重庆理工大学学报2025，Vol.39Issue(5)：10-16,7.DOI:10.3969/j.issn.1674-8425(z).2025.03.002
复杂交通环境下智能车多目标重识别跟踪方法研究
Research on multi-target tracking method of target reidentification of intelligent vehicle in complex traffic environment
摘要
Abstract
To address the target ID association assignment failure and low tracking accuracy due to occlusion of the target in front of intelligent vehicle in complex traffic environment,we propose a multi-target tracking method.Based on deep learning theory,we build a ReID model for target recognition,which fully acquires target global feature and foreground feature.The global feature selection model strengthens the learning of feature region and effectively extracts discriminant global features.The spatial local feature selection model divides the global feature into multi-scale,distinguishing foreground feature from background one as well as learning foreground feature from multi-scale spatial one.The correlation cost matrix of cosine distance and IoU distance is established to realize the correlation between detection object,tracking trajectory and ReID feature similarity.VeRi-776 vehicle rerecognition data set and Market1501 pedestrian rerecognition data set are used to train our ReID model.Multi-target tracking performance comparison experiments are conducted on MOT16 data set.Our results show the multi-target tracking method by introducing ReID model and improving cost matrix better deals with the dense occlusion of the target and tracks the road target accurately in real-time complex traffic environment.关键词
智能汽车/目标重识别/多目标跟踪/深度学习Key words
intelligent vehicle/target reidentification/multi-target tracking/deep learning/simulation分类
交通运输引用本文复制引用
延世龙,陈学文,贾远鹏..复杂交通环境下智能车多目标重识别跟踪方法研究[J].重庆理工大学学报,2025,39(5):10-16,7.基金项目
国家自然科学基金面上项目(62373175) （62373175）
辽宁省自然科学基金项目(2022-MS-376) （2022-MS-376）
辽宁省教育厅重点攻关项目(JYTZD2023081) （JYTZD2023081）
辽宁省属本科高校基本科研业务费专项资金资助项目(LJZZ232410154016) （LJZZ232410154016）
