An Enhanced YOLOv11-based Algorithm for Vehicle and Pedestrian Detection in Complex Traffic Scenarios

With the rapid development of intelligent transport ation systems and autonomous driving technology, accurate and efficient vehicle and pedestrian detection has become an important research direction in the field of computer vision. This paper pro poses an improved detection model called GFL-YOLO to address the challenges of multi-scale targets and complex environments i n object detection for complex traffic scenarios. The model prima rily includes: proposing a GEIT architecture, which effectively ex tracts global edge information through two key modules, MSEIG and CEF; designing a FDPN-DASI architecture, which utilizes a dual attention mechanism to interact and fuse features at differen t scales; proposing a LSDECD detection head, which enriches the diversity of feature representation through shared convolution an d detail enhancement convolution modules. Experimental results on the KITTI dataset show that GFL-YOLO improves recall rate, mAP50, and mAP50-95 by 3.30%,2.7%, and 4% respectively compared to YOLOv11n; ablation experiments further verify the positive contribution of each module to model performance; visualization analysis demonstrates that GFL-YOLO has superior detectio n capabilities for distant targets, occluded targets, dense pedestria ns, and multi-vehicle scenarios, providing an efficient and accurat e solution for object detection in complex traffic scenes.