AutoTrack: A Deep Learning Approach on a Novel Dataset for Vehicle Detection and Route Tracking from CCTV Images

Hit-and-run incidents remain a critical public safety challenge, with thousands of fatalities reported annually. In Bangladesh alone, 7,902 road accident deaths occurred in 2023, with motorcycles involved in 32.4% of cases. Traditional investigative approaches, such as manual CCTV analysis, are inefficient due to poor camera quality, sparse coverage, and human error, leading to significant delays in offender identification. This paper presents an automated system integrating YOLOv5 for vehicle color-based detection and YOLOv8 for license plate recognition to enhance hit-and-run investigations. The proposed system reconstructs vehicle paths using timestamped footage, resolves identification ambiguities, and provides real-time geospatial mapping for law enforcement intervention. Preliminary evaluations demonstrate an 83% accuracy rate in tracking vehicles across multiple surveillance feeds, reducing investigation time from hours to minutes. While challenges such as infrastructure costs and inconsistent CCTV coverage persist, our approach offers a scalable solution for improving road safety and justice delivery. By automating vehicle tracking and identification, this system strengthens law enforcement capabilities, increasing accountability and reducing hit-and-run incidents.