Development of a Driver Safety Reward System Incorporating You Only Look Once (YOLO)-Based Traffic Violation Detection at Roundabouts under Highly Heterogeneous Traffic

Abstract

This study introduced a novel driver safety reward system (DSRS) leveraging You Only Look Once (YOLO)-based traffic violation detection and an incentive mechanism to promote safer driving behaviors at roundabouts. Unmanned aerial vehicle (UAV)-captured aerial videos at unsignalized roundabouts were used for detecting vehicles and extracting trajectory data to analyze various traffic violations such as wrong-way driving, anti-clockwise driving, and improper U-turns. The YOLOv8m model, coupled with ByteTrack, facilitated robust vehicle tracking using high-quality aerial video data. Driver behavior was scored based on detected violations, overspeed events, and sharp acceleration and deceleration instances. Additionally, a MERN stack-based reward system was developed to assign and visualize reward points for drivers. The system provided a clear and accessible way for drivers to understand their driving performance. Furthermore, a multinomial logistic regression (MLR) model was developed to predict violation types, demonstrating good prediction accuracy. This study highlighted the potential of cutting-edge image processing technologies to enhance road safety by identifying problematic traffic scenarios and developing a proactive driver reward disbursement system. The system provided valuable insights into violation patterns predominantly occurring at roundabouts under mixed traffic conditions for different vehicle types. The research, uniquely combining aerial surveillance, machine learning, and a user-friendly reward platform, aimed to reduce traffic violations and promote safer driving practices significantly. This innovative approach contributed to the broader goal of enhancing road safety through technological innovation.