A Vision-based System for Traffic Anomaly Detection using Deep Learning and Decision Trees

Any intelligent traffic monitoring system must be able to detect anomalies\nsuch as traffic accidents in real time. In this paper, we propose a\nDecision-Tree - enabled approach powered by Deep Learning for extracting\nanomalies from traffic cameras while accurately estimating the start and end\ntime of the anomalous event. Our approach included creating a detection model,\nfollowed by anomaly detection and analysis. YOLOv5 served as the foundation for\nour detection model. The anomaly detection and analysis step entail traffic\nscene background estimation, road mask extraction, and adaptive thresholding.\nCandidate anomalies were passed through a decision tree to detect and analyze\nfinal anomalies. The proposed approach yielded an F1 score of 0.8571, and an S4\nscore of 0.5686, per the experimental validation.\n