Efficient Object Detection and Depth Estimation Algorithms for Transportation Scenarios

Object detection and depth estimation are fundamental tasks for autonomous driving systems. In traffic scenarios, existing object detectors often exhibit low recall rates when processing distant targets. Furthermore, the inherent depth ambiguity of monocular vision limits its application in three-dimensional spatial perception tasks. To address these issues, this paper proposes a collaborative perception framework that integrates improved YOLO11 detection (YOLO-DM) with depth estimation. The YOLO-DM network builds upon YOLO11 by introducing a novel C2PSA-MSDA module to expand the receptive field and enhance object features, while integrating a Dynamic Detection Head (DyHead) to improve scale robustness. Furthermore, the YOLO-DM network is combined with the PSMNet stereo matching network to achieve efficient acquisition of target depth information. Evaluation on the KITTI dataset demonstrates that compared to the baseline, YOLO-DM improves mAP@0.5 by 2.2% and mAP@[0.5:0.95] by 2.7%, while reducing computational complexity and parameter count by 4.9% and 6.7%, respectively. The proposed framework significantly enhances perception accuracy while maintaining high model efficiency, offering a highly promising perception solution for autonomous driving systems.