Dynamic Environmental Target Detection and Ranging Based on Improved Yolov5s

In order to improve the detection effect of vehicles in dynamic environment and further obtain accurate distance information, this paper proposes an improved target vehicle detection algorithm with YOLOv5s and ranging of the vehicle in front with binocular depth camera. Based on the YOLOv5s detection network, firstly, the CA attention module is introduced into the Backbone network to effectively extract the contour features of the detection target, secondly, the PANet network in Neck is replaced with BiFPN to improve the fusion ability of the features, and then the Head detection layer is replaced with the decoupled head method to accelerate the convergence speed of the network, and then the loss function is optimized using the SIoU to enhance the accuracy of the vehicle detection accuracy. The improved YOLOv5s network model is used to constrain the feature matching points to obtain the optimal parallax value, and finally the distance information of the front vehicle is obtained through the principle of binocular vision ranging. Tests show that the vehicle recognition rate accuracy is 79.50% with an improvement of 1.50% and the average error rate of ranging is 2.552% in a distance range of 14m.