计算机技术与发展2026，Vol.36Issue(6)：77-84,8.DOI:10.20165/j.cnki.ISSN1673-629X.2025.0345
基于改进RT-DETR模型的道路减速带检测算法
Road Speed Bump Detection Algorithm Based on an Improved RT-DETR Model
摘要
Abstract
In complex urban road environments,uneven road structures such as speed bumps pose potential safety threats to autonomous vehicles.Inaccurate or delayed detection not only affects ride comfort but may also cause vehicle damage or safety incidents.Therefore,the robustness of perception systems under varying illumination,occlusion,and texture conditions is crucial for decision-making and control accuracy in autonomous driving.To this end,we propose an improved algorithm,SDFP-DETR,based on the RT-DETR model to enhance the performance and efficiency of speed bump detection.Specifically,SPCA attention module is integrated into the backbone to strengthen feature representation in key regions and improve detection stability in complex scenes.In the neck network,an AIFI-DHSA module based on a dual-head self-attention mechanism is designed to enhance multi-scale semantic feature modeling.Furthermore,a Focaler-PIoU loss function is adopted to improve gradient response to low-quality samples and accelerate model convergence.Experimental results on a self-built complex road scenario dataset show that the proposed model achieves 5.2 percentage points increase in precision,2.0 percentage points increase in recall,and3.3 percentage points improvement in mAP50.Compared with existing methods,the proposed approach maintains high detection performance under challenging illumination and occlusion conditions,demonstrating excellent environmental adaptability and application potential.关键词
减速带/自动驾驶/SDFP-DETR/SPCA模块/DHSAKey words
speed bump/autonomous driving/SDFP-DETR/SPCA module/DHSA分类
信息技术与安全科学引用本文复制引用
杨智勇,张佳斌,许沁欣..基于改进RT-DETR模型的道路减速带检测算法[J].计算机技术与发展,2026,36(6):77-84,8.基金项目
重庆市自然科学基金创新发展联合基金(CSTB2025NSCQ-LZX0125) （CSTB2025NSCQ-LZX0125）
重庆市教育委员会科学技术研究计划项目(KJZD-M202303401,KJQN202403437) （KJZD-M202303401,KJQN202403437）
