Traffic Flow Statistics Algorithm Based on Improved YOLOv5

Aiming at the problems of poor real-time and low accuracy of traffic flow statistics, this paper proposes an end-to-end traffic flow statistics algorithm based on improved YOLOv5. Firstly, in order to improve the detection effect of YOLOv5, the regional context module is proposed to capture the global information. Secondly, alpha-CIoU Loss is used as the localization loss to improve the bounding box regression accuracy. Finally, we connect the improved YOLOv5 detection algorithm with Byte data association method and set virtual coils in the video frames to complete the traffic flow statistics. The experimental results show that the improved YOLOv5 algorithm on the UA-DETRAC dataset improves the average accuracy by 3.8% compared with the original YOLOv5 algorithm. Combined with the Byte data association method, the traffic flow statistics accuracy reaches 99.2% and the speed reaches 73frame/s, which can effectively improve the accuracy while meeting the real-time requirements.