An Intelligent Application for Detecting and Alerting on Dangerous Driving Behaviors

An Intelligent Application for Detecting and Alerting on Dangerous Driving Behaviors

An Intelligent Application for Detecting and Alerting on Dangerous Driving Behaviors

1st Roshan M Thomas

Dept. Computer Science And Engineering Mangalam College Of Engineering Ettumanoor, Kottayam roshan6799thomas@gmail.com

2nd Divya S B

Dept. Computer Science And Engineering Mangalam College Of Engineering Ettumanoor, Kottayam divya.sb@mangalam.in

AbstractWorldwide, traffic accidents are a major source of fatalities and injuries, resulting in financial losses and societal expenses. One of the main factors contributing to traffic accidents is human error, which includes driver weariness, distraction, and other risky driving practises. AI and computer vision develop- ments have created new possibilities for tracking driver behaviour and warning of risky driving practises. Numerous academics have investigated the use of computer vision algorithms to identify and categorise driving behaviour, including detecting attention and sleepiness. These issues have also been addressed by machine learning algorithms, enabling the creation of systems that can recognise and categorise various driving behaviours. Some exist- ing systems use sensors, such as cameras and accelerometers, to monitor driver behavior and provide feedback. However, these systems can be limited in their effectiveness and may require significant calibration and maintenance. The proposed application to monitor driver visual analytic using AI aims to overcome some of these limitations by using advanced computer vision algorithms and machine learning techniques to provide real-time feedback to the driver.

Index TermsArtificial Intelligence, Visual Analytic, Com- puter Vision

1. INTRODUCTION

   Road safety is a pressing concern worldwide, as road accidents are a major cause of fatalities, injuries, and economic losses. One of the primary causes of road accidents is human error, particularly unsafe driving habits such as driver fatigue, distraction, and other forms of impaired driving. According to the WHO, around 1.35 million people die each year globally due to road accidents, with driver fatigue, distraction, and other unsafe driving habits being primary contributing factors. While various technologies exist to monitor driver behavior and provide feedback, such as sensors and cameras, these systems can be limited in their effectiveness and may require significant calibration and maintenance. Advances in computer vision and machine learning techniques have opened up new opportunities to monitor driver behavior and alert them to unsafe driving habits in real time. This study suggests a driver visual analytics program that employs artificial intelligence (AI) to analyze a drivers visual behavior and notify them if they display indicators of fatigue, distraction, or other risky driving practices. A camera that is positioned on the dashboard or steering wheel of the car is used to implement the system. It records the drivers face and eye movements. If the driver

   is found to be drowsy or distracted, the application could alert them with a sound, vibration, or message to encourage them to take a break or refocus their attention on the road. The proposed system has the potential to improve road safety by lowering the number of collisions brought on by driver inattention, weariness, and other risky driving practices. The proposed application also aims to improve the accuracy of existing systems and focus on detecting bad driving behaviors while introducing an alarming system to prevent accidents while driving. The proposed system holds the potential to make a substantial contribution to road safety and decrease the frequency of accidents due to human error by tackling these crucial concerns

2. RELATED WORK

   1. An HMI Concept to Improve Drivers Visual Behavior and Situation Awareness in Automated Vehicle

      This paper introduces a concise HMI (Human Machine Interface) concept that utilizes an LED ambient light located at the bottom of the windscreen to convey critical information such as the status and intention of the automation, detected potential hazards, and warning for a take-over request (TOR) by altering the color, frequency, lighting position, and anima- tion of the LED. The primary aim is to enhance situational awareness (SA) during automated driving, improve the take- over quality, and enable the driver to perform non-driving related tasks (NDRTs) without distraction or annoyance. The effectiveness of the new HMI was evaluated in a between- subject-design experiment using a static driving simulator with 50 participants, who performed a visual motoric task on a smartphone during a 45-minute automated drive with or without the new HMI [1].

   2. Dynamic Human Behaviour Pattern Detection and Classi- fication

      The focus of this study is on analyzing and classifying four types of detailed human behaviors – walking, standing, running, and sitting – through videos. To achieve this, the study proposes the use of a Convolutional Long Short-Term Networks (CLSTM) model that combines the CNN and LSTM models to facilitate the learning, detection, and classification

      of dynamic human behavior patterns. Initially, the visual rep- resentation of human behaviors is learned through AlexNet, a traditional CNN architecture that processes time-based images. These results are then fed into the LSTM model to learn time- based sequence features that can be used for detailed behavior pattern classification. The study also reports a prototype of a Human Behavior Detection System that utilizes the CLSTM model, and some preliminary case study results are presented [2].

   3. Drivers Visual Search Behavior: Eye Tracking Analysis Approach

      This study presents an analysis of how drivers visually process information displayed along Ir H Juanda street Depok, and how eye-tracking techniques can be used to evaluate the comprehension of visual displays presented in video format. The study utilized fixation and attention maps to report on the drivers attention during the experiment. The research findings highlight the potential of eye movement recordings as a valuable tool for gaining insights into drivers attention while acquiring visual information [3].

   4. Effects of Vehicle Simulation Visual Fidelity on Assessing Driver Performance and Behavior

      In this study, we developed two immersive virtual envi- ronments – a low graphic fidelity driving simulation that is representative of most current research simulation testbeds and a high graphic fidelity environment created using Unreal En- gine, which represents state-of-the-art graphical presentation. A user study was conducted with 24 participants who were required to navigate a route in a virtual urban environment, using AR graphical cues for direction and monitoring the road scene for pedestrian hazards. Driving performance, gaze patterns, and subjective feedback were recorded via situational awareness survey (SART) [4].

   5. XGBoost Algorithm-Based Monitoring Model for Urban Driving Stress: Combining Driving Behavior, Driving Envi- ronment, and Route Familiarity

   The focus of this study is to monitor a drivers driving stress levels by developing a model based on driving behavior, driving environment, and route familiarity. To extract data for the model, a real driving task was designed. The driving behavior is described by measuring the speed and acceleration of the vehicle, while the driving environment is quantified using a dilated residual networks (DRN) model. This model divides th video image from the full region into sub-regions based on the distribution of the drivers attention [5].

3. EXISTING METHODOLOGY

   In this study, a real driving task was designed to extract data, and a drivers driving stress monitoring model was proposed based on driving behavior, driving environment, and route familiarity [6]. The speed and acceleration of the vehicle describe the driving behavior, while the driving environment is quantified using a dilated residual networks (DRN) model.

   Fig. 1. Different layers of XGBoost Method

   This model divides the video image from the full region into sub-regions based on the distribution of the drivers attention. To improve the models accuracy, a hierarchical method based on weighted extreme gradient boosting was used [7]. This method continuously adds and trains new trees in each iteration to fit the residuals of the predicted values of the previous decision tree and the sum of the predicted values of all previous decision trees. Finally, the predicted values of all decision trees are summed together to obtain the final result.

4. PROPOSED METHODOLOGY

   Fig.2 details the proposed architecture for the application to monitor driver visual analytics using AI based on CV and ML techniques. The implementation of the system involves utilizing a camera that is mounted on either the dashboard or steering wheel of the vehicle. The camera is responsible for capturing the movements of the drivers face and eyes. The camera is connected to a processing unit, such as a microcontroller or a dedicated processing board, which runs the computer vision algorithms and machine learning models. The CV algorithms analyze the drivers eye movements and facial expressions to detect signs of fatigue, distraction, or other unsafe driving habits. The ML models leverage this data to categorize the behavior of the driver and ascertain the necessary course of action to be taken. For example, if the driver is found to be drowsy or distracted, the application could alert them with a sound, vibration, or message to encourage them to take a break or refocus their attention on the road. The system can also be integrated with other sensors, such as accelerometers and GPS, to provide additional data on the vehicles speed, acceleration, and location. Utilizing this data can enhance the precision of the system and offer addi- tional background information regarding the drivers actions. The proposed architecture also includes a user interface that allows the driver to configure the system settings and view feedback on their driving behavior. The user interface could be implemented using a mobile application or a dashboard display, depending on the vehicles configuration. Overall, the proposed architecture for the application to monitor driver visual analytics using AI is designed to be accurate, reliable, and easy to use. There is the possibility that it could enhance road safety and lessen the occurrences of accidents caused by human error, to a great extent.

   The system can also be integrated with other sensors, such as accelerometers and GPS, to provide additional data on the vehicles speed, acceleration, and location. This data can be used to further improve the accuracy of the system and provide more context for the drivers behavior.

   Fig. 2. Different Steps in the Architecture

   The proposed architecture also includes a user interface that allows the driver to configure the system settings and view feedback on their driving behavior. The user interface could be implemented using a mobile application or a dashboard display, depending on the vehicles configuration.

   Overall, the proposed architecture for the application to monitor driver visual analytics using AI is designed to be accurate, reliable, and easy to use, with the potential to significantly improve road safety and reduce the number of accidents caused by human error.

5. CONCLUSION

   In conclusion, the development of an AI-based application for monitoring driver visual analytics is a promising area of research with the potential to enhance road safety and decline accidents. The proposed architecture, which includes a camera, processing unit, computer vision algorithms, machine learning models, alarming system, and user interface, offers a robust and flexible approach to detecting and responding to unsafe driving behaviors.

   While XGBoost is a popular machine learning algorithm that has been successfully applied in many domains, including driving behavior monitoring, it is not without its limitations. The algorithms reliance on large amounts of data, potential for overfitting, limited interpretability, high computational requirements, and need for careful feature engineering may make it less suitable for some driving behavior monitoring applications.

   Overall, selecting the most appropriate machine learning algorithm and approach for monitoring driver behavior re- quires careful consideration of the specific requirements and constraints of the application. With further research and development, AI-based driving behavior monitoring systems hold great promise for improving road safety and reducing accidents.

   ACKNOWLEDGMENT

   We would like to express our deepest gratitude to all those who have contributed to the completion of this paper. Our sin- cere thanks go to the authors of the relevant literature, whose insightful contributions have been invaluable in shaping our research. We also extend our appreciation to the reviewers and editors for their constructive comments and suggestions, which have helped us to improve the quality of the paper. We would like to acknowledge the support of our institution, which has provided us with the necessary resources and facilities to conduct this research. Finally, we are grateful to all those who have supported us throughout this endeavor, including our colleagues, friends, and families, whose encouragement and assistance have been instrumental in our success.

   REFERENCES

   [1] Yucheng Yang, Burak Karakaya, Giancarlo Caccia Dominioni, Kyosuke Kawabe, and Klaus Bengler An HMI Concept to Improve Drivers Visual Behavior and Situation Awareness in Automated Vehicle, 21st International Conference on Intelligent Transportation Systems (ITSC) Maui, Hawaii, USA, November 4-7, 2018.
   [2] Shuqin Wang , Jerry Zeyu Gao, Hanping Lin, and Mayur Shitole, Dynamic Human Behavior Pattern Detection and Classification, 2019 IEEE Fifth International Conference on Big Data Computing Service and Applications (BigDataService).
   [3] Dian Kemala Putri, Karmilasari and Mohammad Iqbal, DriversVisual Search Behaviour: Eye Tracking Analysis Approach, IEEE 2020, vol.91.
   [4] Coleman Merenda, Chihiro Suga, Joseph Gabbard, and Teruhisa Misu, Effects of Vehicle Simulation Visual Fidelity on Assessing Driver Performance and Behavior, IEEE Intelligent Vehicles Symposium (IV) Paris, France. June 9-12, 2019.
   [5] YUE LU, XINSHA FU, ENQIANG GUO, and FENG TANG, XGBoost

   Algorithm-Based Monitoring Model for Urban Driving Stress: Combin- ing Driving Behaviour, Driving Environment, and Route Familiarity, IEEE Access VOLUME 9, 2021.

   [6] H. Rahman, M. U. Ahmed, S. Barua, and S. Begum, Non-contactbased drivers cognitive load classification using physiological and vehicular parameters, Biomed. Signal Process. Control, vol. 55, Jan. 2020,Art. no. 101634, doi: 10.1016/j.bspc.2019.101634.
   [7] G. Matthews, A. Tsuda, G. Xin, and Y. Ozeki, Individual differences in driver stress vulnerability in a Japanese sample, Ergonomics, vol. 42,no. 3, pp. 401-415, Mar. 1999, doi: 10.1080/001401399185559.
   [8] M. Elgendi and C. Menon, Machine learning ranks ECG as an optimal wearable biosignal for assessing driving stress, IEEE Access, vol. 8, pp. 3436234374, 2020, doi: 10.1109/ACCESS.2020.2974933.

6. FUTURE WORKS

There are several potential areas for future research in the development of an AI-based application for monitoring driver visual analytics. These inclde incorporating additional data sources, exploring alternative machine learning algorithms, developing more advanced computer vision algorithms, imple- menting real-time feedback and intervention, and conducting field studies and user testing. By addressing these areas, researchers and developers can further improve the effective- ness and usability of AI-based driving behavior monitoring systems, ultimately contributing to improved road safety and reduced accidents.