Automobile Accident Detection and Automated Emergency Alert System Using YOLOv8, ByteTrack, and Deep Learning

Abstract

This paper presents a real-time automobile accident detection and emergency-alert framework for CCTV-based road surveillance. The system uses YOLOv8 to detect vehicles and produce bounding boxes, class labels, and confidence scores. ByteTrack maintains vehicle identities across frames, allowing the computation of center-point trajectories, displacement, velocity, acceleration, relative motion, bounding-box overlap, and time-to-collision. A multi-parameter accident decision module combines proximity, intersection-over-union, relative velocity, sudden motion change, and multi-frame persistence to reduce single-frame false alarms. ConvLSTM-based temporal risk analysis can be incorporated to learn accident-related motion evolution. After a collision is confirmed, the system generates a structured alert containing the timestamp, location, vehicles involved, confidence, severity, and required emergency services. The supplied implementation provides a Flask-based real-time dashboard, local network access, evidence-file management, automatic refresh, incident statistics, and formatted accident alerts. Experimental screenshots demonstrate successful generation of high-severity alerts with 95% confidence, dashboard logging of 14 incidents and 68 evidence recordings, and persistent web-based monitoring. The framework converts passive CCTV infrastructure into an active road-safety system capable of accelerating emergency notification and supporting intelligent transportation and smart-city applications