Advanced Traffic Object Detection in Complex Road Conditions Using an Optimized YOLOv8 Algorithm

To tackle the challenges of false positives and missed detections caused by dense occlusions and small objects in complex road settings, this paper presents an enhanced YOLOv8-based model named YOLO-RC. The backbone network is strengthened using MBConv modules to improve feature extraction and enhance localization for densely occluded objects. A new C3FB structure (Efficient Fusion of C3 modules and FocalNextBlock) replaces the C2f module in the neck network, reducing parameter count and increasing accuracy. Experimental results on the DAIR-V2X-I dataset demonstrate that YOLO-RC achieves an mAP50 of 91.1%–a 4.5% improvement over the original model–with a recall rate of 81.8%. Comparative analysis against mainstream detection algorithms confirms the superiority of YOLO-RC in terms of accuracy and robustness.