To address the issues of low detection accuracy, false detection, and missing detection, as well as the challenge of modeling lightweight scenes caused by the overlapping occlusion of roadside targets and distant targets in autonomous driving scenarios, an improved small target detection algorithm for autonomous driving based on YOLO11 is proposed. Firstly, it embedded the Channel Transposed Attention in the C3k2 module, proposed the C3CTA module, and replaced the C3k2 module in the Backbone network to improve the feature extraction ability and strengthen the detection ability in the case of target occlusion. Secondly, the Diffusion Focusing Pyramid Network is introduced to improve the Neck part, enhance the understanding ability of small targets in complex scenes, and effectively solve the problem that it is difficult to extract vehicle target features. Finally, a Lightweight Shared Convolutional Detection Head is introduced to reduce the number of model parameters and achieve lightweight requirements. The experimental results show that the Precision, Recall, mAP@0.5, and mAP@0.5–95 of the improved algorithm on the world's first DAIR-V2X-I dataset reach 85.7%, 79.4%, 85.3%, and 61.3%, which is 4.0% higher than the baseline model. 3.1%, 2.4%, 2.8%. Higher detection accuracy is achieved, which proves the effectiveness of the improvement. Proposed a new solution approach for target detection in complex autonomous driving scenarios.
Liang et al. (Tue,) studied this question.