Lightweight Vehicle and Pedestrian Fog Detection Model Based on Improved YOLOv5
Vehicle and pedestrian detection is an important part of intelligent transportation.Aiming at the problem that the existing vehicle detection algorithm model has large capacity,large number of parameters,and large memory occupation,which is difficult to be applied to edge devices with limited computing power and memory in intelligent transportation scenarios,thus an improved lightweight object detection network based on YOLOv5s algorithm has been proposed.Firstly,the convolution module of YOLOv5s network is replaced with Ghost convolution to reduce the amount of computation and parameters.Secondly,the improved weighted bidirectional Feature Pyramid Network(BiFPN)structure and non-maximum suppression(NMS)is adopted to improve the accuracy of the model.Lastly,the model is trained and verified via Real-world Task-Driven Testing Set(RTTS)foggy data set to test the effectiveness of the model.The experimental results show that the lightweight fog detection model of the improved YOLOv5 has an average detection accuracy of 88.5%on the image with a resolution of 640×640,whose size is about 7.5 M,and the floating-point calculation amount is 8.20 GFLOPs.Compared with the original YOLOv5s network,the size of the improved model is reduced by 46.4%,the floating-point computation is compressed to 52%,the accuracy is improved by 0.9%,the regression rate is increased by 0.5%,and the average accuracy is improved by 1.1%.The improved vehicle detection algorithm not only ensures high detection accuracy while being lightweight in the model,but also meets the needs of vehicle detection in edge devices with limited computing resources.
国家科技期刊平台
NSTL
万方数据
