See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/315711962

A survey on moving object tracking using image processing

Conference Paper · February 2017

DOI: 10.1109/ISCO.2017.7856037

CITATIONS READS
9 2,777

2 authors, including:

Balaji S R
Panimalar Engineering College, Chennai, India
3 PUBLICATIONS   9 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

VLSI Design View project

All content following this page was uploaded by Balaji S R on 08 January 2018.

The user has requested enhancement of the downloaded file.

2017 11 th International Conference on Intelligent Systems and Control (ISCO)

A SURVEY ON MOVING OBJECT

TRACKING USING IMAGE
PROCESSING
S. R. Balaji Dr. S. Karthikeyan,
Research Scholar Asst. Professor
Department of ECE Department of ECE
Sathyabama University Sathyabama University
Chennai, India Chennai, India
balasrb 1983@gmail.com skarthi 1879@gmail.com

Abstract - Image processing is a method of extracting

some useful information by converting image into
digital inform by performing some operations on it.
Video Sequence
Object detection and tracking are the task that is
important and challenging such as video surveillance
and vehicle navigation. Video surveillance is a Frame Difference
technology which works in dynamic environment in
various events such as sports, public safety, and
management of traffic. This paper reviews the
various challenges and aspects of detection and Background
tracking of objects.
Object Detection
Subtraction
Keywords - Object detection, classification, tracking

I. Introduction
Optical Flow
Object tracking plays a vital role in the field
of computer vision [1]. Object tracking algorithms
have acquired priority due to the availability of
highly sophisticated computers, good quality and
inexpensive cameras. The basic steps for object Motion based
detection and tracking are shown in fig 1.
In the object tracking, the video analysis
involves, moving object detection, object Object Texture based
classification, frame to frame object tracking. Classification
Object detection deals with the identification of Shape based
objects from video frame [2] and to cluster pixels of
these objects [3]. In object classification, the
objects are classified as birds, humans, vehicles and Color based
other moving objects [3]. While object tracking
involves the selection of Region of Interest (ROI)
and keep tracking of motion and its positions from
the video frames [2]. The paper is sorted in the
following ways: Section II describes the methods of Point based
object detection, Section III describes the methods
of object classification, and Section IV describes
the methods of object tracking. Section V provides Kernel based
conclusion. Object Tracking
II Object Detection methods
Moving object detection is the initial step for the
process of analyzing a video. This is done either in Silhouette based
each and every frame or when the object first
appears in the video [4]. Fig 1. Basic steps for object detection and tracking

978-1-5090-2717-0/171$31 .00 ©2017 IEEE 469

 It handles with the elimination of slow-moving traffic, the storage requirement will be
stationary background objects form the moving of prominent.
object of interest [5]. The following techniques B. Optical Flow:
basically focus on the detection of moving objects This method is to calculate the image optical flow
which are majorly the primary source of fIeld and do a process called clustering based on the
information. optical flow distribution characteristics of image.
[10]. The complete detection and movement
II Object Detection methods information of object from the background can be
Moving object detection is the initial step for the obtained from this method. While the complexity in
process of analyzing a video. This is done either in calculation, poor anti noise performance make it not
each and every frame or when the object fIrst reliable for real time demanding applications.
appears in the video [4]. It handles with the
elimination of stationary background objects form TABLEI. COMPARATIVE STUDY OF OBJECT DETECTION
METHODS [9]
the moving of object of interest [5]. The following
Methods Merits Demerits
techniques basically focus on the detection of
moving objects which are majorly the primary It req u ires a
Easiest way.
background
source of information. Frame Difference Perform well for
without moving
A. Frame Difference Method: static background.
obiects
The difference between two consecutive images is I. Low memory I. It computation
calculated, which further identifIes the moving requirement. requires a buffer
2. It does not with the recent
object. The calculation made here is easy and require sub pixel values.
Background
simple for implementation. For various dynamic Subtraction sampling of 2. It does not cope
environments, this method has a strong adaptability, fi·ames for creating with multimodal
yet the fmal result of non-stationary object detection an adequate background
background
is not accurate as it is tough to obtain complete model.
outline of that moving object [6]. It can produce the Require large amour
B. Background Subtraction method: Optical Flow
complete of calculation
Background modeling is the fIrst step of background movement
infonnation
subtraction. It is used to obtain the reference model.
This reference model is compared with each video
sequence to determine possible variation. The III. Methods of Object ClassifIcation
existence of moving object is determined by the Here, the classifIcation of objects is done based on
difference between the current video frame and the their shape features of the motion region [6]. The
pervious frame in terms of pixels [7]. The two classifIcation is as follows:
common approach [8] available in background A. Motion based classification:
subtraction are: Moving object classifIcation has a strong cue of
i. Recursive Algorithm: periodic property shown by a non-rigid articulated
This technique does not involve buffering for the motion of the object. Optical flow is also helpful for
estimation of background [8]. But, recursively the object classifIcation. Rigidity analysis and
update a single background model based on each periodicity of moving entities are done by residual
given input frame. So that, the current background flow. The non-rigid dynamic objects such as human
could be affected by input frames from distant past. being can have greater average residual flow and
The merit of this algorithm is that, it requires less display a periodic component, while the rigid objects
storage. In case of background error, it would be are expected to have little residual flow. [11].
existing for longer time. The different methods B. Texture-based classification:
used in this technique are adaptive background, The gradient orientation that occurs in the localized
Gaussian mixture and approximate median. portions of an image is counted in texture based
ii. Non-Recursive Algorithm: technique. It is then computed using overlapping
The background estimation involved in a non- local contrast normalization on a dense grid of
recursive algorithm [8] uses a sliding window uniformly spaced cells for improving accuracy [9].
approach. It stores a buffer of previous L video C. Shape-based classification:
frames and variation of each pixel within the buffer Representation of points, box and blob are various
is used to estimate the background image. This motion regions available for classifying the moving
techniques does not depend on history beyond those objects based on the shape information. The
frames stored in the buffer, thus they are highly mixture of image based and scene based parameters
adaptable. In case of large buffer need to cope with like image blob area, camera zoom and apparent

470
aspect ratio of blob bounding box acts as the input
features to the network. Each blob at every frame The objects are tracked in this method is based on
undergoes the classification and the histogram keeps the performance criteria. Based on the criteria that
the results [11]. make sense, optical point will be taken [16].
D. Color-based classification Prediction and correction are the two phases present
Colour is easily acquired and is relatively stable in this method. Prediction involves the usage of
under viewpoint variations, unlike other image current set of observations and update for the next
features. Low computational cost of algorithms stage which is the second step that gives a better
proposed makes a colour a desirable feature to approximation of next state. The balance between
exploit when appropriate. In real time to detect and the predicted values and noisy measurements are
track the vehicles, colour histogram based algorithm discovered using Kalman filter. Modeling state
is used. To segment image into background and equation is used to determine the value of the
object and to describe the colour distribution within weights. This method is to track the system being
the sequence of images, a model is created measured at discrete intervals of time [17]. The
according to [12] a Gaussian Mixture method. The basic step involved in Kalman filter is as shown in
occlusion buffer is used to handle the object fig 3.
occlusion.
Object tracking
Table 2: Comparative Study of Object Classification methods [9]
I
Methods Accuracy Comments
Does not require
predefined pattern
I I
Motion based Medium templates but struggles to Point Kernel Silhouette
identi fy a
non-moving human
Tracking Tracking Tracking
Provides improved quality
Texture based High with the expense of
additional computation
time.
Simple pattern-matching Kalman Simple
approach can be applied I- Contour
Shape based Medium filter - Template r-
with appropriate Matching
templates. Matching
It creates a Gaussian
Mixture Model to describe
Particle
I-
the color distribution filter Mean
Color based High within the sequence of - Shape
shift ......
images and to segment the Matching
image into background
and objects Multiple
Hypothesis Support
IV. Methods of Object Tracking Tracking - Vector
Object tracking is the next step followed by the Machine
object detection. It is a technique used to track and
also the travelling direction of objects [14].
According to the author [13], this method is
classified as point, kernel and silhouette based Layer
tracking. The following diagram briefs the - based
classification of object tracking methodology. tracking
A. Point Tracking:
It is the common task of the computer vision with
various applications involved in it. During tracking, Fig 2. Types of Object Tracking [3]
the moving objects are shown by their feature points ii. Particle Filtering:
Point tracking is a major problem especially in false Before moving to the next variable, all the models of
detection of object and in incidence of occlusions that variable will be generated using this filter. Major
[15]. The recognition can be made simple by advantage of this method is that there can be
thresholding, at of identification of these points. numerous variables which can be generated
dynamically.

471
 and non-rigid objects. The tracking techniques
involve representation of object, object features,
shape and appearance of object.
i. Simple Template Matching:
Prediction of Update predicted Examining of Region of Interest (ROI) in the
state variable state variable ongoing video is done by a method called simple
temple matching [9]. A frame is separated from a
video which is verified in the template matching.
This method can track only a single object in the
video. In template matching, translation of motion
only can be done [13].
ii. Mean Shift method:
Fig 3. Basic step of Kalman Filtering [13]
Initially, ROI is defined from moving object by
New operation of resampling is also segmentation and then tracking the object is made
allowed [13]. The assumption of state variables are from one frame to the next. In a initial frame, ROI
normally distributed in this filter which is it is defined by the rectangular window. The
drawback, making it to poor approximations of background object is separated from the tracked
stage which do not Gaussian distribution. This object by this algorithm [13] . Chamfer distance
drawback came be overcome by using this particle transform is used to improve the accuracy of target
filtering. Contours, colour features on texture representation and localization. The distance
mapping are generally used in this algorithm. This between two colour distributions is minimized
filter is a Bayesian sequential importance sample using Bhattacharya Co-efficient and also by
technique that uses a fmite set of weight trials, to chamfer distance transform.
recursively approach the later distribution [\3]. The iii. Support Vector Machine (SVM):
two fundamental phases are prediction and update This method gives a set of positive and negative
which are similar to that of Kalman Filtering. training values. The positive samples include
iii. Multiple Hypothesis Tracking (MHT): tracked image while the negative samples contain
There is always a limited chance of an incorrect all other remaining things that are left untracked. It
correspondence if motion correspondence is can also handle single image but necessity of
recognized using only two frames. If several frames physical initialization and necessity of training [\9] .
have been observed, then better tracking outcomes iv. Layer based Tracking:
can be acquired. Several suggestions for each object Multiple objects are tracked by this method which
at each time frame can be uphold using MHT comes under Kernel tracking. Shape
algorithm. The final track of the object is the most representation, movement such as translation and
likely set of correspondences over the period of time rotation and layer appearance are involved in each
of its observation [13] . This is an iterative layer based on intensity. Initially, compensation of
algorithm. Iteration starts with the set of existing the background motion is done in layering such that
track hypothesis. Each hypothesis is a crew of the objects motion can be evaluated from the
disconnect tracks. A prediction of each object rewarded image by means of 2D parametric motion.
position in succeeding frame is made for each Based on the objects foregoing motion and shape
hypothesis. Distance measured is calculated and features every pixels probability is calculated [\9].
compared to the predictions. C. Silhouette Tracking:
B. Kernel Tracking: Hands, head and shoulders are the objects having
Kernel tracking [\8] is generally done by composite shapes cannot be well defined by
calculating the non-stationary object, which is geometric shapes. A perfect distribution of shape
shown by the embryonic object region, from one for those objects will be given by this approach
frame to another frame. The movement of object is [\8]. The objective of this tracking mechanism is to
in parametric form, such as translation affine etc. detect the object region in every frame with the help
These algorithms vary in terms of the number of of an object model obtained by the previous frames.
objects tracked, representation used and the Dealing with various object shapes, occlusions and
approximation method for object movement [I]. object split and merge is possible. Colour
The illustration of object using geometric hologram, object edges or contour are used to
shape is common in real time. The parts of the represent the model.
objects will be left outside of the prescribed shape i. Contour Tracking:
while parts of the background exist inside which is In the contour tracking methods, an iterative
one of the drawbacks that can be detected in rigid progress of a primary contour in the previous frame

472
 to its new position in the current frame is done [18]. V. CONCLUSION
Certain amount of object in the current frame In this paper, a wide spread survey of object
overlay with the object region in the previous detection and tracking methods have been
frame. Contour tracking is in the form of State presented. Merits and demerits of available methods
Space models [l3]. Parameters of shape and motion of object detection, classification and tracking have
of contour are used to name the state of object. been discussed in detail. The background
According to the maximum probability, the state is subtraction method is concluded to be the easiest
updated for each time. method which gives the complete details about the
In this method, the representation of tracking object compared to frame difference and optical
is done either explicitly or implicitly. The function flow detecting methods.
is defined by the grid in case of implicitly whereas The kernel or contours based tracking
representation is based on explicitly defines the require detection only when the object first appears
boundaries of the silhouette. on the screen whereas the point tracking involve
ii. Shape Matching: detection in every frame. Future work should be
These approaches examine for the object model in efficient for proper understanding of different
the existing frame. This shape matching algorithms to reduce errors in tracking and to use
performance is same as the template based tracking most appropriate algorithm to decrease the time
in kernel approach. required for tracking of objects contammg
Finding matching silhouette which are detected in diversified characteristic videos. We believe that,
two successive frames is another approach to shape this paper can give reliable vision into this
matching [l3]. Point matching and silhouette particular research topic and encourages future
matching are generally alike. Background work.
subtraction is used for detection based on Silhouette
density functions, silhouette boundary and object REFERENCES
edges are the forms of models of object. Hough [1]Alper Yihnaz, Ohio State University, Omar Javed,
Object Video, Inc., Mubarak Shah, University of Central
transform techniques are used to perform and deal Florida,"Object Tracking: ASurvey", ACM Computing
with single object and occlusion [3]. Surveys, Vol. 38, No. 4, Article 13, Publication date: December
2006.
Table 3. Compalison table for Silhouette and Point tracking [13] [2] Payal Panchal, Gaurav Prajapati, Savan Patel, Hinal
Shah and Jitendra Nasriwala, " A Review on Object
Type of Region Appearance Detection and Tracking methods", International Journal for
trackin~ features Research in Emerging Science And Technology, Volume - 2,
Silhouette Complete Object model Issue-I , January 2015.
tracking [3]Himani S. Parekh, Darshak G. Thakore,Udesang K. JaJiya, " A
Point tracking Points Position based Survey on Object Detection and Tracking Methods",
International Joulllal of Innovative Research in Computer and
Table 4. Comparison of tracking methodologies [13] Communication Engineering, Vol. 2, Issue 2, February 2014.
[4]Kinjal A. Joshi, Darshak G. Thakore, "A Survey on Moving
Method Ref. Occlusion Training No. of Object Detection and Tracking in Video Surveillance
System"International Journal of Soft Computing and Engineering
rule objects (USC E) Volume-2, Issue-3, July 2012.
Kalman ...j
[5]N. Paragios, and R. Deriche.. Geodesic active contours and
[I] - Single level sets for the detection and tracking of moving objects. IEEE
filter
Palticle Trans. Patt. Analy. Mach. Intell. 22,3,266-280,2000.
filter [21] ...j - Multiple [6]RupaJi S.Rakibe, Bharati D.Patil, "Background Subtraction
Algorithm Based Human Motion Detection", Intelllational
Journal of Scientific and Research Publications, May 2013.
Method Ref. Occlusion Training No. of
[7]J.Joshan Athanesious, P.Suresh, "Systematic Survey on Object
rule ob.iects
Tracking Methods in Video", Intelllational Journal of Advanced
MHT [I] -V - Multiple
Research in Computer Engineering & Technology (UARCET)
Simple October 2012, 242-247.
Template [I] Paltially X Single [8]Sen-Ching S. Cheung and Chandrika Kamath, "Robust
Matching techniques for background subtraction in urban traffic video",
Mean Shift [I] Partially X Single Proc. ofSPIE-IS&T Electronic hnaging, SPIE Vol. 5308.
[9]Jae-Yeong Lee; Wonpil Yu, "Visual tracking by partition-
SVM [I] Paltially -V Single based histogram back projection and maximwn support criteria,"
Layering Robotics and Bio-mimetics (ROBIO), 20 II IEEE International
based [I] Full X Multiple Conference on , vol. , no. , pp.2860,2865, 7-1 1 Dec. 201.
Tracking [IO]Abhishek Kumar Chauhan, Prashant Krishan, "Moving
Shape Object Tracking Using Gaussian Mixture Model And Optical
[I] X X Single
Matching Flow", International Journal of Advanced Research in Computer
Contour ...j Science and Software Engineering, April 2013.
[I] F Multiple
Matching

473
 [11]Hitesh A Patel, Darshak G Thakore,"Moving Object
Tracking Using Kalman Filter", International Journal of
Computer Science and Mobile Computing,April 2013, pg.326 -
332.
[12] M.Sankari, C. Meena, "Estimation of Dynamic Background
and Object Detection in Noisy Visual Surveillance",
International Journal of Advanced Computer Science and
Applications, 20 11 , 77-83.
[13]J.Joshan Athanesious, P.Suresh, "Systematic Survey on
Object Tracking Methods in Video", International Journal of
Advanced Research in Computer Engineering & Technology
(UARCET) October 2012, 242-247.
[14]Ruolin Zhang, Jian Ding, "Object Tracking and Detecting
Based on Adaptive Background Subtraction", International
Workshop on Infonnation and Electronics Engineering, 2012,
1351-1355.
[15]Patrick Sebastian, Yap Vooi Voon, Comley"Colour Space
Effect on Tracking in Video Surveillance", International Journal
on Electrical Engineering and Infonnatics - Volume 2, Number
4, 2010.
[16] Maria Isabel Ribeiro "Introduction to Kalman Filtering",
Institue of Super Tehnico, June 2000.
[17] Salil P. Banerjee, Kris Pallipuram, "Multi person Tracking
Using Kalman Filter", December 10, 2008.
[18]Mr.Joshan Athanesious J; Mr. Suresh.P "Implementation
and Comparison of Kernel and Silhouette Based Object
Tracking", International Journal of Advanced Research in
Computer Engineering & Technology, March 2013, pp 1298-
1303.
[19]. Rahul Mishra, Mahesh K. Chouhan, Dr. Dhiiraj
Nitnawwre, " Multiple Object Tracking by Kernel Based
Centroid Method for Improve Localization", International
Joumal of Advanced Research in Computer Science and
Software Engineering, July-2012, pp 137-140.
[20] Qinghai Bai, "Analysis of Particle Swarm Optimization
Algorithm", Computer and Information Science, Vol.3 , February
2010, pp 180-184.

474

View publication stats