A Comparative Study of YOLOv8 and YOLOv5 in Traffic Environments

In the Intelligent Transportation System (ITS), high - precision real - time vehicle detection is crucial for ensuring traffic efficiency and safety. Traditional methods have problems such as high false - detection rates and poor adaptability in complex scenarios. This research focuses on the technological evolution of the YOLO series of models and systematically compares the performance differences between YOLOv8 and YOLOv5 in traffic vehicle detection. Through experimental verification using the KITTI dataset and public data, the results show that YOLOv8 has improvements of 4.7%, 3.7%, and 4.1% in mean average precision ([email protected]), recall rate, and inference speed respectively compared to YOLOv5. Its improved non -maximum suppression (NMS), dynamic anchor box generation, and task - aligned loss function effectively reduce the rates of missed and false detections, making it more suitable for traffic scenarios with high real-time requirements.