvement of the SmartGlove.
2. Novel Sensing Unit for SmartGlove
As introduced in the literature review, several technologies have been adopted to do the
hand motion capture for the last three decades. Nowadays, the most popular
technologies used are strain gauges and fibre optics, with advantages and disadvantages.
Aiming at a low-cost but high performance glove-based hand motion capture device,
- 104 -
this project introduces a novel way which uses the optical mouse sensors to capture
human hand joint’s flexion. The special designed OLE based on the optical mouse
sensor has characteristics such as high resolution (can detect the strip’s motion up to 20
inch/s in linear speed and 80 m/s2 in acceleration), fast speed (at least 150Hz), low
power (3.6mA), and low cost (US$50 per OLE). In the characterizations tests, the OLE
also showed good linearity (99.44% in flat condition and 99.42% in bending condition),
repeatability and accuracy (within 1°compared to the accelerometer) in deployment
status. Additionally, the OLE has a compact size (13mm×12mm×4mm) and light
weight (10g) which make it easy to attach on the glove. The way of using Velcro to
attach OLEs also brings convenience for the OLE’s replacement and maintenance.
3. Multi-point Sensing Method for SmartGlove
The multi-point sensing method for SmartGlove is developed based on the inverted
design of the OLE (compared to the OLE for SmartSuit [34]) and the five natural arches
in human hand. Multiple OLEs chained by a flat strip are placed on the 1st and 2nd
knuckles of each finger to measure the flexion of MCP and PIP joints. The multi-point
sensing principle is based on the simplified kinematic model so that it can only capture
the FE motion of the finger joints now. It is compact in design and is verified to be
effective on the SmartGlove. Moreover, the application of the multi-point sensing can
be expanded into more areas (such as measuring the curvature of the spine or the
deformation of the skin) rather than hand motion capture.
4. Effective Calibration Method
Calibration is critical to the accuracy of hand motion capture because different people
come with different hand sizes. In this project, two methods are adopted for the
calibration of SmartGlove. One is a commonly used method for roughly calibration.
The other method uses a special designed calibration block to calibrate the OLEs one by
one for precise calibration. The calibration verification tests show that both methods
can improve the accuracy of the result, and the second method proposed in this project
is more effective.
- 105 -
8.2. Future Work
Although the basic issues relating to the development of the SmartGlove have been
addressed and initially solved, there are still several unexplored aspects in this area.
1. Abduction/Adduction Motion Tracking
In the first prototype, only the flexion/extension movement of the joint is captured.
However, as introduced in the hand bio-mechanism, our hand is so highly articulated
that only the FE motion cannot produce lifelike simulation of the real whole hand
movements. Consequently, the abduction/adduction motion is critical for the vivid
whole hand motion capture. Some of the other data gloves can provide the measurement
of joint’s AA motion using strain gauges (CyberGlove [8]) and fibre optics (ShapeHand
[7]). In this project, as the optical mouse sensor ADNS-3530 is capable to do the 2D
tracking, it is possible to improve the design of the OLE in order to make it able to
measure the AA motion while no additional sensors are adopted.
2. Robustness in Design
Because of the hand’s complex structure and deformable skin, a robust design becomes
very important, which includes issues like: find a better way to firmly attach the OLEs
and straps to the glove so that the influence of skin deformation can be minimized;
re-design the strap to increase toughness and reduce friction and kinks. All of these can
contribute to increase the robustness of the system, and further increase the accuracy of
the measurement.
3. Integration with Other Sensors
Hand motion capture is the major function of the SmartGlove, but will not be the only
one because there are a number of other glove products competent for it. The
integration with other sensors can enormously expend the usage. For example, add
touch sensors on the finger tip to sense the contact between fingers and other objects;
apply force sensors to measure the bending force of the finger; use accelerometers to
- 106 -
track the gesture of the palm. Different sensors can be adopted based on different
application, which can make our glove-based hand motion capture device exceptional.
4. Packaging Design
Our final goal is to develop a low cost, high performance, glove-based, un-tethered
wearable full hand motion capture device with ergonomics and durability. Thus, the
aim of packaging design is to make the SmartGlove unit, durable, comfortable to wear,
and easy to use. OLEs will be packaged with cloth and still be able to be removed; the
microcontroller will be sealed in small box; cables and wires will be hidden; and the
glove will be designed easily to wear and remove.
- 107 -
References
REFERENCES
[1] G. Welch, E. Foxlin, “Motion Tracking: No Silver Bullet, but a Respectable Arsenal,”
IEEE Computer Graphics and Applications, Vol. 22, No. 6, pp. 24-38, 2002.
[2] V. Medved, “Measurement of Human Locomotion”, CRC Press, 2001
[3] W. Goebl and C. Palmer, “Anticipatory Motion in Piano Performance”, Journal of the
Acoustical Society of America, Vol.120, Issue5, pp.3004, November 2006.
[4] K. Mitobe, T. Kaiga, T. Yukawa, T. Miura, H. Tamamoto, A. Rodger and N.
Yoshimura, “Development of a motion capture system for a hand using a magnetic
three-dimensional position sensor”, ACM SIGGRAPH Research posters: motion
capture and editing, Boston, Massachusetts, USA, 2006.
[5] Power Glove User Guide, Mattel Inc., September, 2007.
[6] 5DT Data Glove User Manual, Fifth Dimension Technologies, September, 2007.
[7] ShapeHand User Manual, Measurand Inc., September, 2007.
[8] CyberGlove Reference Manual, Immersion Corporation, September, 2007.
[9] Vhand User Guide, DGTech Inc., September, 2007.
[10] HumanGlove User Manual, Humanware S.R.L., May, 2008.
- 108 -
References
[11] T. Aaron, “A Novel Method for Joint Motion Sensing on a Wearable Computer”,
Proceedings of the 2nd International Symposium on Wearable Computers, pp.
158-159, October, 1998.
[12] N. Karlsson, B. Karlsson and P. Wide, “A Glove Equipped with Finger Flexion
Sensors as a Command Generator Used in a Fuzzy Control System”, IEEE
Transactions on Instrumentation and Measurement, Vol.47, Issue 5, pp. 1330–1334,
October 1998.
[13] L.H. Jose, K. Nicholas and W.L. Robert, “The AcceleGlove: a Whole-hand Input
Device for Virtual Reality”, International Conference on Computer Graphics and
Interactive Techniques, ACM SIGGRAPH, pp.259, San Antonio, Texas, USA, 2002.
[14] T. Alessandro, C. Nicola, Z. Giuseppe and D.R. Danilo, “Characterization of a Movel
Data Glove Based on Textile Integrated Sensors”, Proceedings of the 28th IEEE
EMBS Annual International Conference. pp.2510-2513, New York, USA,
September 2006.
[15] M.N. Noaman, R.A. Ahmmad and A.I. Aula, “Design and Implementation of DHM
Glove Using Variable Resistors Sensors”, Journal of Artificial Intelligence, Vol.1,
Issue 1, pp.44-52, 2008.
[16] D.J. Sturman and D. Zeltzer, “A Survey of Glove Input”, IEEE Computer Graphics
& Applications, Vol.14, Issue 1, pp.30-39, January 1994.
- 109 -
References
[17] D.L. Quam, “Gesture Recognition with a DataGlove”, Proceedings of the IEEE 1990
National on Aerospace and Electronics Conference, Vol. 2, pp. 755-760, Dayton,
Ohio, USA, 1990
[18] T. Kuroda, Y. Tabata, H. Ikuta and M. Murakami, “Consumer Price Data-glove for
Sign Language Recognition”, Proceedings of the 5th International Conference on
Disability, Virtual Reality & Assoc. Tech, pp. 253-258, Oxford, UK, 2004.
[19] Cyber Glove Applications, Immersion Corporation, September, 2007.
[20] Poared Hand Exoskeleton, Department of Computer Science, TU Berlin, March,
2008.
[21] T.G. Zimmerman, J. Lanier, C. Blanchard, S. Bryson, Y. Harvill, “A Hand Gesture
Interface Device”, Proceedings of the SIGCHI/GI conference on Human factors in
computing systems and graphics interface, pp. 189-192, Toronto, Ontario, Canada,
April 1987.
[22] W.-C. Lam, F. Zou and T. Komura, “Motion Editing with Data Glove”, Proceedings
of the ACM SIGCHI International Conference on Advances in computer
entertainment technology, Vol. 74. pp. 337-342, Singapore, 2004.
[23] L. Dipietro, A. M. Sabatini and P. Dario, “Evaluation of an Instrumented Glove for
Hand Movement Acquisition”, Journal of Rehabilitation Research and
Development, Vol.40, Issue 2, pp.179-190, March 2003.
- 110 -
References
[24] Y. Yasumuro, Q. Chen and K. Chihara, “Three-dimensional Modelling of the Human
Hand with Motion Constraints”, Image and Vision Computing, Vol.17, Issue 2, pp.
149-156, February 1999.
[25] Y. Wu and T.S. Huang, “Human Hand Modelling, Analysis and Animation in the
Context of HCI”, Proceedings of the International Conference on Image Processing.
Vol.3, pp. 6-10, 1999.
[26] T. Rhee, U. Neumann and J.P. Lewis, “Human Hand Modeling from Surface
anatomy”, Proceedings of the 2006 symposium on interactive 3D graphics and
games. pp. 27-34, Redwood City, California, USA, 2006.
[ 27 ] W. Ying, T.S. Huang, “Hand Modelling, Analysis and Recognition”, Signal
processing magazine, IEEE, Vol. 18, Issue 3, pp.51-60, May 2001.
[28] J. Lee and T.L. Kunii, “Model-Based Analysis of Hand Posture”, Computer graphics
and Applications, IEEE, Vol.15 Issue 5, pp.77-86, September 1995.
[29] J. Lin, Y. Wu and T.S. Huang, “Modeling the Constraints of Human Hand Motion”,
Proceedings of the Workshop on Human Motion, IEEE, pp. 121-126, Los Alamitos,
USA, 2000.
[30] J. J. Kuch and T.S. Huang, “Vision Based Hand Modeling and Tracking for Virtual
Teleconferencing and Telecollaboration”, Proceedings of the Fifth International
Conference on Computer Vision, pp. 666-669, 1995.
- 111 -
References
[31] J. Denavit and R.S. Hartenberg, “A Kinematic Notation for Lower-pair Mechanisms
Based on Matrices”, Journal of Applied Mechanics, ASME, Vol.22, pp.215-221,
June 1955.
[32] C.-C. Lien, “A Scalable Model-based Hand Posture Analysis System”, Journal of
Machine Vision and Applications, Vol.16. Issue 3, pp. 157-169, May 2005.
[33] M. Veber and T. Bajd, “ Assessment of Human Hand Kinematics”, Proceedings of
the 2006 IEEE International Conference on Robotics and Automation, pp.
2966-2971, Orlando, USA, May 2006.
[34] K. Y. Lim, Y. K. Goh, W. Dong, K. D. Nguyen, I.-M. Chen, S. H. Yeo, H. B. L. Duh,
and C. G. Kim, “A Wearable, Self-Calibrating, Wireless Sensor Network for Body
Motion Processing”, IEEE International Conference on Robotics and Automation
(ICRA 2008), pp. 1017-1022, Pasadena, California, May 2008.
[35] H.-L. Yu, R. A. Chase and B. Strauch, “Atlas of Hand Anatomy and Clinical
Implications”, Mosby, 2004.
[36] Datasheet of ADNS-3530, Avago Technologies, November, 2007.
[37] Arduino Diecimila User Guide, Arduino SmartProject, March, 2008.
[38] SPI Block Guide, Freescale Semiconductor, March, 2008
- 112 -
References
[39] M. Sama, V. Pacella, E. Farella, L. Benini and B. Ricco, “3dID: a Low-power,
Low-cost Hand Motion Capture Device”, Proceedings of the conference on Design,
automation and test in Europe, Vol.2, pp.136-141, Munich, Germany, 2006.
[40] M. Fischer, P. Van de Smagt and G. Hirzinger, “Learning techniques in a dataglove
based telemanipulation system for the DLR hand”, International Conference on
Robotics and Automation, pp.1603-1608, Leuven, Belgium, 1998.
[41] T.-S. Chou, A. Gadd, D. Knott, “Hand-eye: A Vision Based Approach to Data Glove
Calibration”, Proceedings of Human Interface Technology, 2000.
[42] W.B. Griffin, R.P. Findley, M.L. Turner, M.R. Cutkosky, “Calibration and Mapping
of a Human Hand for Dexterous Telemanipulation”, ASME IMECE Conference
Haptic Interfaces for Virtual Environments and Teleoperator Systems Symposium,
2000.
[43] K. Rachid, M. Mounir, M. Marc, “Modeling of the Residual Capability for People
with Severe Motor Disabilities: Analysis of Hand Posture”, International
conference on user modelling, Vol.3538, pp.231-235, Ediburgh, 2005.
[44] Datasheet of LIS3LV02DQ, ST Microelectronics, May, 2008
[45] W. Dong, K.Y. Lim, Y.K. Goh, K.D. Nguyen, I.-M. Chen, S.H. Yeo and B.L. Duh,
“A Low-cost Motion Tracker and Its Error Analysis”, International Conference on
Robotics and Automation, pp.311-316, 19-23 May 2008.
- 113 -
References
[46] S. Wise, W. Gardner, E. Sabelman, E. Valainis, Y. Wong, K. Glass, J. Drace and J.
Rosen, “Evaluation of a Fiber Optic Glove for Semi-automated Goniometric
Measurements”, Journal of Rehabilitation Research and Development, Vol.27,
Issue 4, pp.411-24, 1990.
[47] N.W. Williams, J.M.T. Penrose, C.M. Caddy, E. Barnes, D.R. Hose and P. Harley,
“A Goniometric Glove for Clinical Hand Assessment”, The Journal of Hand
Surgery: Journal of the British Society for Surgery of the Hand, Vol.25, Issue 2,
pp.200-207, April 2000.
[48] L.K. Simone, N. Sundarrajan, X. Luo, Y.C. Jia and D.G. Kamper. “A Low Cost
Instrumented Glove for Extended Monitoring and Functional Hand Assessment”,
Journal of Neuroscience Methods, Vol.160, Issue 2, pp.335-348, 2007.
[49] R. Gentner and J. Classen, “Development and Evaluation of a Low-cost Sensor
Glove for Assessment of Human Finger Movements in Neurophysiological
Settings”, Journal of Neuroscience Methods, Vol.178, Issue 1, pp.138-147, March
2009.
[50] Koch, G. Gary, “Intraclass correlation coefficient” in “Encyclopedia of Statistical
Sciences”, S. Kotz and N. L. Johnson, John Wiley & Sons, New York, 1982.
- 114 -
Appendix
Appendix A: OLE Circuit Board Design

Schematics design
- 115 -
Appendix
PCB layout
- 116 -
Appendix
Considerations of PCB assembly
① Surface mount the sensor and all other electrical components into PCB;
② Reflow the entire assembly in a no-wash solder process;
③ Place the lens onto the base plate. Care must be taken to avoid contaminating
or staining the lens.
④ Remove the protective kapton tape from optical aperture of the sensor and
LED. Care must be taken to keep contaminants from entering the aperture.
Recommend not placing the PCB facing up during the entire assembly process
and holding the PCB first vertically for the kapton removal process.
⑤ Insert PCB assembly over the lens onto the base plate aligning post to retain
PCB assembly. The lens piece has alignment posts to retain PCB assembly. The
lens piece has alignment posts which will mate with the alignment holes on the
sensor aperture.
⑥ The optical position reference for the PCB is set by the base plate and lens.
Note that the PCB motion due to button presses must be minimized to maintain
optical alignment.
- 117 -
Appendix
Appendix B: OLE Housing Design
- 118 -
Appendix
Appendix C: OLE Assembly
- 119 -
Appendix
Appendix D: Interface Board Design

Schematics design
- 120 -
Appendix
PCB layout
- 121 -
Appendix
Appendix E: Calibration Block
- 122 -

Li Kang 2009

Uploaded by

Copyright:

Available Formats

Li Kang 2009

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Li Kang 2009

Uploaded by

Copyright:

Available Formats

This document is downloaded from DR‑NTU (https://dr.ntu.edu.

Smartglove : a multi‑finger sensing system based

Downloaded on 24 Mar 2024 06:46:49 SGT

SMARTGLOVE – A MULTI-FINGER SENSING SYSTEM

SMARTGLOVE – A MULTI-FINGER SENSING SYSTEM

School of Mechanical and Aerospace Engineering

A thesis submitted to the Nanyang Technological University

My parents and my girlfriend, for their unwavering support, encouragement, and

LIST OF FIGURES ...................................................................................... 2

LIST OF TABLES ........................................................................................ 5

LIST OF ABBREVIATIONS ...................................................................... 6

CHAPTER 1. INTRODUCTION ................................................................ 7

1.1. Motivation ..................................................................................... 7

1.2. Objectives .................................................................................... 10

1.3. Scope ............................................................................................ 12

1.4. Report Organization .................................................................. 12

CHAPTER 2. LITERATURE REVIEW.................................................. 15

2.1. Glove-based Hand Mocap Technologies .................................. 17

2.2. Applications ................................................................................ 25

2.3. Discussion .................................................................................... 30

CHAPTER 3. BIOMECHANICS OF HUMAN HAND ......................... 31

3.1. Skeleton Model ........................................................................... 31

3.2. Constraints of Human Hand Motion ....................................... 33

3.3. Kinematic Model ........................................................................ 36

3.4. Discussion .................................................................................... 50

CHAPTER 4. SENSING PRINCIPLE ..................................................... 51

4.1. Inverted OLE Sensing Principle ............................................... 51

4.2. Multi-point Sensing Principle ................................................... 53

4.3. Discussion .................................................................................... 58

CHAPTER 5. SMARTGLOVE DEVELOPMENT ................................ 59

5.1. Hardware Design........................................................................ 60

5.2. Firmware Design ........................................................................ 68

5.3. Application software Design ..................................................... 73

5.4. Discussion .................................................................................... 76

CHAPTER 6. CALIBRATION OF SMARTGLOVE ............................ 78

6.1. Notations ..................................................................................... 78

6.2. Data Mapping ............................................................................. 79

6.3. Calibration Model ...................................................................... 80

6.3.1. One-to-One Mapping .............................................................. 80

6.4. Calibration Methods .................................................................. 81

6.5. Validation of Calibration Model............................................... 83

6.6. Discussion .................................................................................... 86

CHAPTER 7. EXPERIMENTAL VERIFICATION .............................. 88

7.1. OLE Characterization Test ....................................................... 88

7.2. Glove Testing .............................................................................. 94

7.3. Discussion .................................................................................. 102

CHAPTER 8. CONCLUSION & FUTURE WORK............................. 104

8.1. Contribution ............................................................................. 104

8.2. Future Work ............................................................................. 106

Objective measures of human hands as individuals participate in everyday activities are

Figure 2-1 Optical Tracking [3] ........................................................................................ 15

Figure 4-4 Arches of the Hand [35] .................................................................................. 55

Figure 7-3 Linearity Test Result (Average)...................................................................... 89

Table 2-1 Products’ Comparison Chart ............................................................................ 22

Mocap Motion Capture

With the rapid development of computer technology, we gradually tend to use

Motion capture or motion tracking technology started in the 1970s as a

In order to fulfill the requirements of multi-finger motion capture, several objectives

(1) Kinematic modeling of realistic human hand anatomy.