0% found this document useful (0 votes)
156 views4 pages

Accident Detection and Warning System

Roads are monitored via way of means of CCTV cameras passively i.e. expertise isn't always generated via way of means of the CCTV cameras
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
156 views4 pages

Accident Detection and Warning System

Roads are monitored via way of means of CCTV cameras passively i.e. expertise isn't always generated via way of means of the CCTV cameras
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Accident Detection and Warning System


1 2
Abishek T S Adith Jose
Student, Computer science and Engineering Department Student, Computer science and Engineering Department
Sahrdaya College of Engineering & Technology Sahrdaya College of Engineering & Technology
Thrissur, India Thrissur, India

3 4
Albin Davis Alan Thomas
Student, Computer science and Engineering Department Student, Computer science and Engineering Department
Sahrdaya College of Engineering & Technology Sahrdaya College of Engineering & Technology
Thrissur, India Thrissur, India
5
Jasmy Davies
Asst. Professor, Computer science and Engineering Department
Sahrdaya College of Engineering & Technology
Thrissur, India

Abstract:- Roads are monitored via way of means of I. INTRODUCTION


CCTV cameras passively i.e. expertise isn't always
generated via way of means of the CCTV cameras. More than one million people died in various traffic
Human useful resource constraint i.e. there may be the accidents around the world, and many people suffered minor
very restricted posting of the visitor's police personnel at injuries. Many studies have shown that many developing
road crossings, highways, and far off places. In many and underdeveloped countries have the highest road traffic
cases, injuries are happened because of negligence via accident death rates, even though these countries only
way of means of the pedestrian and being stranded in account for half of the world’s vehicles. According to the
far-off places. Adding to this, even the bulk of human data available in India, there is an average of 13 deaths per
beings surrounding the coincidence scene are busy hour, that is, 140,000 deaths per year. The main goal is to
clicking photographs and videos ignorant of the truth enable the system to detect accidents based on the video
that their little negligence should value a life. The photos sequence transmitted by the camera. A tool that helps
are available to the government agencies after trouble accident victims who need it by detecting the accident early
has happened. To conquer this, we endorse a wise device and notifying the authorities from then on. The goal is to
that could employ the prevailing CCTV cameras with detect accidents in seconds by using advanced deep-learning
nighttime vision. The proposed device captures the video algorithms that use convolutional neural networks (CNN or
stream, computes the input, and the device signals are ConvNet) to analyze the frames from the video generated by
generated in real-time, because of this that no extra the camera. We focus on installing the system on busy
sensors could be required. Using the digital digicam roads. Incomplete and timely assistance to accident victims
itself, the plan is to discover injuries in real-time and is rare. On the street, we can install CCTV cameras. In this
additionally send signals to ambulances or firefighting camera, we can set up a proposed system that will take shots
services; in order that suitable assets wished for saving from CCTV cameras and run them on the proposed collision
lives are to be had in time. In many cases, injuries are detection model to detect accidents. In the event of an
suggested as a result of negligence on the part of street accident, response time is of the utmost importance, and
customers and being stranded in remote locations. assistance is provided every additional minute since arrival
Furthermore, the vast majority of people in the vicinity can determine life and death. A system is needed that can
of the accident scene are preoccupied with taking automatically detect and track traffic accidents, and quickly
photographs and videos, oblivious to the fact that their report them to relevant departments to respond to
inattention could cost a life. To overcome this, we emergencies more quickly. Thousands of lives The
recommend a clever gadget that can make use of existing computing power of modern CPUs makes many complex
CCTV cameras with night vision. real-time applications possible. One of the most common
real-time applications is video surveillance systems. CCTV
Keywords:- Convolutional Neural Network, ResNet, Traffic systems reduce the presence of people following the actions
net, openCV, LSTM, GPS, REST. recorded by cameras. One of the advantages of the visual
observation system can be saved and analyzed. Future
reference One of the most important applications of video
surveillance systems is traffic monitoring. The traffic
monitoring system is used to detect, track, evaluate traffic
flow, determine vehicle speed, classify vehicles, etc.

IJISRT21JUN878 www.ijisrt.com 1532


Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. RELATED WORK C. Ubiquitous GPS Vehicle Tracking and Management
System.
A. Accident Detection using Convolutional Neural Iman M. Almomani, Nour Y. Alkhalil, Enas M.
Networks. Ahmad and Rania M. Jodeh[3], “Ubiquitous GPS Vehicle
Sreyan Ghosh, Sherwin Joseph Sunny and Rohan Tracking and Management System”, This document
Roney[1], ”Accident Detection using Convolutional Neural proposes a system to make vehicle tracking more accessible
Networks”. CNN's are used in the modeling of spatial data to individual users and fleet companies. The proposed
such as photographs. CNN had been successful in tasks such system provides not only traditional web-based tracking
as photo classification, item detection, and so on. LSTMs software but also mobile software for tracking anytime,
are algorithms that are used to model sequential data and anywhere. Services for a variety of users. This system
make predictions based on it. LSTMs are widely used in the allows people to track the position, speed, stop, and
textual content category for creating language models, series movement of their vehicles. The monitoring process
generation, and so on. Standard LSTMs can be applied to includes real-time tracking of location or collection of
sequential data with a spatial center right away. As a result, historical reports on vehicle movement, setting speed, and
a CNN-LSTM structure is desired to carry out tasks geographic limits. Services (GPRS) Global System for
involving photo or video sequences. For non-stop video Mobile Communications (GSM) Internet or World Wide
taken from a camera, the proposed version is a fusion of Web and Global Positioning System (GPS)
CNN and LSTM layers. The CNN portion of the proposed
version was particularly stimulated by Inception v3, but with D. Improving Estimation Of Vehicle’s Trajectory Using the
a few tweaks, it has fitted well to our education Latest Global Positioning System With Kalman Filtering
photographs. The LSTM layers have been added to the “Improving Estimation Of Vehicle’s Trajectory Using
existing Convolution Network to recall temporal as well as the Latest Global Positioning System With Kalman
spatial capabilities. This is further divided into the version's Filtering”[4] Predicting a vehicle's future location with
convolution and recurrent elements. The CNN is frequently accuracy is a critical but difficult problem in intelligent
used in a CNN-LSTM network for function extraction from transportation systems.It can be advantageously employed in
photographs, which is then passed directly to the LSTM for car or robot obstacle avoidance systems.Many of the
series prediction. They are widely used in tasks such as currently being researched obstacle-avoidance systems are
activity recognition, image description, and video limited to line-of-sight sensors. Much research looks into the
description, among others. possibility of using GPS data received from various vehicles
to forecast where each car will be in the future.
B. Accident Detection, Severity Prediction, Identification of
Accident Prone Areas in India and Feasibility Study using The methods used to construct these predictions are
Improved Image Segmentation, Machine Learning and rather simple, and they do not produce particularly accurate
Sensors. results in cases where the predicted future position of the
“Accident Detection, Severity Prediction, vehicles is not a straight route, such as curves. The current
Identification of Accident Prone Areas in India and research shows that we need a better technique to anticipate
Feasibility Study using Improved Image Segmentation, the trajectory of vehicles in a variety of conditions. The
Machine Learning and Sensors”[2], In this paper, The Kalman filter (KF) is used in this situation. The KF has a
system provides a three-stage solution that uses machine long history of reliably predicting future states of moving
learning and computer vision to analyze traffic accidents in objects, and it has been used in a variety of applications.
India. The solution is through the classification of car
accidents. This step can be accomplished by using any III. PROPOSED SYSTEM
object detection and image segmentation algorithm. We
experimented with the You Look Only Once (YOLO) The proposed model is a fusion of CNN and LSTM
algorithm. However, the probability of accident layers for the classification of continuous video captured
misclassification is as high as, which becomes a major from the camera. The CNN part of the proposed model was
problem when using real-time data. Image segmentation is mainly inspired by the ResNet-50 but fit well with the
an alternative method of classifying car accidents. We training image through certain adjustments. The LSTM
experimented with the watershed algorithm, the cunning layer has been added to the existing Convolution Network to
edge detection, and the cunning automatic algorithm for take into account spatial and temporal characteristics. This is
separating cars from accidental cars. The YOLO algorithm divided into a convolution part and a recurrent part of the
found that the limitation is that the region of interest is model. CNN-In the LSTM network, CNNs are primarily
satisfactory, but it was misclassified and lost some of the used for image feature extraction and are passed to LSTMs
cars indicated in the input video image. Cunning edge for sequence prediction. Widely used for tasks such as
detection performs best in car accident classification and has activity recognition, image description, video, description,
a very interesting area, making it suitable for good etc. CNN is used to model spatial data such as images. CNN
preprocessing technology for image enhancement of car has been successful in tasks such as image classification and
accident images. object detection. The LSTM is used to model sequential
data and make predictions based on it. LSTMs are widely
used in areas such as text classification, language modeling,
and sequence creation. The standard LSTM can be used

IJISRT21JUN878 www.ijisrt.com 1533


Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
directly on the input space, sequential data. So, if you want B. pre-processing
to work with a series of photos or videos, you need to use The collected data set is pre-processed before feeding
the CNN-LSTM architecture. In the GPS, various it into the CNN model. This preprocessing makes the data
characteristics such as current location data, vehicle speed, compatible for the model and also keep it homogeneous.
and direction are taken into account in the prediction. When The pre-processing is carried out using the openCV library.
the machine learning model detects an accident, the This library is available in python.
proposed system sends an alert to the appropriate authorities
and nearby vehicles. C. Building And Training The ResNet-50 Model
For the main implementation of the ResNet-50 model
IV. METHODOLOGY architecture, an open source library called imageai is used.
ResNet-50 is a 50-layer deep convolutional neural network.
We collected datasets from an online data store called A pre-trained version of the network trained on over a
traffic-net, which categorizes datasets as accidents, crowded million images from the ImageNet database can be loaded.
traffic, fires, and sparse traffic. This is part of DeepQuest AI The pretrained network can categorise images into 1000
that trains you to recognize, understand, and fix problems different object categories.As a result, the network has
accordingly in all environments where machine learning learned detailed feature representations for a diverse set of
systems are deployed. Contains 4400 images containing 4 images. The network accepts images with a resolution of
classes. There are 1,100 images in each of the categories, 224 by 224. The below image show a conceptual
and 900 educational images, and 200 test images. An architecture of ResNet.
Android application was used to alert real-time. The
coordinates of the incident are sent to the Android client
user and compared with the user's location. this helps users
make appropriate decisions.

A. Architecture
The video stream from CCTV is directly sent to a pre-
processing part. This preprocessing is carried out by
openCV library, which converts the video into frames of
images. These images will be sized and formatted in order to
be compatible with the ResNet-CNN model.

The CNN model used here is a ResNet50[5]. which is Fig 2. ResNet-50


implemented on a highly functional computer system with a
high clock rate. These CNN models are system independent This model is trained using the collected datasets. The
hence migration and implementation to any other machine training is conducted in a supervised learning fashion where
including cloud computing systems are hassle free. An the example data sets along with its class value is fed into
android application will be used for the user interface. The the model for its learning. for training the model image
alert system uses REST API for communicating with users frames of collected data sets are used, which are
and appropriate authorities over HTTP protocol. A relational preprocessed to fit the models dimensions and pixel
database stores user's data. resolution.

D. Alerting system
The alerting module checks the ResNet model’s output
in realtime, if the probability of an accident is high it sends
alerts towards the user. We use an android application for
client side. The coordinate of accident location is forwarded
to the android application, along with it alert messages are
passed to emergency response teams like ambulance and fire
engine. To communicate with the android application the
alerting system which is written in python uses REST API.
When the location of an accident is passed, the user can then
choose to take a different route if he wishes to.

Fig 1. Architecture

IJISRT21JUN878 www.ijisrt.com 1534


Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. RESULT responders with essential information as fast as possible but
unavailability of this system are restricted by their non-
The following are the results we got during the testing portability and costs.The proposed vehicle accident
of model on unseen data. detection system can track an accident at its moment of
occurrence. Compared with other deployment systems
composed of expensive sensors and unnecessary hardware,
the proposed system is more economical, more reliable, and
more accurate than similar systems, mainly due to the
model-based approach.

VII. FUTUREWORK

In the future , we can combine both supervised and


unsupervised methods together to improve the system . We
can use supervised learning models to identify the accidents
from the frames which are flagged anomalously by
unsupervised models.

More sophisticated methods for tracking accident


happened area. Algorithms like kelman can be used to
predict the potential of a vehicle passing through the
accident happened place. this will help the driver to avoid
Fig 3. image frame with Dense traffic incident happen area and evade a potential traffic jam.

REFERENCES

[1]. Sreyan Ghosh, Sherwin Joseph Sunny and Rohan


Roney, “Accident Detection using Convolutional
Neural Networks” ,2019 International Conference on
Data Science and Communication (IconDSC),
Bangalore, India,DOI:
10.1109/IconDSC.2019.8816881
[2]. Vipul Gaurav, Sanyam Kumar Singh and Avikant
Srivastava, “Accident Detection, Severity Prediction,
Identification of Accident Prone Areas in India and
Feasibility Study using Improved Image Segmentation,
Machine Learning and Sensors ”, 22-10-2019 ,IJERT
[3]. Iman M. Almomani, Nour Y. Alkhalil, Enas M.
Ahmad and Rania M. Jodeh, “Ubiquitous GPS Vehicle
Tracking and Management System”, 2011 IEEE
Jordan Conference on Applied Electrical Engineering
and Computing Technologies (AEECT), DOI:
10.1109/AEECT.2011.6132526
Fig 4. image frame with accident [4]. Cesar Barrios and Yuichi Motai, “Improving
Estimation Of Vehicle’s Trajectory Using the Latest
VI. CONCLUSION Global Positioning System With Kalman Filtering”,
IEEE Transactions on Instrumentation and
Accident detection operation is not an easy task to Measurement , 19 May 2011, DOI:
handle; it can be an extremely complicated process when it 10.1109/TIM.2011.2147670
comes to real time applications, which is the main reason [5]. ResNet50:https://in.mathworks.com/help/deeplearning
why it is not implemented yet on a large scale. The proposed /ref/resnet50.html;jsessionid=8ba3118b0f95ca5feb6af
system will help to improve the present scenarios. Although 9abc537
an in-vehicle accident detection system provides emergency

IJISRT21JUN878 www.ijisrt.com 1535

You might also like