Machine Learning-Based Intelligent Accident Detection and Notification System in IoT Network

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

2023 11th International Conference on Emerging Trends in Engineering & Technology - Signal and Information Processing (ICETET - SIP)

| 979-8-3503-4842-2/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICETET-SIP58143.2023.10151583

2023 11th International Conference on Emerging Trends in Engineering & Technology - Signal and Information Processing
(ICETET - SIP)

Machine learning-based intelligent accident detection and


notification system in IoT network
Gopal Sunil Rathod Dr. Kapil Jajulwar Mr. Umesh Kubde
Department of Electronics and Department of Electronics and Managing Director
Telecommunication Engineering, Telecommunication Engineering, LGPS Hybrid Energy Pvt. Ltd.
G H Raisoni College of Engineering, G H Raisoni College of Engineering, Narendra Nagar, Nagpur,
Nagpur, Maharashtra, India Nagpur, Maharashtra, India Maharashtra, India – 442001
gopalrathod189@gmail.com kapil.jajulwar@raisoni.net lgpsenergy@gmail.com

Abstract - The number of traffic accidents is increasing every Once the system identifies an incident, it can notify all major
day as the number of cars increases. It is said that 3 million solutions such as medical facilities, police departments, etc.
people die each year worldwide and 55 million people are injured. In the past decades, several mechanisms for accident
Lack of emergency medical care at the scene of an accident or situation monitoring and warning have been published. Global
long post-accident response times are the leading causes of death.
positioning systems are predominantly used in most of the
Cognitive aids can be used to detect incidents and alert traffic
watchers or rescue teams, reducing waiting times for lifesaving current methods of locating vehicles. Your car may have a
operations. Transportation systems have received considerable GPS that detects your accidental location and sends it to a
attention from industry and academia due to the growing demand server in the cloud. Another popular collision warning solution
for urban exploration and have been seen as a way to improve that every car should have an facility for emergency
road safety in these areas. In this study, an IoT and machine notifications. The European Council has formed an emergency
learning-based intelligent rescue team accident detection model call facility that should be present in all cars manufactured
was presented. An IoT module has been developed that can from 2016 onwards. To collect accident data and send it to a
recognize an accident, collect all accident data such as position,
cloud server, this study suggest an intelligent accident-
pressure, force and speed, and transmit all relevant data to the
cloud. Machine learning models are used in the cloud to validate
detection & alert system. I'm here. It uses machine learning
IoT module data and activate emergency modules after a failure techniques on cloud servers to assess the severity of accidents
is detected. and notify emergency services and medical services if
essential.
Keywords - Accident Detection System, Machine Learning,
IoT.
Accident data is sent to rescue teams for emergency medical
I. INTRODUCTION
care. The main contributions of this research are:
As the population grows, the need for transportation
systems grows geometrically. The sharp increase in traffic • Proposal of an alarm system that utilizes a machine
accident rates over the last few decades should be of concern learning model for detecting traffic accidents and IoT.
to everyone. According to a detailed study by the WHO, road • A Compare the performance of the suggested system
accidents are becoming her 7th leading cause of death with the present system.
globally, killing billions of folks each year. ASIRT guesses Abstract of this paper: Section 2 introduces various present
that road accidents will be the fourth leading cause of death in techniques and methods for accident detection. Section 3
the future. The study found that in 2020 he recorded 61 presents the complete architecture of the proposed accident
accidents, 18 deaths and 78 injuries every hour. Lives are at detection and warning system. Section 4 presents performance
risk on the autobahn through irresponsible behavior and estimation factors for measuring the performance of the
deliberate disregard for driving rules. Accidents can occur for suggested system. Section 5 presented an analysis of the
many reasons, including speeding, driving through tight results of the suggested classifier. Lastly, we discuss
spaces, cell phone use, and bad weather. Previous research completion and future scope.
found that high speeds were a major cause of accidents.
Geographical localization of the accident site is extremely II. RELATED WORK
important for rescue. Emergency assistance is provided Investigators have focused on the important issue of traffic
quickly in high traffic areas and urban areas, but it can be accident identification and warning systems. Over the past
difficult to provide quickly in rural areas and highways. years, many dissimilar accident detection and warning methods
Delayed treatment has been shown to result in serious injuries have been suggested by several authors. This section describes
leading to death. The time need for an ambulance to reach at some of them. A accelerometer and GPS information were
the scene of an accident and take the person to a doctor has a used to determine the location of the accident. This technique
important effect on a victim's chances of survival. It's getting a does not actually alert emergency services. Just identify the
lot of attention Intelligent methods for accident detection and accident. An accident detection technique has been proposed
notification are needed to decrease the total traffic casualty that uses a GPS controller to locate the accident and send an

979-8-3503-4842-2/23/$31.00 ©2023 IEEE


rized licensed use limited to: Vignan's Foundation for Science Technology & Research (Deemed to be University). Downloaded on January 19,2024 at 06:21:08 UTC from IEEE Xplore. Restrictions a
alert notification to the nearby emergency medical hospital. A An collective deep learning method to camera crash
major disadvantage of this learning is that only a single sensor detection was suggested in Audio and visual data from
was used. Hence, if the sensor is disabled, the whole structure cameras are collected in the proposed approach, and once
will not work. It can also lead to more false positives. The accidents are detected and classified, we use CNN and a deep
proposed system uses vibration sensors and microprocessors to learning classifier called GRU. The main disadvantage of this
detect and prevent car crashes. Microcontrollers are connected technique was camera was mounted on the dash and could be
to elements such as sensors, GSM, and GPS. Once the location spoiled in an accident. Since this technique depend on only on
of the disaster is identified, the sensor converts the information cameras and does not practice an IoT gateway, it can rise the
from the sensor, including speed, pressure, vibration, etc., and number of wrong/false alarms. In we suggested a mixture of
sends an alert to any location earlier entered emergency machine and deep learning classifiers SVM and Convolutional
number. Emergency efforts cannot be held via such Neural Network, correspondingly, to offer an automated
mechanisms. An IOT established alarm and emergency system, system for noticing car crashes. Various feature selection
using GPS and accelerometers to gather complete data and talk methods are used in the research work to choose the maximum
with a cloud server, was proposed. significant types from the delivered set of types. According to
In the event of an accident, an alert notification will be researchers, the maximum accuracy during the test period is
directed to all listed mobile phone numbers. Proposed deep 86%, which is not enough for use in industrial cases.
learning model for accident detection and saving systems Wi-
Fi cameras are already installed on highways. Continuously III. PROPOSED METHODOLOGY
record traffic video and send to cloud server. Visual For automatic accident detection and notification. It
examination is done in the cloud using deep neural networks. utilizes Internet of Things and machine learning to achieve
Once an accident is discovered, the data will be sent to the higher prediction performance. The recommendation system
nearest rescue team. A mobile-based app has been developed he consists of two steps. The first stage, which utilizes Internet
for accident detection and warning. Only GPS systems and of Things and machine learning model. So it can detect
wireless accelerometers were used to find the accident. When accidents, and the second stage is to respond to emergencies
an accident is discovered, a message is sent to a registered notifications.
mobile phone and sent. Since the proposed application uses a
single sensor, failure of that sensor leads to failure of the entire A. Stage-1
application.
Developed with another Android-based application that Machine Learning and IoT subsection should look like
uses a mobile phone to detect accidents and produce alerts by this:
an accelerometer and GPS system. The suggested application • Hardware Section: The main feature is the hardware
will be used to alert the nearest emergency crew and saving at the heart of the proposed accident detection system.
them from the accident sight. IoT-based car crash detection Uses many IoT sensors and gadgets. To power the
and emergency saving mechanisms are suggested in The entire chassis board, he needs his second 5V DC
design uses several hardware for example GPS devices, power supply connected to the mains. The following
buzzers, and vibration sensors to find and alert an accident. section provides an overview of the many sensors and
After the location of the accident is determined, the alarm will other components used in the proposed system.
sound for 2 minutes and, if not stopped, the online facility will • Sensors: This forms the basis of the proposed
send an alarm announcement to your listed mobile phone collision detection system. FlexiForce sensors are
number. Computer Intelligence Models for Incident Detection used to measure vehicle forces. It has the largest
and Emergency Systems The pre-trained model was used. This sensor area of 19.8mm with a 2-pin connector. This
model was trained to identify accidents in dissimilar weather sensor has a force range of 0-25 pounds only. The
conditions, including monsoons, heavy fog, and poor visibility. operating range can be changed by adjusting the
An image augmentation approach was adopted to extend the control voltage and the magnitude of the output
database. resistance.
Combines IoT and artificial intelligence algorithms to • Microcontroller: Microcontrollers are used to direct
identify incidents and activate emergency subsystems. connections to sensors and manage multiple devices.
Researchers first collect data about incidents involving IoT An impact is recorded when the force exceeds 400 Pa.
devices and send it to the cloud for analysis. Such methods If a collision detection triggers an alarm.
process actual information on-site by fog handling instead of • Raspberry Pi: board is connected to a Wi-Fi network
directing it to the cloud for estimation. Based on the and a cloud server to upload photos and videos to
complexity of the input video system information, the assess the severity of events.
proposed method uses machine learning techniques to • GPS/GSM: Support rescue teams and communicate
recognize accidents that might lead to incorrect results. with other rescue services such as police stations,

rized licensed use limited to: Vignan's Foundation for Science Technology & Research (Deemed to be University). Downloaded on January 19,2024 at 06:21:08 UTC from IEEE Xplore. Restrictions a
ambulances, and families. It is connected to a
microcontroller and used to identify problems.
• Machine Learning Module: Thanks to decades of
progress, machine learning techniques can now be
used more efficiently across all industries. Most
incident detection systems in use today simply use
machine learning and various sensors to identify
events. So, random data is sent to the rescue team. As
there is usage of sensors hence, More accurate in false
positives. If the output of the machine learning model
is an accident, only the accident scene is reported to
the rescue team.

Fig.1. Block Diagram of accident detection system.


Fig.3. flow accident detection system
B. Stage-2

Dataset Description: The dataset contains a total of 600 IV. PERFORMANCE ESTIMATION
random and non-random images.
Efficiency for suggested model was estimated by means of
Alert System: Precision, Recall, F1-Score and Accuracy. This is the equation
that provides the measure of evaluation.
In the accident detection for the video input using the
camera or the pre-recorded videos as input data. Processing
section is used to read the input data frame by frame and
processed data should give to the convolution neural network.
The will do the object detection on the input data and it will
give us the bounding boxes also need to calculate the depth
estimation. This data will go to an accident detection
algorithm.

Where,
TP - true positive,
FP - false positive,
TN - true negative,
FP - false positive.

Fig.2. Alert system.

Based on this algorithm accidents detection will happen.


Once the accident is detected the alert system will send alters
to the medical teams and the police station.

rized licensed use limited to: Vignan's Foundation for Science Technology & Research (Deemed to be University). Downloaded on January 19,2024 at 06:21:08 UTC from IEEE Xplore. Restrictions a
V. RESULT ANALYSIS

The machine learning model implementation with basic


Python programming and test images are scaled to 150 × 150
with random forest and support vector machines.

Fig.7. Non-accident image data bars represent.

Fig.4. Non accident detection


VI. CONCLUSION AND FUTURE SCOPE

Due to the perfect road conditions in the modern city,


more road accidents happen when drivers drive their vehicles
at higher speeds. Despite the advent of many accident
identification and warning systems, many deaths continue to
occur. The problem is compounded, only in part, by poor
automated incident detection, inadequate alarms, and
inadequate emergency response. Two machine learning-based
models are used to evaluate outcomes on the IoT network and
perform emergency services. In the first phase of the proposed
system, use IoT components for accident detection.
Fig.5. Accident detection
IoT sensor for vehicle impact assessment and GPS module
for vehicle speed determination and accident detection, and
Figures show the performance estimation for this
sends all relevant vehicle data to a cloud server. Pre-trained
suggested machine learning models and random and non-
models with random forests and support vector machine are
random image data. random forest classifiers performed well
used in the second stage to reduce false detection rates and
for random and non-random image data. It gives 99% accident
send alerts to the closest tracing groups. In the future, safety
and non-accident detection accuracy.
features will be added to the proposed machine learning model
and to address fatigue detection.

REFERENCES
1] World Health Organization. Global Status Report on Road Safety;
World Health Organization: Geneva, Switzerland, 2018.
[2] Asirt.org. Road Crash Statistics. Available online:
https://www.asirt.org/safe-travel/road-safety-facts/ (accessed on 30
January 2019).
[3] Statistics. Available online: https://morth.nic.in/road-accident-in-
india (accessed on 13 June 2022).
[4] Comi, A.; Persia, L.; Nuzzolo, A.; Polimeni, A. Exploring Temporal
and Spatial Structure of Urban Road Accidents: Some Empirical
Evidences from Rome. Adv. Intell. Syst. Comput. 2018, 879, 147-155.
[5] Liu, J.; Khattak, A.J.; Li, X.; Nie, Q.; Ling, Z. Bicyclist injury
severity in traffic crashes: A spatial approach for geo-referenced crash
data to uncover non-stationary correlates. J. Saf. Res. 2020, 73, 25-35.
Fig.6. Accident image data bars represent.

rized licensed use limited to: Vignan's Foundation for Science Technology & Research (Deemed to be University). Downloaded on January 19,2024 at 06:21:08 UTC from IEEE Xplore. Restrictions a
[6] Cuena, J.; Hernández, J.; Molina, M. Knowledge oriented design of [26] Raspberry Pi 3B+ Specifications. Available online:
an application for real time traffic management: The TRYS system. Eur. https://magpi.raspberrypi.org/articles/raspberry-pi-3bplus-
Conf. Artif. Intell. 1996, 96, 308-312. specsbenchmarks (accessed on 10 March 2021).
[7] Martinez, F.; Manzoni, P.; Garrido, P.; Fogue, M. Automatic [27] Yan, L.; Cengiz, K.; Sharma, A. An improved image processing
Accident Detection: Assistance Through Communication Technologies algorithm for automatic defect inspection in TFT-LCD TCON. Nonlinear
and Vehicles. IEEE Veh. Technol. Mag. 2012, 7, 90-100. Eng. 2021, 10, 293-303.
[8] Fogue, M.; Garrido, P.; Martinez, F.J.; Cano, J.C.; Calafate, C.T.; [28] Zhang, X.; Rane, K.P.; Kakaravada, I.; Shabaz, M. Research on
Manzoni, P.; Sanchez, M. Prototyping an Automatic Notification Scheme vibration monitoring and fault diagnosis of rotating machinery based on
for Traffic Accidents in Vehicular Networks. In Proceedings of the IFIP internet of things technology. Nonlinear Eng. 2021, 10, 245-254.
Wireless Days (WD), Niagara Falls, ON, Canada, 10-12 October 2011; [29] Xie, H.; Wang, Y.; Gao, Z.; Ganthia, B.P.; Truong, C.V. Research
pp. 1-5. on frequency parameter detection of frequency shifted track circuit based
[9] Zhao, Y. Mobile phone location determination and its impact on on nonlinear algorithm. Nonlinear Eng. 2021, 10, 592-599.
intelligent transportation systems. IEEE Trans. Intell. Transp. Syst. 2000, [30] Liu, C.; Lin, M.; Rauf, H.L.; Shareef, S.S. Parameter simulation of
1, 55-64. multidimensional urban landscape design based on nonlinear theory.
[10] Reddy, M.; Tulasi, J. Accident detection depending on the vehicle Nonlinear Eng. 2021, 10, 583-591.
position and vehicle theft tracking, reporting systems. Int. J. Sci. Eng. [31] Kapil K. Jajulwar ; Design of Mobile Robot Navigation System
Technol. Res. 2014, 3, 2359-2362. Using vSLAM and Distributed Filter Techniques.2013 6th International
[11] Tushara, D.B.; Vardhini, P.A. Wireless Vehicle Alert and Collision Conference on Emerging Trends in Engineering and Technology.
Prevention System Design Using Atmel Microcontroller. In Proceedings [32] Kolhe, Pranita, Kamlesh Kalbande, and Atul Deshmukh. "Internet of
of the International Conference on Electrical, Electronics, and Thing and Machine Learning Approach for Agricultural Application: A
Optimization Techniques (ICEEOT), Chennai, India, 3-5 March 2016; Review." In 2022 10th International Conference on Emerging Trends in
pp. 2784-2787. Engineering and Technology-Signal and Information Processing
[12] Shaik, A.; Bowen, N.; Bole, J.; Kunzi, G.; Bruce, D.; Abdelgawad, (ICETET-SIP-22), pp. 1-6. IEEE, 2022.
A.; Yelamarthi, K. Smart Car: An IoT Based Accident Detection System. [33] Kawade, Sharayu Sharad, and Kalyani Akant. "Real Time Image
In Proceedings of the 2018 IEEE Global Conference on Internet of Things Processing System for Crop Segmentation." In 2021 6th International
(GCIoT), Alexandria, Egypt, 5–7 December 2018; pp. 1-7. Conference for Convergence in Technology (I2CT), pp. 1-6. IEEE, 2021.
[13] Rajesh, G.; Benny, A.R.; Harikrishnan, A.; Abraham, J.J.; John, [34] Kumari, Kriti, Pankaj H. Chandankhede, and Abhijit S. Titarmare.
N.P. A Deep Learning based Accident Detection System. In Proceedings "Design of Human Activity Recognition System Using Body Sensor
of the 2020 International Conference on Communication and Signal Networks." In 2021 6th International Conference on Communication and
Processing (ICCSP), Chennai, India, 28–30 July 2020; pp. 1322-1325. Electronics Systems (ICCES), pp. 1011-1016. IEEE, 2021.
[14] Patel, K. Utilizing the Emergence of Android Smartphones for [35] Jajulwar, Kapil K., and A. Y. Deshmukh. "Design of SLAM based
Public Welfare by Providing Advance Accident Detection and Remedy adaptive fuzzy tracking controller for autonomous navigation system." In
by 108 Ambulances. Int. J. Eng. Res. Technol. 2013, 2, 1340–1342. 2016 10th International Conference on Intelligent Systems and Control
[15] Khot, I.; Jadhav, M.; Desai, A.; Bangar, V. Go Safe: Android (ISCO), pp. 1-5. IEEE, 2016.
application for accident detection and notification. Int. Res. J. [36]Kolhe, Mukul, Roshan Paturkar, Umesh Sahu, Sofia Pillai, and
Eng.Technol. 2018, 5, 4118–4122 Abhijit Titarmare. "Analytic for Temperature and Humidity–Cloud based
[16] Nalini, C.; Swapna, R.N. IoT based Vehicle Accident Detection & Forecasting and Dashboard." In 2020 4th International Conference on
Rescue Information System. Eurasian J. Anal. Chem.2018, 13, 911–916 Intelligent Computing and Control Systems (ICICCS), pp. 674-679.
[17] Wang, C.; Dai, Y.; Zhou, W.; Geng, Y. A Vision-Based Video Crash IEEE, 2020
Detection Framework for Mixed Traffic Flow Environment Considering [37] Kshirsagar, Pravin R., Sudhir G. Akojwar, and Nidhi D. Bajaj. "A
Low-Visibility Condition. J. Adv. Transp. 2020, 2020, 9194028. hybridised neural network and optimisation algorithms for prediction and
[18] Bhakat, A.; Chahar, N.; Vijayasherly, V. Vehicle Accident Detection classification of neurological disorders." International Journal of
& Alert System using IoT and Artificial Intelligence. In Proceedings of Biomedical Engineering and Technology 28, no. 4 (2018): 307-321.
the 2021 Asian Conference on Innovation in Technology (ASIANCON), [38] Tadse, Surekha, Muskan Jain, and Pankaj Chandankhede.
Pune, India, 27–29 August 2021; pp. 1–7. "Parkinson's Detection Using Machine Learning." In 2021 5th
[19] Choi, J.G.; Kong, C.W.; Kim, G.; Lim, S. Car crash detection using International Conference on Intelligent Computing and Control Systems
ensemble deep learning and multimodal data from dashboard cameras. (ICICCS), pp. 1081-1085. IEEE, 2021.
Expert Syst. Appl. 2021, 183, 115400. [39] Kosarkar, Nidhi, Pallavi Basuri, Poonam Karamore, Prachi Gawali,
[20] Pour, H.H.; Li, F.; Wegmeth, L.; Trense, C.; Doniec, R.; Grzegorzek, Pradnya Badole, and Pranjali Jumle. "Disease Prediction using Machine
M.; Wismüller, R. A Machine Learning Framework for Automated Learning." In 2022 10th International Conference on Emerging Trends in
Accident Detection Based on Multimodal Sensors. Cars. Sens. 2022, Engineering and Technology-Signal and Information Processing
2022, 1–21. (ICETET-SIP-22), pp. 1-4. IEEE, 2022
[21] Park, E.S.; Fitzpatrick, K.; Das, S.; Avelar, R. Exploration of the [40]Kumari, Kriti, Pankaj H. Chandankhede, and Abhijit S. Titarmare.
relationship among roadway characteristics, operating speed, and crashes "Design of Human Activity Recognition System Using Body Sensor
for city streets using path analysis. Accid. Anal. Prev. 2021, 150, 105896. Networks." In 2021 6th International Conference on Communication and
[22] Singh, G.; Pal, M.; Yadav, Y.; Singla, T. Deep neural network-based Electronics Systems (ICCES), pp. 1011-1016. IEEE, 2021.
predictive modeling of road accidents. Neural Comput. Appl. 2020, 32,
12417–12426.
[23] Detailed Specification FlexiForce A401 Sensor. Available online:
https://www.tekscan.com/products-solutions/force-
sensors/a401?tab=specifications-performance (accessed on 10 March
2021).
[24] NodeMCU ESP8266 Specifications. Available online:
https://components101.com/development-boards/nodemcu-esp8266-
pinout-features-and-datasheet (accessed on 21 April 2021).
[25] Fogue, M.; Garrido, P.; Martinez, F.J.; Cano, J.C.; Calafate, C.T.;
Manzoni, P. A system for automatic notification and severity estimation
of automotive accidents. IEEE Trans. Mob. Comput. 2014, 13, 948–963.

rized licensed use limited to: Vignan's Foundation for Science Technology & Research (Deemed to be University). Downloaded on January 19,2024 at 06:21:08 UTC from IEEE Xplore. Restrictions a

You might also like