ABSTRACT
In developing countries like India, the number of both vehicles and accidents is increasing day by day. The main cause of the annual growth of the accident rate is the lack of appropriate surveillance. In heavy traffic zones such as junctions, tracking vehicles in real time might reduce the death rate from accidents. To overcome the low accuracy of current traffic accident prediction systems, we need a system that analyzes camera footage at selected junctions, detecting and tracking vehicles entering and leaving the junction using DeepSORT (simple online and real-time tracking), accurately establishing their speeds frame by frame, and assigning a fixed ROI (region of interest) to each vehicle. The system, an improvised version of current prediction architectures, is capable of tracking vehicles and predicting the chances of an accident if they exceed the threshold.
