CROWD MONITORING SYSTEM USING

IMAGE PROCESSING
At

Sathyabama Institute of Science and Technology(Deemed

to be University)

Submitted in partial fulfillment of the requirements for the award of Bachelor of Engineering
Degree in Computer Science and Engineering
By

[Link]
([Link].38110527)
&
[Link]
([Link].38110638)

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING SCHOOL OF

COMPUTING

SATHYABAMA INSTITUTE OF SCIENCE AND TECHNOLOGY JEPPIAAR

NAGAR, RAJIV GANDHI SALAI, CHENNAI – 600119, TAMILNADU

1
 SATHYABAMA
INSTITUTE OF SCIENCE AND TECHNOLOGY
(DEEMED TO BE UNIVERSITY)
Accredited with Grade “A” by NAAC
(Established under Section 3 of UGC Act, 1956)
JEPPIAAR NAGAR, RAJIV GANDHI SALAI
CHENNAI– 600119
[Link]

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

BONAFIDE CERTIFICATE

This is to certify that this Project Report is the bonafide work of [Link] (38110527) and
[Link] ([Link].38110638) who carried out the project entitled “Crowd Monitoring
System Using Image Processing” under my supervision from September 2021 to November 2021.

Internal Guide

Dr. P. JEYANTHI.,

Head of the Department

Dr. L. LAKSHMANAN, M.E., Ph.D.,

Submitted for Viva voce Examination held on

Internal Examiner External Examiner

2
 DECLARATION

[Link] and [Link] hereby declare that the Project Reportentitled CROWD
MONITORING SYSTEM USING IMAGE PROCESSING is done by me under the
guidance of [Link],.[Link].D, Associate professor, Dept of CSE at
SATHYABAMA INSTITUTE OF SCIENCE AND TECHNOLOGYis submitted in partial
fulfillment of the requirements for the award of Bachelor of Engineering degreein Computer
Science and Engineering.

DATE:

PLACE:

SIGNATURE OF THECANDIDATE

3
 ABSTRACT
A huge crowd of people visit public places seasonally. Crowd management
is critical in these situations in order to avoid public disasters like stampede,
confusion, social distance, danger of pandemic etc. Therefore, a need arose
in order to manage these crowds to minimize the rate of pandemic and
avoid spreading. This framework would be smart and efficient. The
framework is based on computer vision with a fairly simple and easy to use
interface. The framework is comprised of three layers: sensor, management,
and interface layers. The sensor layer is responsible for gathering the data
of number of people getting into a room. The management phase is
responsible for checking whether the restricted count of a particular room is
attained or not. The interface layer is responsible for alerting the security
officials with a buzzer to convey that the restricted count has been reached
and no more civilians are allowed inside the room. This setup is applicable
in closed space like a room, shop, conference room, lift, etc. In wide area,
we implement OpenCV concepts using python to measure the distance
between each individual and warn them if they exceed the distance limit
among the crowd. The assumed distance between two individuals is
calculated based on the pixel position of a particular person in the scene.
We use image processing to generate the result. If the limit among the
individuals exceed, then it will show warning in red color, otherwise they
are good to go.
Keywords: OpenCV

4
 LIST OF FIGURES
4.1 Architecture Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.2 Data Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

6.1 Output 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.2 Output 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

5
 TABLE OF CONTENTS

Title Page. No

ABSTRACT v

LIST OF FIGURES vi

LIST OF TABLES vii

LIST OF ACRONYMS AND ABBREVIATIONS viii

1 INTRODUCTION 12
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 LITERATURE REVIEW 16

3 Aim and Scope of the present investigation 21

3.1 Aim of project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Project domain . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 22
3.3 Scope of project. . .. . . .. . .. . . . . . . . .. . . . . . . . . . . . 24

3.4 System Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.4.1 Hardware Specification . . . . . . . . . . . . . . . . . . . . 25
3.4.2 Software Specification . . . . . . . . . . . . . . . . . . . . . 25

4 Experimental or materials and methods; algorithms used 27

4.1 General Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
 4.2 Design Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
IMPLEMENTATION AND TESTING 33
5.1 Input and Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.1.1 Input Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.1.2 Output Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.2 Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.3 Types Of Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.3.1 Unit Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.3.2 Functional Testing . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3.3Integration Testing . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3.4 White Box Testing . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3.5 Black Box Testing . . . . . . . . . . . . . . . . . . . . . . . . 36
5.4 Testing Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

5 RESULTS AND DISCUSSIONS 37

6.1 Efficiency of the Proposed System . . . . . . . . . . . . . . . . . 37
6.3 Advantages of the Proposed System . . . . . . . . . . . . . . . . 39
6.4 Sample Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

6 CONCLUSION AND FUTURE ENHANCEMENTS 45

7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.2 Future Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
SOURCE CODE 48
9.1 Source code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
References 56
 Chapter 1

INTRODUCTION

1.1 Introduction

Crowding is a common phenomenon observed during major events

such as concerts, festivals, sports, games, and entertainment. One of the
most interesting and active research topic in computer vision is the
analysis of crowd behavior. Crowd is a group of people gathered in a
certain location. Crowd differs in different situations like crowd in a
temple will be different from the crowd in a shopping area. Crowd is a
group of individuals sharing a common physical location. Now a day’s
increase in human population tends to increase the crowd in public areas
[3].Thus it is required to analyze the surveillance system with several
closed circuit Television is used to monitor the crowd. The human eye
cannot observe all the cameras simultaneously. Thus an automated
technique must be used for continuously monitoring the crowd for a long
period. Challenging problems in detecting the desired events automatically
are that, simultaneous occurrence of more than one events, large number of
data must be processed, occlusions and real time detection. The proposed
method can be applied from small group of object. The Internet of Things
is a powerful industrial system of radio-frequency identification and
wireless devices that have the ability to transfer data over a network
without needing human interaction [1]. Analysis of a crowd behavior using
surveillance videos is an important issue for public security, as it allows
detection of dangerous crowds and where they are headed. [4] Computer
vision based crowd analysis algorithms can be divided into three groups
people counting, people tracking and crowd behavior analysis. Mainly,
IoT consists of three layers, the sensing layer to gather data from real
world via existing hardware e.g. sensors, next the network layer to transfer
the collected data over wired or wireless network, and the application layer
which is responsible for twoway communication between user and
systems.

Chapter 2
LITERATURE REVIEW

S. Hashish, and M. Ahmed (2015). "Efficient wireless sensor

network rings overlay for crowd management in Arafat area of
Makkah." In 2015 IEEE International Conference on Signal
Processing, Informatics, Communication and Energy Systems, pp. 1-6.
Researchers have started utilizing sensor enabled Smartphones as a tag for
large scale human sensing. With the increasing usage of Smartphones,
more persons can be tracked without providing any tag in future. Some of
the Smartphone based location tracking systems require an application to
be installed on the client’s Smartphone. The installed application obtains
the location using GPS sensor of Smartphone and continuously updates the
location to the remote server using Internet connection. However, it is rare
that users in the large crowd and in remote locations will have the Internet
connection all the time. There are many operating systems and versions for
Smartphones, which makes the development and distribution of
application a difficult task. By embracing the Smart City paradigm, crowd
sensing becomes a solution able to cope with air pollution monitoring
since this novel paradigm assumes that a significant number of users
perform collaborative sensing tasks, thereby collecting data from different
populated locations while doing their daily activities. The collected data is
periodically transmitted to a central server for data storage and processing.

D. C. Duives, T. van Oijen, and S. P. Hoogendoorn (2020), “Enhancing

crowd monitoring system functionality through data fusion:
Estimating flow rate from Wi-Fi traces and automated counting
system data,” IEEE Sensors, vol. 20, no. 21, p. 6032.
With the development of sensor technology, communication technology,
and big data science, smart city-oriented intelligent applications have
become important services in human life. People’s living standard has
increased with improved infrastructure and intelligent applications such as
smart home furnishing, smart building. With the increasing expansion and
prosperity of urban business zones, some people choose to go shopping
and seek entertainment in business zones. These large central business
zones have become representative of the city image and are the zones with
the most economic vitality. Besides,of people. By estimating the number of
people, the degree of crowd gathering can be judged for accurate and
effective management and planning at a monitored site. Traditional crowd
monitoring systems depend on vision-based monitoring technology, the
most commonly used method of which is closed-circuit television
monitoring. In other words, the signal can be transmitted from the data
source to a prearranged specific broadcast device connected to the
[Link] fusion is a kind of information fusion technology that
associates, correlates, and combines the information from multiple sensor
to obtain more timely and accurate decision making support. From
low-level data collection to high level services, data fusion offers feasible
and high-efficiency support for deep fusion and mining massive
multi-source data in heterogeneous networks
Sebastian Bek and Eduardo Monari (2015), “The Crowd Congestion
Level A New Measure For Risk Assessment In Video Based Crowd
Monitoring”, International Journal of Information Technology and
Decision Making 3(4):2353- 2361.
There are more categories of crowd counting in video sensing, one is
region of interest counting which estimates the total number of people that
regions at an instance and the other is line of interest which counts number
of people who crosses a detecting line in an instance. The line of interest
counting can be developed using feature tracking techniques where they
are either calculated into trajectories and these trajectories are clubbed into
object tracks or based on counting crowd densities from a temporal frame
of the [Link] in a crowded scene should be regarded as endangered
in case in a certain local region of the scene a too large number of people
is observed. However, on one hand absolute density is very difficult to
extract from crowd videos, and on the other hand density alone is not a
sufficient characteristic figure, since for example in a concert or public
festival, a very high density is usual and not critical per se. We observed
that even in scenarios with high people density, the situation can be
regarded as non-critical, in case people can still move freely and smoothly
through the crowd. As a consequence, we believe that information on the
flow dynamics should be taken into account for risk assessment. Our
approach assumes, that a local spot in the crowd might be critical, if the
density is continuously increasing over time, and simultaneously a
significant reduction of motion dynamics is observed.
Rosario Fedele and Massimo Merenda (2017), “An IoT System for
Social Distancing and Emergency Management in Smart Cities Using
Multi-Sensor Data”, In 2017 IEEE International Conference on Signal
Processing, Informatics, Communication and Energy Systems.
Recommender systems technology aims at reducing the consumer
over-choice due to the huge amount of information available on the web.
These systems use information gathered through few interactions with its
user, to create a customized selection of items that can interest the user and
can facilitate a better user experience. Recommender systems could be
further improved if Machine Learning and Deep Learning algorithms were
used for information retrieval. Noteworthy applications of recommender
systems in smart cities include, a mobile IoT recommender system for
users that need to find Park-and-Ride infrastructures to switch from a
private to a public transportation mode, an autonomous situation-aware
evacuation route recommenderarchitecture, optimized in real time, to
obtain smart buildings, reuse data generated by various IoT applications,
and adapt the services provided by the single IoT system, to improve the
user experience. Traditionally, the problems above can be solved using
Shortest Path Finding algorithms that belong to the mathematical field of
Combinatorial Optimization. Noteworthy examples are, Prim’s algorithm,
which can be used to derive the Minimum Spanning Tree and the
Dijkstra’s and the A* algorithms, which allow one to derive the shortest
path between two points. In the Matlab environment, it is possible to find
different tools that allow one to solve this problem.
Marwa F. Mohamed, Abd El Rahman Shabayek, Mahmoud ElGayyar
(2018), “IoT-based Framework for Crowd Management”, Computer
Science Department, Faculty of Computes and Informatics Suez
Canal University, Ismailia, Egypt, IEEE International Conference on
IoT.
The Internet of Things is a powerful industrial system of radio-frequency
identificationand wireless devices that have the ability to transfer data over
a network without needing human interaction. Mainly, IoT consists of
three layers, the sensing layer to gather data from real world via existing
hardware, the network layer to transfer the collected data over wired or
wireless network, and the application layer which is responsible for two
way communication between user and systems. IoT applications are
rapidly evolving and growing in various fields. Sensing the crowd by
normal sensors and managing it is a challenging problem. Crowd
management requires several stages including crowd data acquisition via
sensor layers, data transferring via network layers, data analysis for
decision making and applying crowd control measures via application
layer. Using different sensor devices, it is possible to gather information
about visitor’s crowds and determine which areas are overcrowded. This
information is then transmitted to the management layer where the admin
can decide to close some doors and roads. The management layer shall be
equipped by a smart service that can recommend which doors to be opened
or closed to direct the crowd flow. The admins then decide whether to
publish this information to the public or not. In order to increase the user
end application usability, the application provides different languages and
auser friendly interface.

Chapter-3

1.2 Aim of the Project

The aim of this project is to detect and react to overcrowding issues

at an early stage. This will help us to assess how effective our crowd safety
precaution is and the considerations which are looked into are as follows,
monitoring the overall number of people to ensure the safe venue capacity
is not exceeded, monitoring the distribution of people to help prevent local
overcrowding, identifying potential crowd problems to prevent the
escalation of public disorder.
1.2 Project Domain

Our project come under the domain of Computer Vision where we have
used both technologies to identify overcrowded areas. The Internet of
things describes the network of physical objects that are embedded with
sensors, software, and other technologies for the purpose of connecting
and exchanging data with other devices and systems over the internet.
Computer vision is an interdisciplinary field that deals with how
computers can be made to gain high-level understanding from digital
images or videos. From the perspective of engineering, it seeks to
automate tasks that the human visual system can do. As a technological
discipline, computer vision seeks to apply its theories and models for the
construction of computer vision systems.

1.4 Scope of the Project

With booming in population and huge number of crowds converging

at common places for meetings, gatherings, and reunions there is increased
scope for these kind of projects. Our approach of identifying crowd in
open space will be definitely useful in controlling the crowds in the public
area, that too in this pandemic world crisis the importance of these kinds of
projects have created necessary need among the governments of various
countries to monitor the public crowd.

1.5 Methodology

Image Processing Method

This method is software oriented approach which is used in open space i.e.
malls, public areas and others, we used the technology of computer vision
to detect the people with the help of cameras in public area, the input to
our model is directly taken from the real time monitoring cameras installed
in public areas where the model will be installed in the system and the
camera video is taken as input, rectangles are used to highlight the crowd
violations where green rectangle insist no violations whereas red rectangle
insist that there is a crowd violation in a particular area, the crowded area
is identified with the help of pixel positions of the people in the video.

3.4 System Specification

3.4.1 Hardware Specification

❖ LCD Display Panel

❖ CCTV Camera
❖ Monitor
❖ Graphics card
❖ Peripheral Devices

3.4.2 Software Specification

❖ Python Programming Language

❖ ScipyLibrary
❖ NumpyLibrary
❖ Imutils Library
❖ Opencv-python Library
❖ Operating System - Windows
 ❖ YOLO Object Detection Library

3.4.3 Standards and Policies

The various standard versions of software which need to be downloaded
and installed are as follows,
⮚ Python - 3.6.X
⮚ Scipy - 1.4.1
⮚ Numpy- 1.18.5
⮚ Imutils- 0.5.3
⮚ Opencv-python - [Link]
⮚ Windows Operating System - 8.1 or higher
⮚ Graphics Card - Intel Graphics 10.18.10.4252 or higher
⮚ Monitor - 1366 X 768 display or higher
⮚ Communication Device - Mobile or Waketalkie

Download the python from WWW and install it, to install python libraries
use the command “pip install library-name version”, Configure the
Adruino board with battery sensor and buzzer, connect the Adruino board
to the computer, commit the necessary code to Adruino with the help of
cable.
 Chapter 4
MODULE DESCRIPTION

4.1 General Architecture

Figure 4.1: Architecture Diagram

The architecture of our project consist of a is closed place where the

architecture explains about monitoring crowds in rooms and conference
halls.
4.2 Design Phase

WORKFLOW DIAGRAM:
The above is the entity relationship diagram for our proposed model.
Where the relation between crowd monitoring system and crowd is a n to n
relationship, the relationship between open space and closed space is n to n
relationship, the relationship between terminate on open space and closed
space is a one to one relationship as the machine can be terminated only
once that is it can be switched off only once to switch off it again we need
to first power it on and then terminate so one to one relationship.

4.3 Module Description

MODULE:

Developing an Image Processing model for monitoring crowd in open

space. This is a software kind of approach where we have used computer
vision to identify crowds, it requires a minimum specification computer to
run the proposed model, the description and required specification for the
computer are already discussed in the hardware component section.

IMPLEMENTATION AND TESTING

5.1 Input and Output

5.1.1 Input Design

The input design for the module is the video captured from the real time
CCTV camera placed in any public place for which we need to monitor the
crowd and make people to spread across the area instead of accumulating
one area.

5.1.2 Output Design

The output design for this module is displayed in the same monitor in
which the video from CCTV camera is used as input it displays the
number of crowd violations in the monitor screen and also differentiates
the people in the scene those who are not violating the crowd by enclosing
them in a blue rectangular box, for those who violated the crowd it
displays them in yellow color rectangular box.

5.2 Testing

Software Testing is a method to check whether the actual software

product matches expected requirements and to ensure that software
product isdefect free. It involves execution of software or system
components using manual or automated tools to evaluate one or more
properties of interest. The purpose of software testing is to identify errors,
gaps or missing requirements in contrast to actual requirements.
5.3 Types of Testing
5.3.1 Unit Testing

The unit testing is done on major important components of the

proposed model in our IoT design, working of it is checked and verified
then the sensors where its working is tested and verified then the battery to
power up adruino and sensor is tested and verified, now the components
related to software working of the model are checked and verified the
camera used is tested for its working and verified then here to run the
python code for crowd detection we have used python which is executed in
the local machine to which the CCTV camera is connected so its setup is
verified.

5.3.2 Functional Testing

The functioning for which the components are being used is tested in
this phase whether the desired result are produced by the components. The
sensor is tested along with the adruino whether it counts the people
correctly and buzzers the alarm when the restricted count is reached. The
python code is executed on the input video from the CCTV camera to
check whether it identifies crowd in public places and distinguishes them
with green and blue rectangular boxes.

5.3.3 Integration Testing

 The combined working of the components is tested and verified in
this testing. The adruino along with the sensor and battery is tested for its
working collaboratively with each other getting attached. The camera and
the code are tested along together whether the image from camera is used
correctly for detecting crowd.

5.3.4 White Box Testing

For checking whether the data is available in the desired folder we

need to specify the correct path for it to detect the crowd from the CCTV
camera available in public places and malls for that we need to specify the
inputs for it assume the expected results you need the program to produce
should be done beforehand for better results.

5.3.5 Black Box Testing

This testing is done without the need of developer, the person who
doesn’t know anything about developing he just checks whether the
expected outputs are produced for the given inputs. For this test give a few
minutes recorded video as input and test whether the program identifies
the crowds in the video and alerts about it.
5.4 Testing Strategy
The strategy we used to test our proposed model was, first to test the
components in closed space where we checked for adruino, sensor, buzzer
integration working on producing the correct the result in the LCD screen
and also the buzzer alarming at correct time, then for closed space we
started with the specifications of the computer first then we installed our
model in the system and also supporting libraries required for our model to
monitor crowd, then we checked for the real time CCTV camera placed in
public places, as we didn’t have access to real time camera we checked our
proposed model with already prerecorded video of format mp4 and it
monitored and displayed the crowd violated area correctly.

Chapter 5

RESULTS AND DISCUSSIONS

6.1 Efficiency of the Proposed System

This project typically is a less cost effective and more economically

benefitable to the organization which builds this project and also to the
society considering the current undergoing pandemic situation. The
analysis of this project ensures in helping government organizations and
other NGO to determine their viability, cost, and benefits associated with
this project before financial resources are allocated to help the needy. So
from a technical point of view we can assure that the idea can be converted
to working system with the already ready available technologies and
hardware systems available to the organization. It also serves as an
independent project assessment and enhances project credibility helping
decision makers determine the positive economic benefits to the
organization that the proposed project will [Link] impact analysis
of our project greatly reduces the overall risks of being inhuman to the
surrounding and its people and also helps to reduce resistance, strengthens
support of the people towards the government organization, and allows for
a more comprehensive understanding of the costs and benefits of the
project.

Output
Figure 6.2: Output 1

Figure 6.3: Output 2

 Chapter 6

CONCLUSION AND FUTURE

ENHANCEMENTS

7.1 Conclusion

We have discussed the crowd management framework for spreading

visitors over a crowded area. Sensors layer aims to track visitors and
gather information about visitor’s crowd. Management layer aims to
analyze the collected data and extract the required information about the
visitors upon administrative acceptance. The interface layer provides an
application that aims to help administrators and visitors to avoid crowd
disasters as it informs them about the count of people in a particular area
and overcrowd area. Also, the proposed framework shall be usable
irrespective of the age and gender if the person knows how to operate a
computer in case of open space proposed model or a one know how to
switch on or off electric circuit can operate our proposed model in closed
space. In wide area, we implement computer vision using python to
measure the distance between each individual and treat them as cause of
crowd violation if they didn’t follow the distance limit among each person
in the [Link] proposed framework will effectively save time and
efforts to help administrators controlling and distributing visitors.

7.2 Future Enhancements

We have discussed two aspects of crowd monitoring in this report.

The most notable point is the exceptional research on vision based crowd
monitoring [Link] order to address the privacy laws, focus needs to
be shifted to less-intrusive and privacy preserving crowd monitoring
techniques. Both, vision and non-vision techniques can compromise the
privacy of an individual, either by facial recognition. While security
oriented technologies might have permissions to identify and track
individuals, general crowd monitoring techniques need to focus more on
intrusion and [Link] monitoring is an important research area and
in order to develop effective technologies, it is necessary to test the newly
proposed techniques in different crowd scenarios. However, the possibility
of getting real life scenarios for testing is either limited or biased with
assumptions. Emulation software like Autodesk, can be really useful to
evaluate the performance of developed techniques under such
[Link] part of our future work, we will deploy Crowd Track system
for large scale environments, such as those occurring in disaster
management scenarios. Also, we plan to perform user studies with safety
and security personnel, to evaluate the applicability of the proposed
measures in practice.

SOURCE CODE

9.1 Source Code

from [Link] distance as dist

import numpyas np
import imutils
import cv2

# import the necessary packages

# initialize minimum probability to filter weak detections along with

# the threshold when applying non-maxima suppression
MIN_CONF = 0.3
NMS_THRESH = 0.3

# define the minimum safe distance (in pixels) that two people can be
# from each other
MIN_DISTANCE = 40

# Detect people function

def detect_people(frame, net, ln, personIdx=0):
# grab the dimensions of the frame and initialize the list of
# results
(H, W) = [Link][:2]
results = []
# construct a blob from the input frame and then perform a forward
# pass of the YOLO object detector, giving us our bounding boxes
# and associated probabilities
blob = [Link](frame, 1 / 255.0, (416, 416), swapRB=True,
crop=False)
[Link](blob)
layerOutputs = [Link](ln)

# initialize our lists of detected bounding boxes, centroids, and

# confidences, respectively
boxes = []
centroids = []
confidences = []

# loop over each of the layer outputs

for output in layerOutputs:
# loop over each of the detections
for detection in output:
# extract the class ID and confidence (i.e., probability)
# of the current object detection
scores = detection[5:]
classID = [Link](scores)
confidence = scores[classID]

# filter detections by (1) ensuring that the object

# detected was a person and (2) that the minimum
# confidence is met
if classID == personIdxand confidence > MIN_CONF:
# scale the bounding box coordinates back relative to
# the size of the image, keeping in mind that YOLO
# actually returns the center (x, y)-coordinates of
# the bounding box followed by the boxes' width and
# height
box = detection[0:4] * [Link]([W, H, W, H])
(centerX, centerY, width, height) = [Link]("int")

# use the center (x, y)-coordinates to derive the top

# and and left corner of the bounding box
x = int(centerX - (width / 2))
y = int(centerY - (height / 2))

# update our list of bounding box coordinates,

# centroids, and confidences
[Link]([x, y, int(width), int(height)])
[Link]((centerX, centerY))
[Link](float(confidence))

# apply non-maxima suppression to suppress weak, overlapping

# bounding boxes
idxs = [Link](boxes, confidences, MIN_CONF, NMS_THRESH)

# ensure at least one detection exists

if len(idxs) >0:
# loop over the indexes we are keeping
for iin [Link]():
# extract the bounding box coordinates
(x, y) = (boxes[i][0], boxes[i][1])
(w, h) = (boxes[i][2], boxes[i][3])

# update our results list to consist of the person

# prediction probability, bounding box coordinates,
# and the centroid
r = (confidences[i], (x, y, x + w, y + h), centroids[i])
[Link](r)

# return the list of results

return results

# load the COCO class labels our YOLO model was trained on
labelsPath = "D:\\final yr
project\\Single-Multiple-Custom-Object-Detection-and-Tracking\data\\labels\\coco.n
ames"
LABELS = open(labelsPath).read().strip().split("\n")

# derive the paths to the YOLO weights and model configuration

weightsPath = "D:\\final yr
project\\Single-Multiple-Custom-Object-Detection-and-Tracking\\yolov3_tf2\\yolov3.
weights"
configPath = "D:\\final yr
project\\Single-Multiple-Custom-Object-Detection-and-Tracking\\[Link]"

# load our YOLO object detector trained on COCO dataset (80 classes)
print("[INFO] loading YOLO from disk...")
net = [Link](configPath, weightsPath)

# determine only the output layer names that we need from YOLO
ln = [Link]()
ln = [ln[i - 1] for iin [Link]()]

# initialize the video stream and pointer to output video file

print("[INFO] accessing video stream...")
# vs = [Link](args["input"] if args["input"] else 0)
vs = [Link]("crowd_monitoring_sample.mp4")

# loop over the frames from the video stream

while True:
# read the next frame from the file
(grabbed, frame) = [Link]()

# if the frame was not grabbed, then we have reached the end
# of the stream
if not grabbed:
break

# resize the frame and then detect people (and only people) in it
frame = [Link](frame, width=900)
results = detect_people(frame, net, ln, personIdx=[Link]("person"))

# initialize the set of indexes that violate the minimum social

# distance
violate = set()

# ensure there are at least two people detections (required in

# order to compute our pairwise distance maps)
if len(results) >= 2:
# extract all centroids from the results and compute the
# Euclidean distances between all pairs of the centroids
centroids = [Link]([r[2] for r in results])
D = [Link](centroids, centroids, metric="euclidean")

# loop over the upper triangular of the distance matrix

for iin range(0, [Link][0]):
for j in range(i + 1, [Link][1]):
# check to see if the distance between any two
# centroid pairs is less than the configured number
# of pixels
if D[i, j] < MIN_DISTANCE:
# update our violation set with the indexes of
# the centroid pairs
[Link](i)
[Link](j)

# loop over the results

for (i, (prob, bbox, centroid)) in enumerate(results):
# extract the bounding box and centroid coordinates, then
# initialize the color of the annotation
(startX, startY, endX, endY) = bbox
(cX, cY) = centroid
color = (0, 255, 0)

# if the index pair exists within the violation set, then

# update the color
if iin violate:
color = (0, 0, 255)

# draw (1) a bounding box around the person and (2) the
# centroid coordinates of the person,
[Link](frame, (startX, startY), (endX, endY), color, 2)
[Link](frame, (cX, cY), 5, color, 1)

# draw the total number of social distancing violations on the

# output frame
text = "Crowded areas identified till now: {}".format(int(len(violate) / 2))
[Link](frame, text, (10, [Link][0] - 25), cv2.FONT_HERSHEY_SIMPLEX,
0.85, (0, 0, 255), 3)

# output frame should be displayed to our screen

[Link]("Frame", frame)
key = [Link](1) &0xFF
# if the `q` key is pressed, break from the loop
if key == ord("q"):
break

References

[1] S. Hashish, and M. Ahmed (2015). "Efficient wireless sensor network rings overlay for crowd
management in Arafat area of Makkah." In 2015 IEEE International Conference on Signal
Processing, Informatics, Communication and Energy Systems, pp. 1-6.
[2] D. C. Duives, T. van Oijen, and S. P. Hoogendoorn (2020), “Enhancing crowd monitoring
system functionality through data fusion: Estimating flow rate from Wi-Fi traces and automated
counting system data,” IEEE Sensors, vol. 20, no. 21, p. 6032.
[3] Sebastian Bek and Eduardo Monari (2015), “The Crowd Congestion Level A New Measure For
Risk Assessment In Video Based Crowd Monitoring”, International Journal of Information
Technology and Decision Making 3(4):2353- 2361.
[4] Rosario Fedele and Massimo Merenda (2017), “An IoT System for Social Distancing and
Emergency Management in Smart Cities Using Multi-Sensor Data”, In 2017 IEEE International
Conference on Signal Processing, Informatics, Communication and Energy Systems.
[5] Marwa F. Mohamed, Abd El Rahman Shabayek, Mahmoud ElGayyar (2018), “IoT-based
Framework for Crowd Management”, Computer Science Department, Faculty of Computes and
Informatics Suez Canal University, Ismailia, Egypt, IEEE International Conference on IoT. [15]
Jianjie Yang, Jin Li, Ye He,“Crowd Density and Counting Estimation Based on Image Textural
Feature”, JOURNAL OF MULTIMEDIA, VOL. 9, NO. 10, OCTOBER 2014.
[16] Aditya vora, Vinay chilaka,“fchd: fast and accurate head detection in crowded scenes”,
arxiv:1809.08766v3 [[Link]] 5 may 2019
[17] Sheng-Fuu Lin, Member, IEEE, Jaw-Yeh Chen, and Hung-Xin Chao, “Estimation of
Number of People in Crowded Scenes Using Perspective Transformation” ieee transactions on
systems, man, and cybernetics—part a: systems and humans, vol. 31, no. 6, november 2001.
[18] Mikel Rodriguez, Ivan Laptev, Josef Sivic, Jean-Yves Audibert, “Density-aware person
detection and tracking in crowds. Proceedings of the IEEE International Conference on Computer
Vision (2011)”.

------------------------------------------------------------------------------------------