PEOPLE COUNTING AND TRACKING USING MACHINE LEARNING

Mr.K.Kalaiarasan1,B.Rabinraj2, S.Karuppusamy3, K.S.Vetrivel4

1
Assistant Professor, Department of Information Technology, M.Kumarasamy College of
Engineering,Karur,TamilNadu,India-639113.
2,3,4
UG Students, Department of Information Technology, M.Kumarasamy College of
Engineering,Karur,TamilNadu,India-639113.

ABSTRACT
The demand for people counting and tracking systems for surveillance applications is rapidly increasing,
necessitating the development of more precise, efficient, and dependable systems. The major goal of this
project is to create an accurate, long-lasting, and efficient system for counting and tracking people in
public spaces. The main goal of this study is to create a system that can work in a variety of
environments, including varying orientations, densities, and backdrops. AI calculations make a model in
light of preparing information, which is a sample of data. The primary goal of this project is to keep track
of and count people and determine the number of things. [11]This hybrid approach has the advantage of
being able to count persons in a crowd with great precision utilizing a high definition camera. [13]. The
HOG (Histogram of Oriented Gradients) technique is utilized to identify detecting, tracking, whereas the
CNN algorithm is utilized to count objects and recognize faces [4], [5], [6]. This research is used to
develop object tracking technologies that are less computationally intensive. In a single algorithm, the
concepts of object detection and object tracking are combined.An ROIs based pedestrian detection system
for single images [7].

Keywords: OpenCV, Detecting, People Counting, Tracking, Surveillance based spotting, Centroid
Tracker, High Definition Camera, CNN, HOG algorithm.

1. INTRODUCTION

People counting and tracking are done using deep learning mechanisms, and thanks to advances in image
processing algorithms and computer technology, the utilization of camcorders to track and count
individuals has expanded significantly in recent years. There are a variety of techniques and technologies
available to do this task, and due to its importance in a wide range of public space applications,
commercial goods offer a variety of options. A Bayesian computer vision system for modeling human
interactions is done by using these algorithms. [15] The functional distinctions between these goods can
be visualised as a pyramid. In terms of functionality, systems that count the number of individuals are the
most basic. A traffic flow detection system combining optical flow and shadow removal is done by the
HOG and CNN algorithms. [9] Another option is to utilise video cameras to count persons entering or
exiting a building through a door. Video cameras are commonly used as sensors in systems that track
individuals. A method of counting the passing people by using the stereo images is implemented by using

1
the HOG and CNN algorithms. [16] It can be tracked while they are in the field of vision of the camera,
resulting in a more accurate representation of their actions.

2. LITERATURE SURVEY

Title:Image Motion Analysis Year: 2020Author:NanangCahyadiAbstract:Individuals counting

frameworks can give factual pattern data used to human conduct investigation. Instinctively individuals
counting framework works in view of succession discovery. In any case, with dynamic exploration
exercises in Artificial Intelligence in Computer Vision, there are numerous choices accessible for carrying
out a group counting framework.

3. METHODOLOGIES

The methods used are profound sort, centroid tracker, CSRT, etc, which track the distinguished objects.
The steps and methodologies followed are:

• Draw an ideal reference line on the information outline.

• Recognize individuals utilizing the item identification.
• Note the centroid on the recognized individual.
• Track the development of that obvious centroid.
• Compute the centroid development Count the quantity of individuals coming in or leaving a reference
line.
• In light of the counting, increase the up or down counter.
4. RELATED WORK

Object tracking and counting were not first offered as machine learning approaches. The present system
detects, tracks, and counts objects using a deep learning method in machine learning. The main goal of
this project is object detection and determining whether there are any instances of things from defined
categories (such as humans, vehicles, bicycles, dogs, or cats) in a picture and, if present, returning the
spatial location and human behavior extent of each object instance.[10] The advantage of this hybrid
technique is that it can count people even in a crowd with great accuracy utilizing a high definition
camera. The HOG (Histogram of Oriented Gradients) technique is utilized to identify tracking, while the
CNN algorithm is utilized to count objects and recognize faces. [1], [12]

5. PROPOSEDSYSTEM

The main purpose of this project is to tracking and counting of people and detect the number of the
objects. The advantage of this half breed approach is that can apply exceptionally accuracy using high
definition camera and count the people even in the crowd. HOG (Histogram of Oriented Gradients)
algorithm is used for detecting, tracking and counting of objects such as body parts and face recognition.
[14] This project is used for object tracking techniques without as a large part of the computational
weight.

2
 5.1 System Setup

An above camera connected via USB link will be mounted pointing downwards or marginally
inclined.The camera's height has a direct impact on the likelihood of detection. The higher the camera
stature, the broader the camera's field of vision and, as a result, the larger the monitoring area.

5.2 Camera choice

There are two kinds of cameras: fixed cameras and PTZ cameras. They can then be partitioned into two
sorts: shading and non-shading. Notwithstanding, any type of camera might give information to a PC or
control unit. A 25 business minimal expense camera was used for this venture. The nearest centering
distance of the camera used is 4.8mm. It has a top notch sensor, high edge rate video playback, and a
base responsiveness of 2.0V/Sec. It additionally upholds great AVI ,as well as picture stockpiling in
BMP and JPG designs. Its result outline has a goal of 640x480, which is great for the recognizing
framework.

5.3 Image Processing Library

The people counting system in this project is built utilizing Microsoft Visual Studio and theOpen Source
Computer Vision Library for identifying social and behavioral sciences. [3] Because of technological
advancements, OpenCV is sufficient to support the whole detection process.[2] There are several
functions in OpenCV's default library that may be used to handle video feeds from various sources.

5.4 Object Detecting

We run our computationally more costly article tracker during the detection phase to determine if new
items have entered our vision and to check in the event that we can find protests that were "lost" during
the following stage. We construct or then again update an article tracker with the updated bounding box
coordinates for each identified object. We only perform this step once every N frames since our item
indicator is all the more computationally intensive.

5.5 Object Counting

We establish an object tracker for each of our identified objects to monitor the object as it travels across
the edge. Our article tracker ought to be more efficient and quicker than the item identifier. We'll
continue to follow until we reach the N-th frame, at which point we'll re-run our item indicator. The
whole procedure is then repeated. We plan to use such a global positioning framework to improve our
people and object counting.

6. IMPLEMENTATION

To execute our people counter. We'll utilize both OpenCV and dlib to assemble our people counter. For
standard PC vision/picture handling capacities, we'll utilize OpenCV, and for individuals counting, we'll
3
utilize the profound learning object indicator. Then, at that point, we'll utilize dlib to carry out
connection channels. We might have utilized OpenCV all things considered, yet the dlib object
following execution. Initially the video sequences of the object is detected based on motion
segmentation and the object is classified with Gaussian filter and then track the human and count the
people as shown in the fig.6.1.

Fig 6.1 Flow chart of Tracking and Counting People

6.1 UseCaseDiagram
A use case diagram depicts a framework's dynamic way of behaving. The object is initially detected
by the people counting and tracking system. By following the input dataset it tracks and counts the
number of projects with high accuracy and determines whether the object is moving or static as shown
in the fig.6.1.1. It is accomplished by employing a high-definition camera that focuses with pinpoint
accuracy. The advantage of this hybrid approach is that it can count people even in a crowd with high
accuracy using a high definition camera. [8], The HOG (Histogram of Oriented Gradients) algorithm
is used for object tracking and the CNN algorithm for object counting and face recognition as shown
in the fig.6.1.2. A single algorithm employs the concepts of objectdetection and object tracking

Fig.6.1.1 Sequence Diagram of people Counting and Tracking

Fig. 6.1.2 UML Diagram of People Counting and Tracking Result Analysis

4
7. RESULT AND DISCUSSION

It was successful in developing a counting system that could detect and count people. This project
made use of a low-cost commercialised upward camera, open source programming, and a control unit
like a CPU.

8. CONCLUSION
This research led in the creation of a counting system that can identify and count individuals. This
project makes use of a low-cost commercialized overhead camera, open source programming, and a
control device such as a CPU. The construction of the passerby counting framework might be based
on the HOG and R-CNN algorithms, which are utilized in object detection and counting.

9. FUTURE SCOPE

In future, effective research is required to remove the complexities by shadows. Counting and tracking
of people even in very much crowded area with high definition of image with very high accuracy can
be implemented by using additional algorithms like Single Shot Detector (SSD), Faster RCNN, YOLO
in future and it can be used to implement real time tracking and counting of people and objects.

10. REFERENCE

[1] A Fuzzy based High-Resolution Multi-View Deep CNN for Breast Cancer Diagnosis through
SVM Classifier on Visual Analysis”, Journal of Intelligent & Fuzzy Systems, IOS Press, DOI:
10.3233/JIFS-189174, Page 1-14, December 2020, vol.39, no.6, pp. 8573-8586, 2020.
[2] R.Kasthuri, A Co-Operative Bait Detection For Against Malvolent NodesAttacks in Manet,
International Journal on Concurrent Applied Research in Engineering and Management, Vol 1(4),
April 2016, ISSN (O):2395-7425, DOI: 10.17950/ijer/v5i4/012.
[3] Cetinkaya, H. and Akcay, M. (2015). People Counting at Campuses. Procedia - Social and
Behavioral Sciences, Vol. 182, pp.732-736. Vol. 10, 2015, DOI: 10.1016.sbspro.2015.04.821.
[4] Yan-yan, C., Ning, C., Yu-yang, Z., Ke-han, W. and Wei-wei, Z. (2014). Pedestrian Detection and
Tracking for Counting Applications in Metro Station. Discrete Dynamics in Nature and Society, 2014,

5
pp. no 1-11. Vol. 3, Issue 5, 2014 Kowcika.A, “People Counting System- A Review”,
DOI:10.1155/2014/712041.
[5] Chen, Y., Guo, S., Zhang, B. and Du, K. (2013). A Pedestrian Detection and Tracking System
Based on Video Processing Technology. 2013 Fourth Global Congress on Intelligent Systems pp.
1205-1208, DOI:10.1109/GCIS.2013.17.
[6] Wang, X., Sun, J. and Peng, H. (2013). Foreground Object Detecting Algorithm based on Mixture
of Gaussian and Kalman Filter in Video Surveillance. JCP, 8(3), vol 08, no.3, pp. 1479-1489, 2013,
DOI:10.4304/ jcp.8.3.693-700.
[7] Shou, N., Peng, H., Wang, H., Meng, L. and Du, K. (2012). An ROIs based pedestrian detection
system for single images. 2012 5th International Congress on Image and Signal Processing, pages 49–
58, DOI: 10.1109/GCIS.2012.17.
[8] G.Xiong,J.Cheng,X.Wu,Y.Chen,Y.Ou,andY.Xu,“A approach for people counting in aovercrowded
behavior detection,” Neurocomputing, 2012 Vol. 83, pp. 121-135, 2012. DOI: 10.1155/2020/669225
[9] Zhaoxiang Zhang, YuqingHou, Yunhong Wang, Jie Qin (2011). A traffic flow detection system
combining optical flow and shadow removal. 2011 Third Chinese Conference on Intelligent Visual
Surveillance. DOI no: 10.1109/IVSurv.2011.6157021, INSPEC Accession number: 12572190.
[10] Candamo, J., Shreve, M., Goldgof, D., Sapper, D. and Kasturi, R. (2010). Understanding Transit
Scenes: A Survey on Human Behavior-Recognition Algorithms. IEEE Trans. Intell. Transport. Syst.,
11(1), pp.206-224.DOI:10.1109/TITS.2030963.
[11] Zhao, X., Dellandrea, E. and Chen, L. (2009). A People Counting System based on Face
Detection and Tracking in a Video. Advanced Video and Signal Based Surveillance, pp.67-72.Vol. 41,
2011, DOI: 10.1109/AVSS.2009.45.
[12] Green-Roesel, R., Diógenes, M., Ragland, D. and Lindau, L. (2008). Effectiveness of a
Commercially Available Automated Pedestrian Counting Device in Urban Environments: Comparison
with Manual Counts, DOI: 10.3141/2140-01.
[13] Velipasalar, S., Tian, Y. and Hampapur, A. (2006). Automatic Counting of Interacting People by
using a Single Uncalibrated Camera. 2006 IEEE International Conference on Multimedia and Expo,
vol. 2, pp. 267-271, DOI: 10.1109, ICME.2006.262768.
[14] R.Bharathi, T.Abirami,” Energy efficient compressive sensing with predictive model for IoT
based medical data transmission”, Journal of Ambient Intelligence and Humanized Computing,
November 2020, https://doi.org/10.1007/s12652-020-02670-z
[15] Oliver, N., Rosario, B. and Pentland, A. (2000). A Bayesian computer vision system for modeling
human interactions. IEEE Transactions on Pattern Analysis and Machine Intelligence, 22(8), pp.831-
843, Print ISSN: 0162-8828, ElectronicISSN: 1939-3539, DOI: 10.1109/34.868684.
[16] Terada, K., Yoshida, D., Oe, S. and Yamaguchi, J. (n.d.). A method of counting the passing
people by using the stereo images. Proceedings 1999 International Conference on Image Processing
(Cat. 99CH36348), Kobe, vol. 2, pp. 338-342, 1999. ISSN: 2302-9285, DOI: 10.11591/eei.v9il.1849.