A PROJECT REPORT ON

CAR ACCIDENT DETECTION USING DEEP

LEARNING

SUBMITTED TO THE SAVITRIBAI PHULE PUNE UNIVERSITY, PUNE

IN THE PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE AWARD OF THE DEGREE

OF

BACHELOR OF ENGINEERING (COMPUTER ENGINEERING)

SUBMITTED BY

Divya Virshetty Badure Seat No: B190484207

Under The Guidance of

Prof. A. R. Patil

DEPARTMENT OF COMPUTER ENGINEERING

TSSM’s
Padambhooshan Vasantdada Patil Institute of Technology
Bavdhan -21
SAVITRIBAI PHULE PUNE UNIVERSITY
2022 -2023
 CERTIFICATE

This is to certify that the project report entitles

CAR ACCIDENT DETECTION USING DEEP

LEARNING

Submitted By

Divya Virshetty Badure Seat No: B190484207

Is a bonafide work carried out by us under the supervision of Prof. A. R.

Patil and it is approved for the partial fulfillment of the requirement of Savitribai
Phule Pune University, for the award of the degree of Bachelor of Engineering
(Computer Engineering).

(Prof. A. R. Patil) (Prof. .......................)

Internal Guide External Examiner

(Prof. Ganesh Wayal) (Prof. R. S. Pawar)

Head Of Department Principal,PVPIT

Stamp Of The College

Place: PVPIT Bavdhan Pune
Date: ..../..../2023
 Acknowledgement

It gives us great pleasure and satisfaction in presenting this project report on

“Car Accident Detection Using Deep Learning”.

We thankful to and fortunate enough to get constant encouragement, support

and guidance from all Teaching staffs of Computer Department which helped us
in successfully completing our project work. Also, We would like to extend our
sincere esteems to all staff in laboratory for their timely support.

We have furthermore to thank Computer Department HOD Prof. Ganesh

Wayal and Guide Prof. A. R. Patil to encourage us to go ahead and for con-
tinuous guidance. We would also like to thank our project team members who
showed immense patience and understanding throughout the project.

We would like to thank all those, who have directly or indirectly helped us for
the completion of the work during this project.

Divya Virshetty Badure

(B.E. Computer Engg.)
 Abstract

At present, vehicular collisions constitute the primary contributor to loss of life.

The rise in vehicular fatalities has been observed to correspond with the growth
in automobile manufacturing. Metropolitan areas also exhibit a heightened mor-
tality rate due to a growing trend of negligence among individuals, resulting in
severe consequences. The wide variety of transportation options available in con-
temporary times is believed to contribute significantly to the increased incidence
of automobile accidents. The observed trend indicates a rise in both vehicular
and passenger traffic with the growing number of automobiles utilizing the road
networks. They present a potential hazard to both physiological and psychological
systems. Thus, there exists a need for a dependable and feasible approach to auto-
matic accident detection that leverages methodologies from the domain of image
processing. In order to achieve this objective, the employed methodology involves
the utilization of image normalization techniques, Convolutional Neural Networks
(CNNs), and a Decision Tree algorithm for the purpose of detecting instances of
cars accidents.
Keywords: Convolutional Neural Networks, car accident detection, Image pro-
cessing, Image normalization.

PVPIT, Department of Computer Engineering, 2022-23 I

 Table of Contents

List of Figures IV
List of Tables V

Sr Title Page
No. No.

1 INTRODUCTION 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 LITERATURE SURVEY 5

3 SOFTWARE REQUIREMENTS AND SPECIFICATION 9

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.1 Project Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.1.2 User Classes and Characteristics . . . . . . . . . . . . . . . . 10
3.1.3 Assumption and Dependencies . . . . . . . . . . . . . . . . . 10
3.2 Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 External Interface Requirements . . . . . . . . . . . . . . . . . . . . 11
3.3.1 User Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.2 Hardware Interfaces . . . . . . . . . . . . . . . . . . . . . . . 11
3.3.3 Communication Interfaces . . . . . . . . . . . . . . . . . . . 11
3.4 Nonfunctional Requirements . . . . . . . . . . . . . . . . . . . . . . 12
3.4.1 Performance Requirements . . . . . . . . . . . . . . . . . . . 12
3.4.2 Safety Requirements . . . . . . . . . . . . . . . . . . . . . . 12
3.4.3 Security Requirements . . . . . . . . . . . . . . . . . . . . . 12
3.4.4 Software Quality Attributes . . . . . . . . . . . . . . . . . . 12

PVPIT, Department of Computer Engineering, 2022-23 II

 3.5 System Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.5.1 Software Requirements(Platform Choice) . . . . . . . . . . . 12
3.5.2 Hardware Requirements . . . . . . . . . . . . . . . . . . . . 13
3.6 SDLC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.7 System Implementation Plan . . . . . . . . . . . . . . . . . . . . . . 14

4 SYSTEM DESIGN 16
4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Data Flow Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3.1 DFD Level 0 . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3.2 DFD Level 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3.3 DFD Level 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.4 UML Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.4.1 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . 19
4.4.2 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . 19
4.4.3 Sequence Diagram . . . . . . . . . . . . . . . . . . . . . . . 20
4.4.4 Component Diagram . . . . . . . . . . . . . . . . . . . . . . 20
4.4.5 Deployment Diagram . . . . . . . . . . . . . . . . . . . . . . 21
4.4.6 State Transition Diagram . . . . . . . . . . . . . . . . . . . 21
4.4.7 Package Diagram . . . . . . . . . . . . . . . . . . . . . . . . 22

5 OTHER SPECIFICATION 23
5.1 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.3 Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 PROJECT IMPLEMENTATION 24
6.1 Implementation Detail . . . . . . . . . . . . . . . . . . . . . . . . . 24
6.2 Tools and Technologies Used . . . . . . . . . . . . . . . . . . . . . . 27

7 TESTING 28
7.1 Type of Testing Used . . . . . . . . . . . . . . . . . . . . . . . . . . 28

PVPIT, Department of Computer Engineering, 2022-23 III

 7.1.1 Performance Testing . . . . . . . . . . . . . . . . . . . . . . 28
7.1.2 System Testing . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.1.3 Recovery Testing . . . . . . . . . . . . . . . . . . . . . . . . 28
7.1.4 Security Testing . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.1.5 Integration Testing . . . . . . . . . . . . . . . . . . . . . . . 29
7.2 Test Cases and Test Results . . . . . . . . . . . . . . . . . . . . . . 30

8 RESULT 31
8.1 Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8.2 Screen Shots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

9 CONCLUSIONS AND FUTURE WORKS 38

9.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
9.2 Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

Appendix A REFERENCES 39

Appendix B PLAG REPORT 41

Appendix C CERTIFICATES 42

PVPIT, Department of Computer Engineering, 2022-23 IV

 List of Figures

3.1 Water fall model Architecture . . . . . . . . . . . . . . . . . . . . . 13

3.2 Timeline Chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4.2 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 DFD Level 0 Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.4 DFD Level 1 Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.5 DFD Level 2 Diagram . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.6 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.7 Activity Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.8 Sequence Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.9 Component Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.10 Deployment Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.11 State Transition Diagram . . . . . . . . . . . . . . . . . . . . . . . 21
4.12 Package Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6.1 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6.2 Convolution Neural Network Architecture . . . . . . . . . . . . . . 26

8.1 Mean Square Error Measurement . . . . . . . . . . . . . . . . . . . 32

8.2 No of proper expected of scores V/s No of Obtained Scores . . . . . 32
8.3 Model Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
8.4 Main Frame to input Videos . . . . . . . . . . . . . . . . . . . . . . 34
8.5 Frames Extracted . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
8.6 Gray scale Converted Frames . . . . . . . . . . . . . . . . . . . . . 35
8.7 1st Epoch of CNN Training . . . . . . . . . . . . . . . . . . . . . . 35
8.8 Last Epoch of CNN Training . . . . . . . . . . . . . . . . . . . . . . 36
8.9 Accuracy and Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

PVPIT, Department of Computer Engineering, 2022-23 V

8.10 Car Accident Detection Frame . . . . . . . . . . . . . . . . . . . . . 37

PVPIT, Department of Computer Engineering, 2022-23 VI

 List of Tables

3.1 Hardware Requirements . . . . . . . . . . . . . . . . . . . . . . . . 13

7.1 Test Cases For User . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

PVPIT, Department of Computer Engineering, 2022-23 VII

 Car Accident Detection Using Deep Learning

CHAPTER 1

INTRODUCTION

1.1 Introduction

The incidence of traffic accidents and deaths has increased along with increasing
consumer demand for and broad usage of automobiles. Nowadays, there are more
fatalities on the roads than ever. Injuries experienced while driving can end up
in enormous financial consequences for people. their homes, and whole nations.
Tragically, a lot of individuals have perished as a result of delayed assistance. Even
a seemingly small accident might be catastrophic if first aid is not administered
right after. In a critical situation, it’s probable that the car’s driver or passengers
won’t be capable to reach out for help.
At any expense, accidental deaths must be avoided. While accidents cannot
be prevented, researchers can ensure that those who are hurt swiftly get medical
attention. In a case like this, each moment matters for rescue staff. There are
several deaths as a result of the ambulance’s delayed arrival. The lapse in time is
mostly caused by the possibility that no one saw what occurred.
Regarding causal factors, automobile collisions are responsible for an overabun-
dance of fatalities and serious injuries. Insufficient emergency care is responsible
for a higher percentage of fatalities compared to those resulting from vehicular
accidents. Although car accidents are typically unforeseen events, they can be
averted through timely and dependable emergency medical attention. Possible
factors that may lead to accidents include mechanical malfunction, unfavourable
highway conditions, and delayed driver response. In the absence of immediate
medical intervention following an injury-inducing event, there is a high probabil-
ity of fatality for the victim. The lengthening of ambulance turnaround times
can be attributed primarily to delays in the notification of emergency services. In
situations where the unfortunate individual is unable to function, their ability to

PVPIT, Department of Computer Engineering, 2022-23 1

 Car Accident Detection Using Deep Learning

initiate an emergency call may be compromised.

The general people have to pay greater attention to and become more aware
of this problem. Individuals who must walk around the city lacking being able to
use cars run a similar danger. The situation is becoming harsher by the minute
and by the hour.
Drive-by traffic has an impact on a variety of goods and hobbies and has grown
to be an integral component of people’s everyday life. Every year a significant
number of people lose their lives in motor accidents, and an even greater number
get injuries or become incapacitated. It has been said that the frequency of traffic
collisions and the annual production of cars have both grown at an exponential
pace, increasing the number of deaths and property damage. Numerous lives
have been taken as a result of missed chances for reporting accidents and the
resulting absence of medical assistance, despite continued attempts to expand
CCTV surveillance via the placement of cameras at traffic junctions and high-
tech radar equipment for detecting speeding automobiles on key roads.
Because car accidents are becoming increasingly frequent, it is crucial yet chal-
lenging to identify accidents using computer vision. The use of artificial intelli-
gence (AI) for roadway surveillance and vehicle accident detection has generated
a lot of enthusiasm and anticipation. As an example, detection of accidents may
be used to alert the autonomous vehicle to slow down or stop if a collision is about
to occur. Additionally, an autonomous vehicle may alert authorities right away
if it identifies a collision that occurred previously. This speeds up the process of
alerting emergency personnel and dealing with the ramifications of accidents.
The one-of-a-kind strategy that Kumaravelrajah Kapilan [1] suggested as a
solution to the issue of insufficient emergency services for those who have been
injured in a car accident is the most realistic and efficient way to deal with the
issue. This system is designed to promptly alert the pre-set contact information
for emergencies in the case of an accident, and it would also supply the contact
with information on the geolocation of the incident. In the case of an accident,
timely action may be taken as a result of this notion. It will hasten the response
of the rescue team to the site of the collision, which could mean the distinction

PVPIT, Department of Computer Engineering, 2022-23 2

 Car Accident Detection Using Deep Learning

among life and death for those who are injured. With this strategy, the time it
takes to get medical assistance after an accident is cut down significantly. The
most significant shortcoming of the system is that it is unable to function in areas
that have dense extensions or even inadequate network connectivity at all.
Liang-Bi Chen [2] suggested employing an in-vehicle amusement informatics
framework as a heterogeneous bridging interface for linking multiple communica-
tion modalities in his concept for an IoV system. This idea was presented in the
context of an IoV system. For the purpose of predicting the occurrence of traf-
fic collision incidents, the TensorFlow framework has implemented deep learning
algorithms on a cloud-based autonomous neural training infrastructure. It has
become abundantly evident, on the basis of actual findings, that deep learning
has the potential to make major advancements in the effectiveness of identifying
road accident incidents. If it is determined that a traffic collision has occurred,
the pertinent information will be uploaded to a cloud-based technology platform
in order to be used in an instant warning as well as additional operations.
It is possible for the safety of both the driver and passengers in a vehicle to
be improved by using the technology proposed by Navin Kumar M [3]. Because
the technology provides everything on its own, there is no requirement for any
kind of involvement from a person. It is possible that putting this method into
action can assist guarantee that the individual gets prompt medical treatment,
which will increase their probability of survivability. The equipment used has the
capacity to determine the precise location of an accident and the circumstances
of the incident, which enables it to compute the number of automobiles that were
involved and, as a result, the number of paramedics that will be necessary to
promptly transfer all of the injured people.

1.2 Motivation

ˆ Increasing safety and security for the car accident scenario

ˆ Enabling effective utilization of image processing to achieve the car accident

detection.

PVPIT, Department of Computer Engineering, 2022-23 3

 Car Accident Detection Using Deep Learning

ˆ Implementing Convolutional Neural Networks for the purpose of achieving

the car accident detection.

1.3 Problem Definition

To enhance the process of car accident detection using image normalization

along with Convolutional Neural Networks and Decision Making.

PVPIT, Department of Computer Engineering, 2022-23 4

 Car Accident Detection Using Deep Learning

CHAPTER 2

LITERATURE SURVEY

In the words of Sonjoy Rana [4,] the information that is being provided here is
intended to make the emergency response unit (ERU) more efficient and speedy
so that it can better assist patients in need. This prototype is able to conduct
an investigation into the events leading up to an accident and promptly alert the
relevant parties through mobile app or text message. A GPS module also relays
the current location of the incident to the responding personnel. The system’s
two smartphone applications include a Driver characteristic for synchronising the
individual’s telephone with the QARS record keeping system, a Pedestrian variant
for providing photos and videos from the location of a collision, in addition to a
Disaster Services connected that gets all notifications, declarations, and data anal-
ysis on incidents. The platform also includes a The sidewalk variant for providing
images and footage from the context of a car crash. Due to the fact that this
is the case of emergency, the Quick Accident Response System (QARS) could be
able to assist in reducing the number of fatalities that are related with vehicular
accidents.
Using a ResNet with a customised this was before the-activation, a spatio-
temporal 3D kernel programme, and an architecture known as a fully-hypothesized
postoperative network (FPN), Hyunwoo Kim [5] proposed a method for identifying
automobile wrecks in CCTV video. This method makes use of a fully-hypothesized
postoperative network (FPN). It is possible that enabling parallel processing would
improve the performance of processing since the proposed architecture is based on
convolutional neural networks. The periodic input helps to lower the overall com-
puting overhead, while the most recent pre-activation and the FPN both contribute
to an increase in accuracy. This tactic might be used in a CCTV administration
system to improve the effectiveness of the system by enabling every leadership
agent to monitor a greater number of cameras at the same time.

PVPIT, Department of Computer Engineering, 2022-23 5

 Car Accident Detection Using Deep Learning

According to Rajvardhan Rishi’s [6] research, the AMS system will make sig-
nificant advances to the field of automobile accidents. The automatic messaging
system makes it possible to give assistance to the person who survived a vehicle
accident as quickly as possible. Because of this, the victim has a better chance of
surviving the attack. Incorporating AMS into the production process of vehicle
gearboxes might potentially improve their overall utility. In the vast majority of
contemporary autos, there is no provision for installing an AMS system. Through
the application of mobile phone technology, an automobile manufacturer or en-
forcement service may immediately react to incidents, hence lowering the overall
number of fatalities and the severity of injuries.
As Aparajith Srinivasan [7] explains, reported work serves as part of a mishap
recognition architecture; this encompasses daytime, evening, misty, snowy, and
dusty conditions. It has been shown via empirical research that the brand-new,
state-of-the-art Recognition Transformer significantly improves the precision with
which incidents may be spotted in surveillance film. The Random Forest method
will be used as a supplementary aid for the classification procedure when it was
decided to do so. We were able to achieve a low latency while maintaining an
excellent rate of detection, which was an improvement over our earlier attempts
by a substantial margin. One of the challenges that the DETR has at the moment
is the reality that it has difficulty locating very minute details.
The technique that was presented by Gokul Rajesh [8] is employed to determine
the causes of vehicle accidents. The Global Positioning System (GSM) module will
transmit an alarm to the several command centres in the area if an event is found
to have occurred. But in contrast to the systems that are in place right now,
this particular one is more trustworthy and economical. Even while the system
design is being trained to utilise the created information, it is already capable of
identifying accidents with anything like an exceptional level of accuracy. An early
examination substantiates the hypothesis that the system operates faultlessly and
can be scaled. If this system is implemented, the appropriate authorities will
be alerted instantly, and the time it takes for first responders to get at the site
of an accident will be cut down significantly. Because of this, the solution that

PVPIT, Department of Computer Engineering, 2022-23 6

 Car Accident Detection Using Deep Learning

was presented will be valuable in a future where accidents involving vehicles are
becoming an increasingly important issue.
As suggested by Prasanmit Nath [9], this could possibly be deployed as a
component of a well-connected medical evacuation assistance, with ambulances
being sent to the sites of incidents in which they are required the most. When
determining what kinds of supplies a paramedic needs to bring to the site of a
vehicle collision, it is helpful to keep in mind the severity of the crash in order
to make the most informed decision possible. It is necessary for the gadget to
be permanently mounted inside the vehicle if it is to be utilised in a setting that
simulates real-world conditions. It is necessary for it to have a robust power supply
that is self-contained in order for it to be able to survive a significant accident and
continue to provide data to the EMS.
The new approach developed by Dhananjai Chand [10] includes a recommenda-
tion to identify vehicle impacts. The accumulation of data from relocating targets
is an essential component of the model. In addition, the video feed from a dash
cam could be utilised in activities like as calculating velocity, recognising lanes,
and detecting convergence of velocity vectors. Because of the myriad of param-
eters that we have included into the framework in order to assess the likelihood
of accidents, the structure is more reliable than it has ever been. The activities
of these individuals contribute to the formation of the foundation for autonomous
cars throughout Asia. However, we do not yet have access to massive datasets
that have been well annotated so that we may test and improve our system. In
addition to this, the limitation of the outcome is the capacity of prediction that
such a Mask R-CNN architecture has.
Durgesh Kumar Yadav has reviewed a number of different research that investi-
gate different strategies for locating accidents [11]. This initiative aims to enhance
the underlying physical network rather than the ability of Intelligent Transporta-
tion Systems (ITS) to recognise congestion issues, accidents, and other events as
they occur. Several scholars and industry professionals have come up with a va-
riety of ideas for the implementation of autonomous accident surveillance. They
detect precisely where there was a crash of crashes in traffic by using means such

PVPIT, Department of Computer Engineering, 2022-23 7

 Car Accident Detection Using Deep Learning

as smartphones, VANET, Global navigation and GSM components, mobile ap-

plications, and deep learning algorithms. This information is then sent to first
responders.
Yasitha Warahena Liyanage [12] proposed a one-of-a-kind Bayesian method for
identifying the most dangerous activities based on speed data in the context of
determining the location of incidents on motorways. The researchers discovered
the ideal technique and built a technique (ATTAIN) for applying it into practice
in the case that important pre- and post-difference calculations are provided. In
addition, the researchers devised recursive formulations using ML and MAP in
order to estimate the missing crucial post-accident model parameters. The re-
searchers relied on the formulations described earlier and the ideal resolution in
order to produce ATTAIN-ML and ATTAIN-MAP, which are two modifications
of the original ATTAIN programmes. The results of the evaluation showed that
ATTAIN and its variants improve not only the current state of the art for optimum
cutoff recognition in the analysis of time series, but also the accuracy of beginning
time estimate for the features used in this research.

PVPIT, Department of Computer Engineering, 2022-23 8

 Car Accident Detection Using Deep Learning

CHAPTER 3

SOFTWARE REQUIREMENTS AND SPECIFICATION

3.1 Introduction

This Software Requirements Specification provides a complete description of

all the functions and constraints of the “Car Accident Detection Using Deep
Learning”. The document describes the issues related to the system and what
actions are to be performed by the development team in order to come up with a
better solution.
In terms of contributing causes, automobile collisions account for a dispropor-
tionate number of fatalities and injuries. A greater proportion of persons die owing
to inadequate emergency care than due to traffic accidents. While car accidents
are usually a complete surprise, they may be prevented with prompt and reliable
emergency care. Mechanical failure, poor road conditions, and slow driver reac-
tion times are all potential causes of accidents. Without prompt medical attention
after an injury-causing incident, the victim is quite likely to die. Due mostly to
delays in alerting emergency services, ambulance response times may be signifi-
cantly lengthened. When the victim is incapacitated, they may not be able to
make an emergency call. Consequently, there is a want for a reliable and practical
method of auto-accident detection that makes use of techniques from the field of
image processing. To this end, we use image normalization, Convolutional Neural
Networks, and a Decision Tree to identify the occurrence of vehicle accidents.

3.1.1 Project Scope

The purpose of this SRS document is to provide a detailed overview of our

software product “Car Accident Detection Using Deep Learning”, its pa-
rameters and goals. This document describes the project’s target audience and
its user interface, hardware and software requirements. It defines how our client,

PVPIT, Department of Computer Engineering, 2022-23 9

 Car Accident Detection Using Deep Learning

team and audience see the product and its functionality. The Main Purpose of
System is to deploy the Voice based email system for visually impaired users to
provide voice based navigation and improved convenience.
The setup should be precise for the capturing of the car input video for the car
accident detection scenario. The constraints felt thus far by the developer have
only been our weekly story cards, the end-to-end side of the interface, and time to
time brushing on methodology of implementation which schedule the completion
of the project in May 2023.
The major scope of this project is as follows

ˆ The training is performed for a limited set of input car accident dataset.

ˆ The application should be easy to use operate and deploy.

ˆ The data properly preprocessed to achieve effective car accident detection.

3.1.2 User Classes and Characteristics

This project is for the average user to enhance the experience while utilizing
a car accident detection system. This is having following characteristics:
End User

ˆ Activate System.

ˆ Initiates the Car accident recognition.

3.1.3 Assumption and Dependencies

Following are the assumptions:

ˆ System is deployed using a single machine.

ˆ Limited number of videos are given input.

Dependencies can be:

ˆ System Speed.

ˆ IDE Configuration.

PVPIT, Department of Computer Engineering, 2022-23 10

 Car Accident Detection Using Deep Learning

3.2 Functional Requirements

ˆ System should grab car accident frames accurately.

ˆ System should yield the results as per user Defined protocols.

ˆ System should display all intermediate results in sober format.

ˆ System should send proper detection of car accident.

3.3 External Interface Requirements

3.3.1 User Interfaces

Our system interacts with user on the following occasions:

1. By the User while activating the system.

2. By the User while providing the car accident video input.

3.3.2 Hardware Interfaces

Our system interacts with hardware components on the following occasions:

ˆ Our system interacts with storage for car accident input video.

ˆ Our system interacts with user for displaying the result.

3.3.3 Communication Interfaces

Our system’s different modules are communicating with one another on the fol-
lowing scenarios:

1. From Car Accident input video Module to Preprocessing module

2. From Preprocessing Module to Image Normalization module

3. From Image Normalization Module to Convolutional Neural Network module

4. From Convolutional Neural Network module to Decision Making module

5. From Decision Making module to Car Accident Detection module

PVPIT, Department of Computer Engineering, 2022-23 11

 Car Accident Detection Using Deep Learning

3.4 Nonfunctional Requirements

3.4.1 Performance Requirements

ˆ High Speed: System should process car accident detection in good speed.
Then system must wait for process completion.

ˆ Accuracy: System should correctly detect the car accident, and display the
result accurately. System output should be in user required format. That
means all the intermediate steps need to be display properly.

3.4.2 Safety Requirements

The system need to be properly converted into executable file format so that
the source code of the system will preserved from any kind of intrusions.

3.4.3 Security Requirements

The system need to be secured using the proper credentials and always the
backup of the system need to be kept to restore the whole system again in worst
case scenario.

3.4.4 Software Quality Attributes

ˆ Number of functionalities provided

ˆ System should be reliable
ˆ User login is secured.
ˆ System provides proper way of car accident detection
ˆ Survivability of the system should be good
ˆ System performs each function accurately.
ˆ System is extensible.

3.5 System Requirements

3.5.1 Software Requirements(Platform Choice)

1. Platform: Python
2. Technology : Python SDK
3. IDE: Spyder 5.0

PVPIT, Department of Computer Engineering, 2022-23 12

 Car Accident Detection Using Deep Learning

3.5.2 Hardware Requirements

1 systems of following minimum configuration

Sr. No. Parameter Minimum Requirement Justification

1 Processor 2.2 GHz For Fast Processing
2 Hard Disk 200 GB For Fast Processing
3 RAM 4 GB For Fast Processing
4 Monitor, Keyboard and UPS 1 Quantity None

Table 3.1: Hardware Requirements

3.6 SDLC

The Project for Car Accident Detection using Deep Learning is developed
under waterfall model architecture as shown in the below figure 3.1.

Figure 3.1: Water fall model Architecture

According to this model the planning about the project is done followed by the
requirement analysis. Once the requirement analysis is fulfilled the design process
was started followed by coding process. After the coding process implementation

PVPIT, Department of Computer Engineering, 2022-23 13

 Car Accident Detection Using Deep Learning

process is complemented. This implementation is verified and validated to con-

tinue with the installation process to deploy the design model. As a last step the
operation and support is provided to incorporate the designed software.

3.7 System Implementation Plan

Figure 3.2: Timeline Chart

ˆ Literature Review On Linear Solver Case : This time, our team performs
the literature review.

ˆ Abstract & Synopsis Submission : This time, our team wrote the abstract
and synopsis.

ˆ Scheme of Implementation : This time, our team developed the basic mod-
ule.

ˆ Performance profiling to identify bottlenecks : This time, our team identified

the bottleneck for performance profiling.

ˆ Study Of Behaviour of solver : Here the behaviour of the user according to

the use case.

PVPIT, Department of Computer Engineering, 2022-23 14

 Car Accident Detection Using Deep Learning

ˆ Layout and Design setup : This time, our team designed a user interface.

ˆ Detailed problem definition, SRS , platform : This time, our team wrote the
detailed problem definition in SRS.

ˆ Modify and optimize the bottlenecks : The bottlenecks are identified and
implementation is modified according to the need.

ˆ Detailed design Document and Methodology : This time, our team wrote
the detailed documentation and designed a project methodology.

ˆ Experimental Result : The obtained results are tabulated for different trials.

ˆ Analysis , Validation of result and conclusion : Here our team analyzed and
validated the result and wrote the conclusion.

ˆ Report Document the entire work : Here, our team reports the entire docu-
mentation of the work.

PVPIT, Department of Computer Engineering, 2022-23 15

 Car Accident Detection Using Deep Learning

CHAPTER 4

SYSTEM DESIGN

4.1 System Architecture

Figure 4.1: System Architecture

The system architecture diagram of figure no 4.1 provides the architecture

for the proposed system in the form of different layers. The User Interface layer
consists of the test video and the output is the Car Accident Detection.

4.2 System Design

Figure 4.2: System Design

PVPIT, Department of Computer Engineering, 2022-23 16

 Car Accident Detection Using Deep Learning

The system overview diagram of figure no 4.2 provides an overview of the sys-
tem with the important modules in the form of blocks. At first the fingerprint
dataset is extracted and preprocessed through Image Normalization. These Im-
age Normalization image is then provided to the CNN processing. The Decision
Making is then implemented to achieve the Car Accident Detection.

4.3 Data Flow Diagrams

4.3.1 DFD Level 0

Figure 4.3: DFD Level 0 Diagram

The DFD 0 diagram of figure no. 4.3 for the data flow diagrams describes the
flow of the approach. The DFD diagram provides the simplest flow where in the
Test Video is preprocessed and the CNN is implemented and the Car Accident
Detection is performed.

PVPIT, Department of Computer Engineering, 2022-23 17

 Car Accident Detection Using Deep Learning

4.3.2 DFD Level 1

Figure 4.4: DFD Level 1 Diagram

The DFD 1 diagram of figure no. 4.4 provides even more details wherein
the user provides the test video which is effectively preprocessed. The Image
Normalization is performed and provided to the CNN which are used to perform
CNN 1st layer and training dataset to achieve the Car Accident Detection.

4.3.3 DFD Level 2

Figure 4.5: DFD Level 2 Diagram

PVPIT, Department of Computer Engineering, 2022-23 18

 Car Accident Detection Using Deep Learning

The DFD 2 diagram of figure no. 4.5 is the most detailed wherein the user
provides the test video which is preprocessed through image resizing and then
the Image Normalization is evaluated. These Image Normalization image is then
provided to the CNN that utilizes the Activation function. The Dense layer is
then implemented on the trained module to achieve the Car Accident Detection.

4.4 UML Diagrams

4.4.1 Use Case Diagram

Figure 4.6: Use Case Diagram

The Use case Diagram of figure no. 4.6 depicts the various use cases that are
performed by the user in the proposed model. The use cases feeding the train
dataset, activate system, feeding the test video, and finally view the Car Accident
Detection result.

4.4.2 Activity Diagram

Figure 4.7: Activity Diagram

PVPIT, Department of Computer Engineering, 2022-23 19

 Car Accident Detection Using Deep Learning

The activity diagram of figure no. 4.7 lists the various activities that
are performed in the proposed methodology, the start state is initiated and the
dataset is fed, the system is trained and the test video is fed after which the result
is viewed.

4.4.3 Sequence Diagram

Figure 4.8: Sequence Diagram

The sequence role diagram of figure no. 4.8 provides a sequence of the
approaches as well as the various roles performed in the intermediate.

4.4.4 Component Diagram

Figure 4.9: Component Diagram

PVPIT, Department of Computer Engineering, 2022-23 20

 Car Accident Detection Using Deep Learning

The component diagram of figure no. 4.9 illustrates the important components
in the proposed system. In our approach the important components consist of the
Test Video which is interlinked with Image Normalization and CNN.

4.4.5 Deployment Diagram

Figure 4.10: Deployment Diagram

The deployment diagram of figure no. 4.10 illustrates the important resources
that are utilized for the deployment purposes.

4.4.6 State Transition Diagram

Figure 4.11: State Transition Diagram

PVPIT, Department of Computer Engineering, 2022-23 21

 Car Accident Detection Using Deep Learning

The state transition diagram of figure no. 4.11 provides the various states
that the proposed system goes through. Initially the start state wherein the user
provides the test video which is preprocessed through image resizing and then
the Image Normalization is evaluated. These Image Normalization image is then
provided to the CNN that utilizes the Activation function. The Dense layer is
then implemented on the trained module to achieve the Car Accident Detection
and then reaches the stop state.

4.4.7 Package Diagram

Figure 4.12: Package Diagram

The package diagram of figure no. 4.12 lists all the important modules and
the various packages that are being utilized to achieve the system.

PVPIT, Department of Computer Engineering, 2022-23 22

 Car Accident Detection Using Deep Learning

CHAPTER 5

OTHER SPECIFICATION

5.1 Advantages

1. Can be effective in faster recognition of a car accident scenario autonomously

2. Can be useful in saving lives of the individuals due to timely detection of

collision.

3. This types of software is helpful in providing quick relief to the accident

victims.

4. Can be useful in identifying the road defects where accidents are happening
oftenly.

5.2 Limitations

1. Approach will be demonstrated on a minimal amount of input car accident

dataset.

2. Performance of the approach is dependent on the health of the deployment

machine among other factors.

5.3 Application

1. Traffic Police Offices

2. Emergency Services

3. Car manufacturers

4. Road contractors

PVPIT, Department of Computer Engineering, 2022-23 23

 Car Accident Detection Using Deep Learning

CHAPTER 6

PROJECT IMPLEMENTATION

6.1 Implementation Detail

Figure 6.1: System Overview

The illustration presented in Figure 6.1 illustrates the proposed approach for
detecting car accidents utilizing Convolutional Neural Networks. The following
steps delineate a systematic approach to executing the methodology are:

1. Step 1: Preprocessing

2. Step 2: Image Normalization

3. Step 3: Convolution Neural Network

4. Step 4: Testing through Decision Making

These steps are elaborated below:

Step 1: Preprocessing – The model designed for the identification of car
accidents employs a manually derived dataset specifically tailored for this task.
The process of generating the dataset initially began by procuring videos from a

PVPIT, Department of Computer Engineering, 2022-23 24

 Car Accident Detection Using Deep Learning

widely used video sharing platform, namely YouTube, that were relevant to the
aforementioned shots. The proposed methodology aims to detect vehicular colli-
sions by means of video input. The videos that have been gathered are efficiently
preprocessed, and subsequently, the frames that encompass the car accident are
gathered. The resulting dataset comprises images depicting diverse types of colli-
sions, which are evenly partitioned into training and testing folders for use in the
subsequent stage of the process.
Step 2: Image Normalization - Prior to the training, the images of car
accidents are downsized to a 170 X 170 measurement.
During this stage of the suggested strategy, a generator object for image data
is constructed using the ratio of 1/255 and the software collections Keras and
TensorFlow are employed for comprehensive analysis. The present protocol is
employed for the purpose of commencing both training and testing procedures
on images of cricket shots. The initialization of the ImageDataGenerator object
involves the specification of certain attributes, including the destinations of the
training as well as testing folders, the dimensions of the images, a batch size of 32,
and the categorical class setting, with grayscale as the designated colour mode.
Step 3: Convolutional neural network training - This steps is explained
with the below mentioned point:

ˆ A sequential neural network framework is built using the TensorFlow kit’s

Sequential class. Subsequently, the topmost layer of the deep neural network
is dedicated solely to the equivalent measurements of the images, wherein
a convolution layer comprising 32 kernels measuring 3 by 3 is incorporated,
accompanied by the ”ReLU” activation function. Subsequently, a supple-
mentary Convolutional layer comprising 64 kernels measuring 3 by 3 is incor-
porated, accompanied by the utilization of the rectified linear unit (ReLU)
activation function. A pooling layer of highest size 2 by 2 is implemented
with a dropout rate of 25

ˆ A third convolutional layer is incorporated into the model, with 128 kernels,
each with a size of 3 x 3. The ReLU activation function is employed for this
specific purpose. The maximum pooling layer is specified to have a dimension

PVPIT, Department of Computer Engineering, 2022-23 25

 Car Accident Detection Using Deep Learning

of 2 by 2. Following the deployment of the third layer, a subsequent layer,

namely the fourth layer, is implemented utilizing 128 kernels of dimensions
3 by 3 and employing the Rectified Linear Unit (ReLU) activation function.
A subsequent Max pooling layer is incorporated into the model architecture,
accompanied by a 25% dropout rate and a 2 by 2 dimensionality.

ˆ Ultimately, the structure of the neural network is concluded by applying

the flatten function and incorporating a dense layer of 1024 units, which is
activated by the rectified linear unit (ReLU) function. At the end of the
convolutional neural network, a dropout rate of 50% is specified, utilizing a
pair of dense layers in addition the ”softmax” activation function.

ˆ The Adam optimization algorithm is frequently employed to enhance the

resultant output over a training duration of 500 epochs. Upon completion
of the training process, the trained data is stored in a file format denoted
as .H5. This file is subsequently utilized by the generated model during the
testing phase. Figure 6.2 depicts the framework of the Convolutional Neural
Network.

Figure 6.2: Convolution Neural Network Architecture

PVPIT, Department of Computer Engineering, 2022-23 26

 Car Accident Detection Using Deep Learning

Step 4: Testing through Decision Making- In the course of the test-

ing process, the neural network object for the testing image loads the previously
trained model parameters in h5 file format. The aforementioned data is utilized
to generate integer-based prognostications. The car accident is identified by the
deep learning model trained data as per the classification dictionary. The user
interface presents this output to the user.

6.2 Tools and Technologies Used

The proposed methodology for detecting car accidents is implemented utilizing

the Python programming language on a workstation that operates on the Win-
dows operating system. The Spyder Integrated Development Environment (IDE)
is utilized for the purpose of coding the aforementioned strategy. The machine
utilized for deployment is equipped with an Intel Core i5 central processing unit,
8 gigabytes of random access memory, and a 1 terabyte hard disc drive.

PVPIT, Department of Computer Engineering, 2022-23 27

 Car Accident Detection Using Deep Learning

CHAPTER 7

TESTING

7.1 Type of Testing Used

7.1.1 Performance Testing

The performance of the system Car accident detection is measured by the Root
mean square error for proper accident detection and improper accident detection
mode.

7.1.2 System Testing

Checked the steadiness of the space provided by the system for the maximum
number of images. This is done by feeding the maximum number of images to
learn and so that set the threshold by the Virtual machine option of the spyder
IDE.

7.1.3 Recovery Testing

Our system can be recovered in span of 2 hour after crashing. Within two hour
we can install all the Software and deploy our system to work as in the past.

7.1.4 Security Testing

7.1.4.1 Stress Testing

The System is well equipped to stand against the breakdown point of maximum
number of images to the cloud decided by the IDE’s virtual machine settings,
beyond that the memory overflow exception may arise.

7.1.4.2 Unit Testing

All the modules are independently handled developed and ran to get proper

PVPIT, Department of Computer Engineering, 2022-23 28

 Car Accident Detection Using Deep Learning

output and finally they are integrated to get the whole output.

7.1.4.3 Black Box Testing

Compatibility analysis is done by passing the output of one module to another

and checking for estimated output of Car accident detection System.

7.1.5 Integration Testing

When all the individual modules are integrated into our system, that yields
a proper system of Car accident detection System that is cross examined for the
desired output.

PVPIT, Department of Computer Engineering, 2022-23 29

 Car Accident Detection Using Deep Learning

7.2 Test Cases and Test Results

ID TEST CASE INPUT PASS CRITERIA

U FRAME EXT Frame Extrac- Accident videos All the frames from
tion the input video is ex-
tracted and stored in
the specific folder.
U IMG REZ Resizing of the Frame Folder All the images are re-
frames sized to the said Di-
mension.
U IMG GRAY Gray scaling of Resized image All the images
the Frames folder are converted into
Grayscale mode.
U IMG TRAIN Training the Dataset All the images are suc-
data cessfully trained and
.h5 file is being gener-
ated.
U VID TEST Video testing Video Video is analyzed for
the accident and then
specific alert will be
sent to the user.

Table 7.1: Test Cases For User

PVPIT, Department of Computer Engineering, 2022-23 30

 Car Accident Detection Using Deep Learning

CHAPTER 8

RESULT

8.1 Outcomes

The proposed methodology for detecting car accidents is implemented utilizing

the Python programming language on a workstation that operates on the Win-
dows operating system. The Spyder Integrated Development Environment (IDE)
is utilized for the purpose of coding the aforementioned strategy. The machine
utilized for deployment is equipped with an Intel Core i5 central processing unit,
8 gigabytes of random access memory, and a 1 terabyte hard disc drive.
In order to ensure the successful implementation of the Convolutional Neural
Network, it is imperative to conduct a thorough evaluation of the reliability of the
car accident detection technique. The approach employed in this study utilizes
Root Mean Square Error as a metric to assess the accuracy of the suggested
application.
The utilization of the RMSE performance metric has proven to be effective
in assessing the efficacy of the car accident Detection system. The section below
provides a discussion of the experimental evaluation.
Performance Evaluation through Root Mean Square Error The suggested ap-
proach’s error rate is determined by computing the root mean square error (RMSE).
In this experiment, the Root Mean Square Error (RMSE) is employed to quan-
tify the discrepancy between the observed and predicted car accident detection
outcomes obtained through the use of the Convolutional Neural Network (CNN)
module. Equation 8.1 illustrates the RMSE methodology.
RM SEf 0 = [ N 2 1/2
P
i=1 (Zf i − Zoi ) /N ] ——————(8.1)
Where,
P
- Summation
(Zfi - Zoi)2 - Differences Squared for the hand gesture identified correctly and

PVPIT, Department of Computer Engineering, 2022-23 31

 Car Accident Detection Using Deep Learning

hand gesture identified incorrectly

N - Number of conducted Experiments
Prior to estimating the Error Rate of the Strategy through RMSE, it is neces-
sary to calculate the Mean Square Error (MSE). The mean squared error (MSE)
represents the discrepancy between the observed and predicted values of car acci-
dent detection. The project is undergoing testing with a progressively increasing
number of trials, and the outcomes are being documented in the table presented as
in below table. The obtained outcomes have been utilized to generate the diagram
depicted in Figure 8.2 as presented below.

Figure 8.1: Mean Square Error Measurement

Figure 8.2: No of proper expected of scores V/s No of Obtained Scores

PVPIT, Department of Computer Engineering, 2022-23 32

 Car Accident Detection Using Deep Learning

The mean squared error (MSE) is computed based on the measured MSE value
obtained from thorough experimentation with the convolutional neural network
(CNN) module of the Car accident detection methodology. The root mean square
error (RMSE) of the magnitude 1.264 is derived by taking the square root of the
mean squared error (MSE) average. The effective implementation of the CNN
model can be inferred from the error rate. Consequently, there is a significant
enhancement in the precision of the execution of car accident detection. Figure 8.3
below depicts the model accuracy attained by the Convolutional Neural Network
(CNN) in the system.

Figure 8.3: Model Accuracy

PVPIT, Department of Computer Engineering, 2022-23 33

 Car Accident Detection Using Deep Learning

8.2 Screen Shots

Figure 8.4: Main Frame to input Videos

Figure 8.5: Frames Extracted

PVPIT, Department of Computer Engineering, 2022-23 34

 Car Accident Detection Using Deep Learning

Figure 8.6: Gray scale Converted Frames

Figure 8.7: 1st Epoch of CNN Training

PVPIT, Department of Computer Engineering, 2022-23 35

 Car Accident Detection Using Deep Learning

Figure 8.8: Last Epoch of CNN Training

Figure 8.9: Accuracy and Loss

PVPIT, Department of Computer Engineering, 2022-23 36

 Car Accident Detection Using Deep Learning

Figure 8.10: Car Accident Detection Frame

PVPIT, Department of Computer Engineering, 2022-23 37

 Car Accident Detection Using Deep Learning

CHAPTER 9

CONCLUSIONS AND FUTURE WORKS

9.1 Conclusion

The present study outlines a methodology for automated analysis of car ac-
cident utilising deep learning techniques, which yields a notably improved out-
come. The technique stated employs Convolutional Neural Networks to leverage
the video input of each car accident. The convolutional neural network (CNN)
model is trained using a dataset comprising diverse videos of car accidents. Prior
to being utilised for training, the dataset undergoes preprocessing, followed by
normalisation of the preprocessed images, before being inputted into the CNN
model. Upon completion of the trained model, it is employed to conduct test-
ing on preprocessed and normalised input video, which has not been previously
subjected to CNN recognition. The categorization of car accident identification
outcomes is achieved through a successful process of decision-making. The re-
sults and discussion portions of the research article provide an overview of the
results that were attained through the utilisation of the recommended approach.
The findings have indicated the effectiveness of the suggested methodology with
a significantly low error rate.

9.2 Future Works

The methodology demonstrated in this study has the potential to serve as a

valuable foundation for future research endeavours. Specifically, it can be trans-
formed into an application programming interface (API) to facilitate seamless
integration and enable real-time applications in the future.

PVPIT, Department of Computer Engineering, 2022-23 38

 Car Accident Detection Using Deep Learning

APPENDIX A

REFERENCES

[1] K. Kapilan, S. Bandara and T. Dammalage, ”Vehicle Accident Detection and

Warning System for Sri Lanka Using GNSS Technology,” 2020 International
Conference on Image Processing and Robotics (ICIP), 2020, pp. 1-5, doi:
10.1109/ICIP48927.2020.9367334.

[2] L. -B. Chen et al., ”An Implementation of Deep Learning based IoV System
for Traffic Accident Collisions Detection with an Emergency Alert Mecha-
nism,” 2018 IEEE 8th International Conference on Consumer Electronics -
Berlin (ICCE-Berlin), 2018, pp. 1-2, doi: 10.1109/ICCE-Berlin.2018.8576197.

[3] M. Navin Kumar, S. Pravin Kumar, R. Premkumar and L. Navaneethakr-

ishnan, ”Smart Characterization of Vehicle Impact and Accident Reporting
System,” 2021 7th International Conference on Advanced Computing and
Communication Systems (ICACCS), 2021, pp. 964-968, doi: 10.1109/ICACC
S51430.2021.9441797.

[4] S. Rana, S. Sengupta, S. Jana, R. Dan, M. Sultana and D. Sengupta,

”Prototype Proposal for Quick Accident Detection and Response System,”
2020 Fifth International Conference on Research in Computational Intelli-
gence and Communication Networks (ICRCICN), 2020, pp. 191-195, doi:
10.1109/ICRCICN50933.2020.9296153.

[5] H. Kim, S. Park and J. Paik, ”Pre-Activated 3D CNN and Feature Pyramid
Network for Traffic Accident Detection,” 2020 IEEE International Confer-
ence on Consumer Electronics (ICCE), 2020, pp. 1-3, doi: 10.1109/ICCE4656
8.2020.9043125.

[6] R. Rishi, S. Yede, K. Kunal and N. V. Bansode, ”Automatic Messaging Sys-

tem for Vehicle Tracking and Accident Detection,” 2020 International Con-

PVPIT, Department of Computer Engineering, 2022-23 39

 Car Accident Detection Using Deep Learning

ference on Electronics and Sustainable Communication Systems (ICESC),

2020, pp. 831-834, doi: 10.1109/ICESC48915.2020.9155836.

[7] A. Srinivasan, A. Srikanth, H. Indrajit and V. Narasimhan, ”A Novel Ap-

proach for Road Accident Detection using DETR Algorithm,” 2020 Interna-
tional Conference on Intelligent Data Science Technologies and Applications
(IDSTA), 2020, pp. 75-80, doi: 10.1109/IDSTA50958.2020.9263703.

[8] G. Rajesh, A. R. Benny, A. Harikrishnan, J. Jacob Abraham and N. P. John,

”A Deep Learning based Accident Detection System,” 2020 International
Conference on Communication and Signal Processing (ICCSP), 2020, pp.
1322-1325, doi: 10.1109/ICCSP48568.2020.9182224.

[9] P. Nath and A. Malepati, ”IMU based Accident Detection and Intima-
tion System,” 2018 2nd International Conference on Electronics, Materi-
als Engineering & Nano-Technology (IEMENTech), 2018, pp. 1-4, doi:
10.1109/IEMENTECH.2018.8465309.

[10] D. Chand, S. Gupta and I. Kavati, ”Computer Vision based Accident Detec-
tion for Autonomous Vehicles,” 2020 IEEE 17th India Council International
Conference (INDICON), 2020, pp. 1-6, doi: 10.1109/INDICON49873.2020.93
42226.

[11] D. K. Yadav, Renu, Ankita and I. Anjum, ”Accident Detection Using Deep
Learning,” 2020 2nd International Conference on Advances in Computing,
Communication Control and Networking (ICACCCN), 2020, pp. 232-235,
doi: 10.1109/ICACCCN51052.2020.9362808.

[12] Y. W. Liyanage, D. -S. Zois and C. Chelmis, ”Near Real-Time Freeway Ac-
cident Detection,” in IEEE Transactions on Intelligent Transportation Sys-
tems, vol. 23, no. 2, pp. 1467-1478, Feb. 2022, doi: 10.1109/TITS.2020.3027
494.

PVPIT, Department of Computer Engineering, 2022-23 40

 Car Accident Detection Using Deep Learning

APPENDIX B

PLAG REPORT

PVPIT, Department of Computer Engineering, 2022-23 41

 Car Accident Detection Using Deep Learning

APPENDIX C

CERTIFICATES

PVPIT, Department of Computer Engineering, 2022-23 42

 Car Accident Detection Using Deep Learning

PVPIT, Department of Computer Engineering, 2022-22 43