| Abstract |
Abnormal drivers are individuals who drive above the speed limit, change speed suddenly, or change vehicle lateral position incessantly. Monitoring these abnormal driving behaviours in real-time is very crucial to improving driving safety. To improve drivers’ awareness of their driving habits, and prevent potential car accidents, the need to consider a fine-grained monitoring approach is a must, which not only detects abnormal driving behaviours but also identifies specific types of abnormal driving behaviours, i.e. weaving, swerving, side slipping, fast U-turn, Turning with a wide radius and sudden braking. All six types of driving behaviours have their unique patterns on inertial acceleration. By studying the driving profile, necessary features to capture the abnormal patterns in driving behaviours are identified, such as acceleration and rotation along X, Y and Z directions, position of Gas Pedal, Light Status and Indicators (turn-light signal). Using these observations, machine-learning models are trained to perform real-time high-accuracy driving behaviour monitoring. Abnormal driving errors centre on human mindsets; as such a national ranking system ‘DriveScore’ has been incorporated that provides tools and statistics for self-analysis, and the use of ‘Reward and Reinforcement’ approach for encouraging good driving behaviour. Two different datasets—Mendeley and Virginia were individually used to determine deviant driving patterns and conclude whether/not a car will crash, models trained achieved an accuracy of 90% and 98%, respectively. This allows us to predict whether the driving is safe or unsafe, and the probability of a crash. This paper presents an intelligent monitoring and ranking system for all kinds of vehicles. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG. |
