The improvement of YOLOv5s algorithm for real-time detection of vehicle targets

In response to the existing target detection algorithm in urban road vehicle detection, there are problems such as low detection accuracy, poor real-time and robustness, etc., and propose a method of vehicle target detection to improve YOLOv5s. First of all, optimize the YOLOv5s algorithm network structure, and use increasing the detection layer of small targets to improve the accuracy of the detection of small targets; then use CIOU instead of Giou as the border frame regression loss function, solving the border frame that contains each other to compare with length and width. Question; then use the Hardswish activation function on the entire network model for convolution operations. In order to verify the effectiveness of the algorithm in this article, we will test the improved algorithm on the UA-DETRAC dataset. The experimental results show that the accuracy of target detection (Precision) increased by 2.46%, the recall rate (Recall) increased by 5.5%, and the average accuracy (mAP) increased by 2.19% Vehicle goals.