Abstract:To address the issue of missed and false detections for distant small objects and occluded objects in current road target detection algorithms in autonomous driving scenarios, a road target detection algorithm based on an improved YOLOv8n is proposed. In terms of feature extraction, the Receptive-Field Attention Convolution is lightweightly improved, and the C2f module is reconstructed to solve the problem of non-shared parameters in convolution calculations, enabling the network to effectively capture critical information. Then, a lightweight point sampling operator is introduced to reduce the loss of feature details during the upsampling process, better preserving image detail information. In terms of feature fusion, a multi-scale feature fusion network is designed to enhance small target feature information and enrich the bidirectional fusion of features at different scales. Simultaneously, a normalization attention mechanism is used to suppress irrelevant background information interference, improving the model′s anti-interference capability. Experimental results show that the proposed improved algorithm achieves detection accuracies of 92.6% and 78.7% on the KITTI dataset and the Udacity dataset, respectively, representing improvements of 2.1% and 1.6% compared to the original algorithm. The model still meets lightweight requirements and enhances adaptability to complex road scenes to a certain extent.