Near Miss Detection Using Distancing Monitoring and Distance-Based Proximal Indicators

Despite efforts to improve road safety, accidents persist due to insufficient evidence from manual police reporting, non-optimized detection algorithms, and technical limitations in real-time video processing and modelling. This study focuses on detecting and tracking vehicles within a monitoring system and analyzing near-miss incidents (black spot and unseen area), specifically examining the influence of video quality on detection performance using advanced model detectors (YOLOv4-tiny, YOLOv5, YOLOv7, and YOLOv7+CNeB). The experiment employed methods for vehicle detection through the monitoring system. Near-miss detection was conducted using two approaches: manual observation (Social Distancing Monitoring and Bird's Eye View) and automatic calculation (using DN indicators). Statistical methods, including descriptive statistics, and one-way ANOVA, were applied to compare datasets obtained from these indicators. The study concludes that YOLOv7+CNeB is effective for vehicle detection and near-miss analysis when video quality is considered in system design and implementation. YOLOv7+CNeB significantly reduces the time required to collect evidence from specific roads, provides visual reports, and addresses technical limitations in current algorithms. Future research should explore additional factors contributing to near-miss events, such as road environment, lane changes, and driver behaviours.