Automatic Vehicle Accident Detection and Classification from Images: A Comparison of YOLOv9 and YOLO-NAS Algorithms

Vehicle accidents have profound implications for both individuals and traffic systems. They not only endanger lives but also disrupt traffic flow, causing delays and economic losses. Effective strategies for accident prevention and management are essential to mitigate these impacts and ensure the safety and efficiency of roadways. To detect and place a quick response, it is important to monitor traffic and detect vehicle accidents by using cameras. In this study, we propose accident detection methods by using artificial intelligence models to improve vehicle accident detection efficiency. It utilized the YOLOv9 and YOLO-NAS algorithms for this task, and a comprehensive evaluation and performance comparison were employed. Both models perform real-time object detection and have been adapted for this specific application. It was observed that the YOLO-NAS-L model showed good performance with a mAP.50 of 85% and a mAP.50-.95 of 70.1%. The YOLOv9-C model, on the other hand, performed better in terms of accuracy, with a mAP.50 score of 92.7% and a mAP.50-.95 score of 86%. Hence, both proposed methods are well-suited to be applied for real-time accident detection to contribute to quick post-accident recovery and save lives.