An Improved Yolov8n-Based Road Traffic Object Detection Algorithm

In the current complex road traffic environment, the interaction between vehicles and pedestrians is intricate.Therefore, it has become crucial for autonomous driving systems to detect road conditions promptly and accurately. An improved YOLOv8n model is proposed in this paper to identify vehicles, pedestrians in the distance, and cyclists on the road. Firstly, FasterNet is used to replace the backbone network of YOLOv8n to cut down redundant computations and improve the accuracy and speed of detection. Secondly, the C2f module of YOLOv8n is replaced with the RCSOSA module to enhance the detection accuracy of small targets, such as pedestrians and cyclists. Finally, the CARAFE upsampling module is adopted to improve the detection accuracy of occluded targets. Ablation experiments are carried out on the KITTI dataset, and the experimental results show that each of the proposed improved modules is effective. The ultimately improved YOLOv8n network has achieved a certain improvement in the detection accuracy of small targets and occluded targets on traffic roads.