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

net/publication/336881210

PEOPLE COUNTING TECHNOLOGY

Article · November 2016

DOI: 10.21884/IJMTER.2016.3142.ABPRK

CITATIONS READS

6 9,053

1 author:

Zahraa salah Dhaief

Mustansiriyah University
17 PUBLICATIONS 26 CITATIONS

SEE PROFILE

All content following this page was uploaded by Zahraa salah Dhaief on 29 October 2019.

The user has requested enhancement of the downloaded file.

 PEOPLE COUNTING TECHNOLOGY
Zahraa Salah Dhaief 1
1
Department of Computer Science, College of Education, AL-Mustansiriyah University, Baghdad-Iraq
Abstract— People Counter based on video is an important field in a Computer Vision. When people
counter is implemented it may accompany by some problems. This paper proposed a method which
may mix some existing technologies to overcome some problems. The people counter requires more
powerful processing since it deals with real-time video, so the proposed method converts a color
image into binary in order to minimize data of image. Also the image may contain a noise, the
proposed method uses Erosion and Dilation processes to erase the noises. Reducing development
time is important term in Software Engineering to build an application. The method depends on
existing packages for reducing development time. The proposed method is implemented in Java
programming language.
Keywords— People counting; Ultrasonic Sensor; Video counter; frame; Blob Detection
I. INTRODUCTION
The owner’s knowledge of how many and when people are inside the building, company or
shopping mall, help him to optimize the scheduling of labor and monitor the promotional events
effectiveness. Security measure can benefit from people counting; this help to determine the number
of guards can be assigned [1].
There are several people counting technologies as:
1. Ultrasonic Sensor: there should be cluster of tree-node sensors; each node mounts an ultrasonic
area. Wide area needs multiple clusters, coordinating node of cluster reads form nodes by RF
link. Distributed algorithm of nodes decides counting process for detected people. Such system
needs clock synchronization at millisecond level to exchange data simultaneously. Protocol of
clock synchronization is imposing a disadvantage of this technology [2].
2. Infrared sensor array: a system requires IR sensors matrix which form the detectors. The matrix
provides sensor signals. The algorithms of pattern recognition are detecting people moving
across sensor area. Such system require power processing and synchronization for detecting
people [3].
3. Infrared Motion Sensors: three PIR sensors are required for each passage monitored. A
coordinator connects to the sensors by a wireless RF link. The coordinator receive the events of
detected motion from the sender. The coordinator concludes people count from correlating phase,
number and time difference of signal peaks. PIR sensors are alternative to previous technology,
however the effort and cost of installing multiple nodes of sensor for each Surveillance area is a
cost-side disadvantage [4].
4. Sensor Fusion: The system consists of PIR sensors, CO2 and camera. It depends on a Hidden
Markovian Model that relays on an Filter of Extended Kalman to derive building occupancy.
This technology combines readings of current sensor and historical data to estimate the true state
of the system, adjusting for stochastic processes and sensor noise [5].
5. Video counter: it based on ceiling mounted camera; people are identified by background
subtraction of image. The objects (blobs) are identified and their size estimated and compared to
pixels dimension of people which established previously. This analysis leads to people count [6].
This technology is better than previous ones because [7]:
1. Count is bidirectional.
2. The counting algorithm is flexible for unexpected situations.

DOI:10.21884/IJMTER.2016.3142.ABPRK 210
 International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 03, Issue 11, [November– 2016] ISSN (Online):2349–9745 ; ISSN (Print):2393-8161

3. Surveillance of large area.

4. Surveillance area of high traffic.
5. It is integrated with online systems or database easily.
6. Installation process is done easily
Honglian et al. (2008) propose a method for segmenting multiple people relay on the
histogram of frame differencing image that has bi-directional projection of gray scale. Use the gray
value information of projection for the detection is insufficient for this system. When the projection
of any movable targets is same as the targets of people, there is a probability to occur an error in
counting [8]. Chen et al. (2008) propose an approach that uses a zenithal camera for counting the
number of persons riding and getting out a bus. The approach at first divided the frame to multiple
blocks then blocks to analogous vectors of motion as concerning same object [9]. Hartono et al.
(2006) propose an approach for counting number of people, it firstly find out the pedestrian as dots
and acts as binary masks after that track the pedestrian using correlation based algorithm for
consecutive frames[10]. Tsong-Yi et al. (2010) propose an algorithm for segmentation and detection
of motion object depends on algorithm of frame difference where the counting feature is taken from
width and height of the detected people. To detect and track moving people [11]. Yan-yan et al.
(2014) present algorithm for template-matching and method for counting pedestrian based on Haar-
like detection. The method is aimed to count pedestrians that are existing in a metro station via
automatic surveillance using video camera[12].This paper introduce an effort to detect, track and
count people through proposed method. The proposed method tries to overcome some problem such
as power of processing, changing background features and noises erasing. This is leading to enhance
the operation of surveillance implementation accurately.
II. METHOD ARCHITECTURE
The proposed method concerns by counting persons who entered a gate. There must be a
camera fixed on the gate ceiling. A scene provided by a camera represents the observing area where
pictures of pedestrian is captured in order to detect, track and count them. It uses the following steps
to track and count people:
1- Observing phase: it represents alerting phase which determine the moving objects when they
are entering the scene.
2- Tracing phase: it tracks the motion of object within observing area.
3- Counting phase: it the most interest for counting the people according to tracing phase.
The proposed method uses data which is funding from camera video which is composed of many
frames. Captured frame must be processed individually to obtain people (blob). After that, blob
detection and tracking are handled. Then blob is counted.
III. APPROACH ALGORITHM
The following algorithm represents abstraction of proposed method:
STEP1: Open the socket to connect to the camera.
STEP2: Send command to set configuration of the camera.
STEP3: Request video via socket.
STEP4: Receive video packets and write them to file as JPEG.
STEP5: Check if frames end THEN GOTO step 4
ELSE
Read the file in Open CV IpI Image.
STEP6: Detect an Object.
STEP7: Check if object is detected THEN Initialize cv Blob and repeat STEPS 4, 5, 6
ELSE
GO TO STEP 4.
STEP8: Track the object via cvBlob.

@IJMTER-2016, All rights Reserved 211

 International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 03, Issue 11, [November– 2016] ISSN (Online):2349–9745 ; ISSN (Print):2393-8161

STEP9: Update size and location of object.

STEP10: Check if the object is lost THEN GOTO STEP 3
ELSE
Check if Delay Over THEN GOTO STEP 4
ELSE
Send pan, tilt and zoom command through socket to the camera.
STEP11: Reset delay.
In order to optimize the tracking and counting people approach there should be three
methods, they are used to perform above phases:
1- Enhancing the Image
2- Detecting the Blob
3- Tracking the Blob
These steps implemented as the following operations:
3.1 Background Segmentation
People counting are a field of Computer Vision systems, they are requiring separation
between background and moving objects (people). This paper focus on background subtraction to do
this separation. The background image should be stored which acts as reference image to compare
with other images (frames). Performance of the approach is affected by background estimation, since
there is factors affect it as, changing light condition with the time, the background update when it
uses slow recursive function to capture changes of background.
People entering observing area should not stop to solve these problem. The solution depends on
accumulator which is image contains accumulation of the images changing with the time. So
subtraction of the accumulator from current image will produce the objects (foreground). But the
result image may contain noises, deletion of noises provide peter detection.
3.2 Binarization
The major drawback of the video people counter system, that it requires more processing and
resources to operate accuracy. So minimizing data of processing can be useful for optimizing the
resources and reducing time of processing. The proposed method uses a threshold or transform the
image to binary, which can be helpful for this purpose. Binary image can be obtained by splitting and
threshold the pixels into white and black on gray image. Value of default threshold that is convenient
for most videos is 70, a special situation can be found such the differences between high brightness
and average value, so there is no default threshold uses. The value is not appropriate to be chose by
user because the high value will cause damage. After Binarization the result image is containing
white and black colors only, the values greater than 70 will become white and the lesser one will
become black.
3.3 Erosion, Dilation processes
The image obtained from Binarization process may contain noises, so the erosion process is
used in the method for clearing the image from this noise. It is used multiple times to erode the
borders between connected components of binary image. The overlap, noises or other link that can
appear between different components are separating using erosion process. Erosion is not sufficient
because the object may contain black area, so the shape is irregular. The black region is filled using
dilation process in iterative manner.
3.4 Object Detection and Tracking
The Object Detection (Blob Detection or Blob analysis) is a process that is used in the
proposed method to analysis an image for detecting foreground objects. It may a difficult process in
some cases such as when the object is same color as the background, the object may contain holes.
Object is detected by finding all adjacent pixels and collect them to form the object. Blob object is
hard to analyze when it composes of multiple people. This case may occur in crowded areas, the
solution in the method is done by splitting the large blob object into smaller one based on height and

@IJMTER-2016, All rights Reserved 212

 International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 03, Issue 11, [November– 2016] ISSN (Online):2349–9745 ; ISSN (Print):2393-8161

width. The features are taken into account when blob object is detecting is a position, height and
width. This features are analyzed according to coordinates of X and Y values. Knowledge of smallest
and largest values of X and Y is helpful way to determine the object features by subtracting the
smallest from the largest one. Blob object is determined as rectangle, it must sort these objects by
putting them in a dynamic list. Link list is used to easily add, delete, merge and split the specified
objects. The rectangle represents counting area, the algorithm scan counting area from most left hand
side to most right hand side. After Binarization, Erosion and Dilation processes the object is
represented as white blob. When scanning process find white pixel, algorithm records it as object
blob. The iterative process of scanning will form entire object. A person is acted by 50-65 width, the
result of counted people obtains by dividing the width of blob by person width and round it to the
nearest integer. Some margin may found, therefore person width specify to 46. For example object
blob is 86, so 86/46= 1.8 is round to 2 (number of people). Object (blob) tracking aims to analyze
object path. It is implemented in this paper depending on cvBlob library, this library puts labels for
detected object to track them among frames. The proposed method is implemented in Java 6.0 as
People Counter Method (PCM) application as in Figure 1.

Figure 1. People counter method implementation

IV. RESULTS
The video counter is dealt with Real-Time images which require high processing ability; this
requires more Central Processing Unit (CPU) overhead. The proposed method reduces CPU
overhead via converting the images to binary image.
The proposed method uses multiple detection processes to obtain better result for people
detection. It uses Background Segmentation and Object Detection techniques. As well as it separate
large blob into small ones and reduce people overlap through camera position. This produces better
people detection.
The proposed method save Real-Time operation, since it uses link list for recording peoples
and fasting of processing. It reduces development time when it uses existing packages to minimize
development overhead.
There are no need for shadow extraction in proposed method, because it deals with shadow as
noises. This leads to more speeding for processing, because it reduces processing operations.

@IJMTER-2016, All rights Reserved 213

 International Journal of Modern Trends in Engineering and Research (IJMTER)
Volume 03, Issue 11, [November– 2016] ISSN (Online):2349–9745 ; ISSN (Print):2393-8161

V. CONCLUSION
The proposed method provides faster processing for people counter application, because it
depends on binary image which represents the image with lesser data than color one. Background
Segmentation is a process uses to separate the background and foreground, but it isn’t sufficient
process because the resulting image may contain noises. In addition to Background Segmentation the
proposed method uses Object Detection method in order to detect people accurately. Background
Segmentation isn’t accurate when the feature of background changes, therefore, the proposed method
uses accumulator image to be convenient with timely changing.
VI. ACKNOWLEDGMENT
The author would like to thank AL-Mustansiriyah University (www. uomustansiriyah.edu.iq)
Baghdad-Iraq for its support in the present work.
REFERENCES
[1] R. Berg, “Real-time people counting system using video camera”, Department of Computer Science and Media
Technology, Gjøvik University College, Norway, 2007.
[2] Q. Chen, M. Gao, J. Ma, D. Zhang, L. M. Ni, and Y. Liu. “Moving Object Counting using Ultrasonic Sensor
Networks”. International Journal of Sensor Networks, vol. 3, 2008.
[3] K. Hashimoto, K. Morinaka, N. Yoshiike, C. Kawaguchi, and S. Matsueda, “People Count System using Multi-
Sensing Application”. In Proceedings of International Conference on Solid State Sensors and Actuators,
TRANSDUCERS, vol. 2, 1997.
[4] P. Zappi, E. Farella, and L. Benini “Enhancing the Spatial Resolution of Presence Detection in a PIR based
Wireless Surveillance Network”, In Proceedings of IEEE Conference on Advanced Video and Signal Based
Surveillance, AVSS, pp. 295 –300, 2007.
[5] S. Meyn, A. Surana, Yiqing Lin, S.M. Oggianu, S. Narayanan, and T.A. Frewen.A “Sensor-Utility-Network
Method for Estimation of Occupancy in Buildings”, In Proceedings of IEEE Conference on Decision and Control,
held jointly with the Chinese Control Conference, CDC/CCC, pp.1494 –1500 , 2009.
[6] S. Velipasalar, Y. Tian, and A. Hampapur, “Automatic Counting of Interacting People by using a Single
Uncalibrated Camera”, In Proceedings of IEEE International Conference on Multimedia and Expo, 2006.
[7] S. Mezei AND A. Sergiu, “A Computer Vision Approach to Object Tracking and Counting”, DARABANT
STUDIA UNIV. BABES_BOLYAI, INFORMATICA, Vol. LV, No. 3, 2010.
[8] M. Honglian, L. Huchuan, Z. Mingxiu , “A Real-time Effective System for Tracking Passing People Using a Single
Camera”, Proceedings of the 7thWorld Congress on Intelligent Control and Automation, Chongqing, China, pp.
6173-6177, 2008.
[9] C.H. Chen, Y.C. Chang, T.Y. Chen, D.J. Wang, “People Counting System for Getting In/Out of a Bus Based on
Video Processing”, IEEE Computer Society, Eighth International Conference on Intelligent Systems Design and
Applications, pp. 565-569, 2008.
[10] S. Hartono, T. Ji, T. Yap-Peng , “People Counting by Video Segmentation and Tracking”, 9th international
Conference on Control, Automation, Robotics and Vision, pp. 1- 4, 2006.
[11] C. Tsong-Yi , C. Chao-Ho , W. Da-Jinn , C. Tsang-Jie , “Real-Time Counting Method for a Crowd of Moving
People”, Sixth International Conference on Intelligent Information Hiding and Multimedia Signal Processing, pp.
643-646, 2010.
[12] C. Yan-yan, C. Ning, Z. Yu-yang, W. Ke-han, and Z. Wei-wei “Pedestrian Detection and Tracking for Counting
Applications in Metro Station”, Hindawi Publishing Corporation, Discrete Dynamics in Nature and Society ,
Article ID 712041, 2014.

@IJMTER-2016, All rights Reserved 214

View publication stats