Enhanced Accident Detection System within Smart Cities Using Deep Learning Models

Accidents are a major cause of traffic accidents, which are increasing every day. Sharing information about the state of roads with nearby vehicles could reduce secondary accidents that result from ignorance of the road conditions ahead. The objectives of this study are to propose an architectural VANET system that can solve the problem of avoiding road traffic collisions. We propose an architecture VANET for accident detection using deep learning methods within this context. Consequently, we evaluated the performance of Mask R-CNN, SOLO, YOLOv6 and YOLOv7 deep learning models in order to reach this goal. YOLOv7, the model selected from this paper, achieved state-of-the-art results with 97.5% in real-time.