AI-Assisted Hit-and-Run Detection and Tracking

With the expansion of the Indian road network, hit-and-run incidents account for a significant portion of the country’s rising road fatalities. This study proposes an AI-based collision detection and vehicle tracking system designed to address this issue by utilizing CCTV cameras. The proposed approach includes image preprocessing techniques such as background subtraction, followed by deep-learning-based feature extraction models such as Vision Transformers (ViT) and a transfer learning approach using MobileNetV2. The tracking methods include Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), OSNet, and ResNet. The system achieved a validation accuracy of 91.84% for detection using ViT and 86.78% using the tracking methods. This research lays the groundwork for developing a system that enables seamless integration with law enforcement agencies, allowing for rapid identification of vehicles involved in accidents and tracking the offenders.