Seminar Report on

Car Crash Detection Using ML

Submitted by

Chandra Sekhar Nayak

Regd. No.: 2101229056

Seminar Report submitted in partial fulfillment of the requirements for

the award of Degree of B.Tech. in Computer Science & Engineering
under DRIEMS University

2021 - 2025

Under the Guidance of

Prof. Shekharesh Barik

Asso. Professor, Dept. of CSE

Department of Computer Science and Engineering

School of Engineering and Technology, Tangi, Cuttack-754022

 Department of Computer Science & Engineering
School of Engineering and Technology, Tangi, Cuttack - 745022

Certificate

This is to certify that this is a bonafide Seminar report, titled “Car Crash Detection
using ML”, done satisfactorily by Chandra Sekhar Nayak (2101229056) in partial
fulfillment of requirements for the degree of B.Tech. in Computer Science & Engineering
under DRIEMS University.

This Seminar report on the above mentioned topic has not been submitted for any other
examination earlier before in this institution and does not form part of any other course
undergone by the candidate.

Prof. Shekharesh Barik Dr. Mamata Rath

Asso. Professor, Dept. of CSE Asso. Professor & Head
Guide Dept. of CSE
 ACKNOWLEDGEMENT

I express my indebtedness to my guide Prof. Shekharesh Barik, Associate Professor of

the Computer Science & Engineering department who spared his valuable time to go
through manuscript and offer his scholar advice in the writing. His guidance,
encouragement and all out help have been invaluable to me. There is short of words to
express my gratitude and thankfulness to him.

I am grateful to all the teachers of Computer Science & Engineering department,

DRIEMS, for their encouragement, advice and help.

At the outset, I would like to express my sincere gratitude to Dr. Mamata Rath, H.O.D
of Computer Science & Engineering department for his moral support extended towards
me throughout the duration of this seminar.

I am also thankful to my friends who have helped me directly or indirectly for the success
of this seminar.

Chandra Sekhar Nayak

Regd. No.: 2101229056
Department of Computer Science & Engineering
School of Engineering and Technology, DRIEMS University
 ABSTRACT

Car crashes are a major cause of human fatalities and property damage worldwide. A reliable
car crash detection system is important for the safety of passengers and drivers. With recent
advances in machine learning, it has become possible to develop intelligent car crash
detection systems that can identify and respond to accidents in real-time. In this system,
various sensors, such as accelerometers, gyroscopes, and cameras, are used to collect data
from the vehicle. This data is then fed into a machine learning model, which is trained to
recognize patterns that are indicative of a car crash. The machine learning model can be
trained on a large dataset of pre-existing car crash data, which enables it to learn the patterns
and characteristics associated with different types of crashes. Once the model has been
trained, it can accurately detect car crashes and alert emergency services, potentially reducing
response times and increasing the chances of survival for the passengers and drivers involved
in the accident. Overall, a car crash detection system using machine learning has the potential
to greatly improve road safety by quickly and accurately detecting accidents, and
automatically notifying emergency services.

Keywords: Machine Learning, accelerometers, gyroscopes, Crash Detection System.

 CONTENTS

LIST OF FIGURES i
CHAPTER 1 1
1 INTRODUCTION 1
1.1 LITERATURE SURVEY 1
1.2 PROBLEM DEFINITION 2
1.3 MOTIVATION OF WORK 2
1.4 OBJECTIVES 3
CHAPTER 2 4
2 SYSTEM ANALYSIS AND DESIGN 4
2.1 PROPOSED METHOD 4
2.1.1 DATASET COLLECTION 5
2.1.3 BUILDING MODEL 5
CHAPTER 3 6
3 ALGORITHMS USED 6
3.1 LOGISTIC REGRESSION 6
3.1.1 TYPES OF LOGISTIC REGRESSION 7
3.2 RANDOM FOREST CLASSIFIER 7
3.2.1 WORKING OF RANDOM FOREST CLASSIFIER 8
3.3 DECISION TREE CLASSIFIER 8
CHAPTER 4 10
4 PROGRAM IMPLEMENTATION IN PYTHON 10

4.1 IMPORTING THE LIBRARIES 10

4.2 LOAD THE DATASET 10

4.3 MODEL BUILDING 10
4.4 USED MODEL 11
4.4 4.4.1 LOGISTIC REGRESSION ALGORITHM 11
4.4.2 DECISION TREE CLASSIFICATION 11
4.4.3 RANDOM FOREST CLASSIFIER 12
 4.5 MODEL EVALUATION 12
4.6 PREDICTIONS 13
CONCLUSION AND FUTURE SCOPE 15
REFERENCES 16
 LIST OF FIGURES

FIG NO. FIGURE TITLE PAGE NO

Fig 2.1 PREDICITION MODEL 4

Fig 2.2 DATASET SCREENSHOT 5
Fig 3.1 LOGISTIC REGRESSION 6
Fig 3.2 RANDOM FOREST CLASSIFIER 7
Fig 3.3 WORKING OF RANDOM FOREST CLASSIFIER 8
Fig 3.4 DECISION TREE CLASSIFIER 9
Fig 4.1 IMPORTING THE LIBRARIES 10
Fig 4.2 LOADING THE DATASET 10
Fig 4.3 SPLITTING THE DATA 11
Fig 4.4 USED ALGORITHMS 11
Fig 4.5 LOGISTIC REGRESSION 11
Fig 4.6 DECISION TREE 12
Fig 4.7 RANDOM FOREST 12
Fig 4.8 ACCURACY OF DATASET 13
Fig 4.9 PREDICTION ON GIVEN SENSOR DATA 13
Fig 4.10 DECISION MAKING ON THE BASIS OF PREDICTION 14

i
 Chapter 1

INTRODUCTION

Vehicle crashes cause injuries and deaths to road users as well as enormous economic and
social costs, and more than 50% of vehicle crashes are caused by careless driving. [1]. In
order to ensure driving safety, in recent years, all countries have researched car crash
avoidance technology. According to statistics, if you can give the driver an additional 0.5
second reaction time in a dangerous situation, you can reduce collision by 45%, so modern
cars are equipped with all kinds of measurement and alarm systems to keep driving safety.
Vehicle crash detection can be detected in a variety of ways, but g-sensors (accelerometer)
are usually used. This method is commonly used because it is cheap and possible to easily
determine the amount of change in the values of x, y, z. Today, being able to support humans
in their daily work by developing autonomous systems is one of the biggest challenges of
modern computer science. One example is an autonomous driving system that helps reduce
fatalities from traffic accidents. A variety of new sensors have been used in the past few years
for tasks such as recognition, navigation and manipulation of objects. A plenty of technology
has been aimed to research the crash detection in vehicle. Collision probability data generated
from Monte Carlo simulation taking driver behavior and vehicle dynamics into account,
tracking algorithm using interactive multi-model particle filter, and threat assessment
algorithm to estimate collision probability. In another method, two models are considered: a
model in which the follower maintains a safe distance and a model in which the follower
maintains a safe time. Analyze distance delays and time delays caused by major vehicles'
impact on followers. There is also a way to develop new challenging benchmarks for stereo,
optical flow, visual odometer/SLAM and 3D object detection tasks using autonomous driving
platforms.

1.1 LITRATURE SURVEY

 V. Machaca Arceda et al [1] In this project there aim is to detect car crash accident
detection in video.The first stage they used is convolutional neural network in this case
they used YOLO algorithm.The second stage tracker in order to focus each car.The final
stage for each car they have used the (VIF) Violent flow descriptor with a support vector
machine in order to detect car crashes. It gives around 89% accuracy.

1|Page
  Deeksha Gour et al [2] This project is based on the neural network and deep learning of
object detection along computer vision technology and several methods and algorithms.
This approach will work on images,recorded videos ,real-time live videos and will
detect, track and compute moving objects, velocity and direction using convolutional
neural networks. They used a YOLO algorithm which achieves its result by applying a
neural network on the image.
 Pratishtha Gupta el al [3] This project is predicated on image processing. This document
presents a replacement work on traffic images during which the traffic image is
processed to seek out if every individual vehicle goes within the straight or diagonal
direction when it crosses the junction.Intensity values are computed on each pixel in row
or column. It gives around 98% accuracy for detecting direction deviation.
 Vaishnavi Ravindran et al [4] The approach is based on detecting damaged vehicles from
footage received from surveillance cameras installed in roads and highways which would
indicate an occurrence of road accident. Detection of damaged cars falls under a category
of object detection in the field of machine vision technique. They used a supervised
learning method which works as a binary classifier to distinguish between images
containing a damaged car as class 1 and images not containing it as class 0.
 Sreyan Ghosh et al [5] The intent is to create a system which would detect an accident
based on the live video from a CCTV camera installed on highways. The idea is to take
each frame of video and run it through a deep learning convolutional neural network,
which has been trained to classify frames of video into accident or non accident. It gives
around 92.3%.

1.2 PROBLEM DEFINITION

The most challenge part of the project is to detection of vehicle crash and rescuing people
from accident by calling hospitality for emergency. Car crashes causes injuries and death due
to lack of medical help in isolated accidental areas and for that this report solves this problem
by detecting the crash and also calling for help using machine learning techniques.

1.3 MOTIVAION OF WORK

There are several potential motivations for using machine learning to detect car crashes using
accelerometer and gyroscope sensors. Here are a few:

2|Page
  Improve safety: By detecting car crashes more quickly and accurately, emergency
services can be notified faster and potentially save lives. Machine learning algorithms
can analyze sensor data in real-time and provide an automated response.
 Reduce insurance fraud: Some drivers may try to stage accidents in order to collect
insurance payouts. By accurately detecting real crashes, insurers can avoid paying out
fraudulent claims and reduce the costs associated with fraud.
 Enhance autonomous vehicle safety: Self-driving cars rely on sensor data to make
decisions and navigate the road. By using machine learning to detect crashes,
autonomous vehicles can more quickly respond to dangerous situations and avoid
accidents.

Overall, using machine learning to detect car crashes has the potential to save lives, reduce
insurance costs, and improve overall road safety.

1.4 OBJECTIVES

To achieve the aim stated above, the objectives of this investigation are to

 Conduct a systematic review of the state of the art in the areas of crash prediction and
constellation extraction algorithms, automotive sensors, sensor data fusion, and
trajectory planning.
 Conduct statistical analyses on accident databases to identify the most significant crash
parameters.
 Define possible collision scenarios for the design, implementation, testing, and
evaluation of the constellation extraction algorithm.
 Devise a novel algorithm or combination of existing algorithms for extracting the
constellation of a crash.
 Conduct experiments to assess the devised constellation extraction algorithm.

3|Page
 Chapter 2
SYSTEM ANALYSIS AND DESIGN

2.1 PROPOSED METHOD

Car Crash Detection System, the arrangement of accelerometer and gyroscope in a vehicle
may vary depending on the vehicle's space and design. In general, the black box is placed in
the center of the vehicle because it is considered to be placed in a safe location. In general, g-
sensor uses the first value read after system booting as the default value and measures the
change of the corresponding value. In this paper, the initial value is specified as 1.000 for
convenience and understanding of calculation. There are countless crash detection
algorithms. In this paper, the method of detecting impact through absolute values is as
follows.

Gyroscope
Sensor

Fig: 2.1 Prediction Model

|x2| – |x1| > x axis threshold
|y2| – |y1| > y axis threshold (2)
|z2| – |z1| > z axis threshold

The G-sensor value can reflect the period by sampling that is the changeable value. In this
paper, the sampling period is 10ms for testing. x1, y1, z1 are the previous values, and x2, y2,
z2 are the changed values after 10ms. When the previous value is subtracted from the
changed value as in Equation 2, if the corresponding value is higher than each predetermined
threshold value, it can be determined as the crash event.

4|Page
2.1.1 DATASET COLLECTION

This module includes data collection and understanding the data to study the sensors data for
a better prediction with accuracy.

The data set is in CSV (Comma Separated Value) format which is further prepared to data
frame as supported by panda's library in python.

Fig: 2.2 Dataset Screenshot

2.1.3 BUILDING MODEL

This is most important phase which includes model building for prediction of car crash
detection. In this we have implemented various machine learning algorithms for crash
detection. These algorithms include Random Forest Classifier, Decision Tree Classifier,
Extra Tree Classifier, Ada Boost algorithm, Perceptron, Linear Discriminant Analysis
algorithm, Logistic Regression, K-Nearest Neighbor, Gaussian Naïve Bayes, Bagging
algorithm, Gradient Boost Classifier.

But we use only three type of algorithms,

 Logistic Regression
 Random Forest Classifier
 Decision Tree Classifier
5|Page
 Chapter 3

ALGORITHMS USED

The main purpose of designing this system is to predict weather a vehicle has crashed or not.
We have used as Logistic Regression, Random Forest Classifier, Decision Tree Classifier in a
machine-learning algorithm to train our system.

3.1 LOGISTIC REGRESSION

Logistic regression is one of the most popular Machine Learning algorithms, which comes
under the Supervised Learning technique. It is used for predicting the categorical dependent
variable using a given set of independent variables. Logistic regression predicts the output of
a categorical dependent variable. Therefore the outcome must be a categorical or discrete
value. It can be either Yes or No, 0 or 1, true or False, etc. but instead of giving the exact
value as 0 and 1, it gives the probabilistic values which lie between 0 and 1. Logistic
Regression is much similar to the Linear Regression except that how they are used. Linear
Regression is used for solving Regression problems, whereas Logistic regression is used for
solving the classification problems. In Logistic regression, instead of fitting a regression line,
we fit an "S" shaped logistic function, which predicts two maximum values (0 or 1). Logistic
Regression can be used to classify the observations using different types of data and can
easily determine the most effective variables used for the classification. The below image is
showing the logistic function:

Fig: 3.1 Logistic regression

6|Page
3.1.1. TYPES OF LOGISTIC REGRESSION
Logistic Regression can be classified into three types
 Binomial: In binomial Logistic regression, there can be only two possible types of the
dependent variables, such as 0 or 1, Pass or Fail, etc.
 Multinomial: In multinomial Logistic regression, there can be 3 or more possible
unordered types of the dependent variable, such as "cat", "dogs", or "sheep"
 Ordinal: In ordinal Logistic regression, there can be 3 or more possible ordered types of
dependent variables, such as "low", "Medium", or "High".

3.2 RANDOM FOREST CLASSIFIER

It can be used for both Classification and Regression problems in ML. It is based on the
concept of ensemble learning, which is a process of combining multiple classifiers to solve a
complex problem and to improve the performance of the model. As the name suggests,
"Random Forest is a classifier that contains a number of decision trees on various subsets of
the given dataset and takes the average to improve the predictive accuracy of that dataset."
Instead of relying on one decision tree, the random forest takes the prediction from each tree
and based on the majority votes of predictions, and it predicts the final output. The greater
number of trees in the forest leads to higher accuracy and prevents the problem of overfitting.

7|Page
 Fig: 3.2 Random Forest Classifier

3.2.1 WORKING OF RANDOM FOREST CLASSIFIER

Random Forest works in two-phase first is to create the random forest by combining N
decision tree, and second is to make predictions for each tree created in the first phase.
Step:1- Select random K data points from the training set.
Step:2- Build the decision trees associated with the selected data points (Subsets).
Step:3- Choose the number N for decision trees that you want to build.
Step:4- Repeat Step 1 & 2.
Step:5- For new data points, find the predictions of each decision tree, and assign the new
data points to the category that wins the majority votes.

Fig: 3.3 Working Of Random Forest Classifier

3.3 DECISION TREE CLASSIFIER

Decision Tree is a Supervised learning technique that can be used for both classification and
Regression problems, but mostly it is preferred for solving Classification problems. It is a
tree-structured classifier, where internal nodes represent the features of a dataset, branches
represent the decision rules and each leaf node represents the outcome. In a Decision tree,

8|Page
there are two nodes, which are the Decision Node and Leaf Node. Decision nodes are used to
make any decision and have multiple branches, whereas Leaf nodes are the output of those
decisions and do not contain any further branches. The decisions or the test are performed on
the basis of features of the given dataset. It is a graphical representation for getting all the
possible solutions to a problem/decision based on given conditions. It is called a decision tree
because, similar to a tree, it starts with the root node, which expands on further branches and
constructs a tree-like structure. In order to build a tree, we use the CART algorithm, which
stands for Classification and Regression Tree algorithm. A decision tree simply asks a
question, and based on the answer (Yes/No), it further split the tree into subtrees.

Fig: 3.4 Decision Tree Classifier

9|Page
 Chapter 4
PROGRAM IMPLEMENTATION IN PYTHON

4.1 IMPORTING THE LIBRARIES

Fig: 4.1 Importing the libraries

4.2 LOAD THE DATASET

The datasets consist of several medical predictor (independent) variables and one target
(dependent) variable, Outcome. Independent variables include the number of pregnancies the
patient has had, their BMI, insulin level, age, and so on.

Fig: 4.2 Loading the dataset

4.3 MODEL BUILDING

A machine learning model is built by learning and generalizing from training data, then
applying that acquired knowledge to new data it has never seen before to make predictions
and fulfill its purpose.

10 | P a g e
 Fig: 4.3 Splitting the Data

4.4 USED MODEL

In this project we are using different algorithms

Fig: 4.4 Used algorithms

4.4.1 LOGISTIC REGRESSION ALGORITHM

Logistic regression predicts the output of a categorical dependent variable. Therefore the
outcome must be a categorical or discrete value. It can be either Yes or No, 0 or 1, True or
False, etc. but instead of giving the exact value as 0 and 1, it gives the probabilistic values
which lie between 0 and 1. After building the model using logistic regression algorithm let’s
check the accuracy of the model on the training dataset.

Fig: 4.5 Logistic Regression

4.4.2 DECISION TREE CLASSIFICATION

Decision trees help you to evaluate your options. Decision Trees are excellent tools for
helping you to choose between several courses of action.Build the model using decision tree
to get the accuracy score.of the model on the training dataset.

11 | P a g e
 Fig: 4.6 Decision Tree

4.4.3 RANDOM FOREST CLASSIFIER

Random Forest is a classifier that contains a number of decision trees on various subsets of
the given dataset and takes the average to improve the predictive accuracy of that dataset.
Build the model using random forest classifier to get the predictions and accuracy from the
testing dataset

Fig: 4.7 Random Forest

4.5 MODEL EVALUATION

Model Evaluation is the process through which we quantify the quality of a system’s
predictions. To do this, we measure the newly trained model performance on a new and
independent dataset. This model will compare labeled data with its own predictions.

12 | P a g e
Accuracy is the number of times you predicted something correctly divided by how many
times you actually predicted it.

Fig: 4.8 Accuracy of Dataset

4.6 PREDICTIONS

After getting all accuracy now we creating a function that will predict that a vehicle met with
crash or not.

Fig: 4.9 Prediction on given sensor data

Prediction Decision

 Then we will take input for prediction. Input sequence: x, y, z coordinates of

accelerometer sensor data and x, y, z coordinates of gyroscope sensor data.

13 | P a g e
Fig: 4.10 Decision making on the basis of prediction

14 | P a g e
 CONCLUSION AND FUTURE SCOPE

Automated accident detection systems help in reducing fatalities emanating from vehicular
accidents by decreasing the emergency response time. Real-time accident detection and
emergency alert can save lives. In this work we have presented a smartphone based accident
detection system which augments data from sensors such as accelerometer, gyroscope and
microphone to reliably detect a vehicular crash. Proposed models based only on acceleration
thresholds have false positives in absence of GPS. In-vehicle airbag deployment is triggered
at acceleration above or equal to 60g whereas mild or medium accidents produce acceleration
only up to 40g which may result in serious injury to the driver or passenger on-board. Hence,
keeping the thresholds high will definitely reduce false positives but it will introduce false
negatives, and keeping it low will bring false positives along with actual true positives (crash
events). Additionally, smartphones are capable of providing rich amount of data for accident
analysis, like data from multiple sensors, including pictures and videos, etc. In accidents
where smartphone itself is completely damaged, this solution may not be able to deliver what
it promises to. Hence, it may not be possible to beat the accuracy of expensive in-vehicle
systems equipped with high-end accident detection technologies with a smartphone sensor
based application, as they have the obvious advantage of direct access to the vehicle.
However, there might be several scenarios where the smartphone is actually not damaged
wherein such a solution can result into life-savings. Especially in developing countries where
people in general cannot afford those pre-installed vehicle accident detection systems because
of the cost involved, this solution becomes even more important and will have a significant
impact on first emergency/medical response team reaching the accident location on time and
thus saving lives. Therefore, having a solution with similar level of accuracy which can make
a difference as big as saving lives of people is better than having no other alternative at all.
As a part of future work we are investigating whether pictures/videos from smartphone and
data from multiple mobile gadgets like smart watch, heart rate monitor or other smart
wearables can be used to identify and improve the accuracy of analyzing severity of an
accident. This data captured during an accident or at the location can help emergency
responders to better analyze the accident and situation of injured people or victims and act
accordingly. Moreover, this solution can be integrated with traffic congestion monitoring
solutions to divert traffic from accident location.

15 | P a g e
REFERENCES

1. Lee, K.B. , Shin, H.S., 2019, August. An application of a deep learning algorithm for
automatic detection of unexpected accidents under bad CCTV monitoring conditions in
tunnels. International Conference on deep learning and machine learning in emerging
applications (Deep- ML) (pp. 7-11), IEEE, 2022
2. T. Kim , H. Jeong, A Novel Algorithm for Crash Detection Under General Road Scenes
Using Crash Probabilities and an Interactive Multiple Model Particle Filter, IEEE
Transactions on Intelligent Transportation Systems, vol. 15, no. 6, pp. 2480-2490, Dec.
2022.
3. V. Naumov, Analysis of Time and Distance Delays in Car Following Models,” 2010
International Conference on Intelligent Systems, Modelling and Simulation, Liverpool,
pp. 296-299, 2010
4. Tian, D., Zhang, C., Duan, X. , Wang, X, An automatic car accident detection method
based on cooperative vehicle infrastructure systems. IEEE Access, 7, pp.127453- 127463,
, 2021
5. H. L. Wang, M. A. Jia-Liang, A design of a smart car accident rescue system combined
with WeChat platform, J. Transp. Eng., vol. 17, no. 2, pp. 48–52, Apr. 2020.
6. V. Machaca, Arceda E., Laura Riveros. Fast Car Crash Detection in Video, Oct-2018.
7. Deeksha Gour, Amit Kanskar. Automated AI based road traffic Accident alert system,
2019.
8. Pratishtha Gupta, Manisha Rathore, G N Purohit. Detection of Direction Deviation of
vehicle using CCTV Cameras, 2014.
9. Vaishnavi Ravindran, Lavanya Vishwanathan, Dr. Shanta Rangaswamy. A novel
approach to Automatic Road Accident detection using machine vision techniques, 2018
10. R. R. Knipling, Assessment of IVHS countermeasures for collision avoidance systems,
National Highway Traffic Safety Administration, Washington, DC, USA, Tech. Rep.
DOT HS 807 995, 2017.

16 | P a g e