汽车工程学报2025，Vol.15Issue(5)：699-706,8.DOI:10.3969/j.issn.2095-1469.2025.05.06
基于路侧视觉感知的交通目标检测及跟踪方法研究
A Method for Traffic Target Detection and Tracking Using Roadside Visual Perception
摘要
Abstract
Visual inspection is an important technology for roadside perception in vehicle-road cooperation.Conventional vision algorithms struggle to balance detection accuracy and computational efficiency.To address this issue,the paper proposes a new visual processing method based on YOLOX and KPP-DeepSORT.First,YOLOX performs target recognition on multiple video streams.KPP-DeepSORT then tracks the bounding boxes of those detected targets in each video.In the tracing algorithm,K-Means++is introduced for improving DeepSORT.Considering the orderly motion of vehicles at intersections,it clusters all targets and applies DeepSORT within each cluster.The above process greatly reduces the probability of the target ID reassigment in DeepSORT cascade matching,and shortens the overall computing time of multi-target tracking.Results show that the proposed method accurately detects and tracks the pedestrians and vehicles at intersections.Especially during rush-hour traffic,it is notably more efficient than several common algorithms.These results suggest its strong potential in internet-of-vehicles applications.关键词
车路协同/路侧感知/图像识别/YOLOX/DeepSORTKey words
vehicle-road cooperative/roadside perception/image recognition/YOLOX/DeepSORT分类
交通工程引用本文复制引用
李晓晖,夏芹,张强..基于路侧视觉感知的交通目标检测及跟踪方法研究[J].汽车工程学报,2025,15(5):699-706,8.基金项目
国家重点研发计划项目(2023YFC3009600):道路运输车辆重大事故风险防范与应急避险技术 （2023YFC3009600）
