Challenges in Real Time Vehicular Accident Detection using YOLOv8 and Trajectory Analysis

Urban traffic surveillance struggles with accident detection due to occlusion, vehicle overlap, camera distortion, and lighting changes. Manual monitoring causes fatigue and delays, while existing automated methods rely on simplistic indicators, yielding high false positives, missed events, and excessive computational load unsuitable for edge devices. We propose a lightweight, real time vehicular accident detection system using YOLOv8 for detection and tracking, followed by spatiotemporal trajectory, velocity anomaly, and collision signature analysis to differentiate true accidents from normal traffic. Optimized for roadside deployment, it ensures low latency inference and reduced false alarms via heuristic post processing. Benchmark evaluation shows 94–96% precision, improved recall and F1 score over prior methods, plus explainable trajectory and confidence visualizations advancing reliable, scalable intelligent transportation for road safety.