A lightweight small-scale pedestrian and non-motorised vehicle target detection algorithm based on improved YOLOv8
To solve the problem of pedestrian and non-motorized vehicle governance at traffic intersections,a small and light-weight small-scale pedestrian and non-motorized vehicle detection algorithm based on improved YOLOv8,ACM-YOLO is pro-posed,for the small size of pedestrian and non-motorized vehicle targets at panoramic intersection monitoring.Firstly,a lightweight and efficient AFPN feature fusion network is proposed to replace the PAFPN algorithm in the original network to improve the recog-nition effect of small targets;secondly,CWPConv channel weight partial convolution is proposed based on PConv,and CWPC2f is further proposed to effectively reduce the number of model parameters and calculation;finally,the MPDIoU function is used to opti-mize the boundary box loss of the network.The results show that compared with YOLOv8m,the mAP50 is increased by 2.2%,8.1%and 3%on the self-built dataset,VisDrone2019 dataset and CityPerson dataset respectively,the number of parameters is re-duced by 17%and the GFLOPs is reduced by 2.
small target detectionYOLOv8pedestrians and non-motor vehicles detectionCWPConv
万方数据
