Research on Vehicle Pedestrian Detection Based on Improved YOLOv8

To address the issues of vehicle pedestrian occlusion, missed detections, and false detections in traffic scene detection, an improved vehicle and pedestrian detection algorithm, YOLOv8_R, is proposed. Firstly, integrating the GAM attention module reduces feature information loss and enhances the focus on small object features. Secondly, the BiFPN is introduced to replace FPN and PAN, enabling adaptive fusion of multi-scale features. Thirdly, the loss function is optimized by replacing CIOU loss function with the Wise-IoU loss function, which increases the weight of difficult-to-distinguish samples. Finally, the dynamic detection head is used to improve the accuracy of detection while adapting to different detection tasks The results show that YOLOv8_R obtains 93.2% mAP on KITTI data set, which is 2.1 % higher than the original model, improving the accuracy of vehicle pedestrian detection, and alleviating the problems of occlusion and false detection to a certain extent.