重庆理工大学学报2026，Vol.40Issue(1)：62-69,8.DOI:10.3969/j.issn.1674-8425(z).2026.01.008
减少道路行人重叠漏检的联合检测模型研究
Research on a joint detection model to reduce overlapping and missed detection of road pedestrians
摘要
Abstract
To address overlapping and missed detections of pedestrians in the autonomous driving scenarios,this paper proposes a joint detection model,improving the accuracy and recall of pedestrian detection.First,the YOLOv8 algorithm is enhanced,and the small-object detection layer is replaced to detect the human heads and entire bodies in the image.Next,an Adaptive Weighted Distance Non-Maximum Suppression(AWD-NMS)method is proposed,which fully considers the detection confidence and the Euclidean distance between center points to remedy the missed detection of non-maximum suppression(NMS)when dealing with high overlap targets.Finally,a head-to-body matching strategy is introduced to analyze the spatial relationship between the detected head bounding box and body bounding box.Experimental results demonstrate,on the Wider Person dataset,the proposed joint detection model improves the average precision and recall by 2.2%and 2.3%respectively,compared with YOLOv8.Real-vehicle experiments further prove the proposed model effectively reduces the occurrence of overlapping and missed detections of pedestrians on the road.关键词
行人检测/小目标检测层/非极大值抑制/联合检测模型Key words
pedestrian detection/small target detection layer/non-maximal inhibition/joint detection model分类
交通工程引用本文复制引用
刘明伯,李刚,张东,王希岳..减少道路行人重叠漏检的联合检测模型研究[J].重庆理工大学学报,2026,40(1):62-69,8.基金项目
中国高校产学研创新基金项目(2024HT007) （2024HT007）
辽宁省教育厅重点攻关项目(JYTZD2023081) （JYTZD2023081）
辽宁省高等学校国(境)外培养项目(2018LNGXGJWPY-YB014) （境）
