面向无人车近距离行人跟踪的自适应双重识别技术  

Self-adaptive double Re-ID technique for closepedestrian tracking of unmanned vehicles

作　　者：黄文艺 王博 孙超[1] 卢兵[1] 彭勇礼 张春霞 HUANG Wenyi;WANG Bo;SUN Chao;LU Bing;PENG Yongli;ZHANG Chunxia(National Engineering Research Center of Electric Vehicles,Beijing Institute of Technology,Beijing 100081,China;Shenzhen Automotive Research Institute,BTT(Shenzhen Research Institute of National Engineering Laboratory for Electric Vehicles),Shenzhen 518000,China)

机构地区：[1]北京理工大学电动车辆国家工程实验室,北京100081 [2]北京理工大学深圳汽车研究院电动车辆国家工程实验室深圳研究院,深圳518000

出　　处：《重庆理工大学学报（自然科学）》2023年第6期110-118,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金重点项目(U1964206);科技部重点研发计划项目(2022YFB2503203)。

摘　　要：智能无人小车的感知系统作为前端输入,对小车的功能应用至关重要,但是现有的传感器均存在视野盲区,导致无人小车对近距离目标不敏感,易引发安全问题。因此,对近距离目标有效识别和跟踪成为了亟需解决的问题。提出一种自适应双重识别技术,该技术采用当前主流目标检测器和目标跟踪器,并结合重识别技术对近距离行人进行实时跟踪。该方法弥补了当前深度学习算法强依赖数据集的不足,提高了算法的通用性,增强了无人小车感知范围,提升了无人车的商业价值。实车实验表明,该算法可实现对近距离行人从远至近、从近至远的持续跟踪,行人ID保持稳定。The perception system of intelligent autonomous vehicles,serving as the frontend input,plays a crucial role in the functional application of these vehicles.However,existing sensors have blind spots in their field of view,leading to a lack of sensitivity towards close-range targets,which can potentially cause safety issues.Thus,effective recognition and tracking of near-distance targets have become pressing issues to address.This paper proposes an adaptive dual recognition technique that employs the current mainstream target detectors and trackers and combines them with re-identification technology for real-time tracking of nearby pedestrians.This method compensates for the shortcomings of current deep learning algorithms that are heavily dependent on datasets,improves the versatility of the algorithms,expands the perception range of autonomous vehicles,and enhances their commercial value.Actual vehicle experiments demonstrate that the algorithm can continuously track pedestrians from far to near and vice versa,while maintaining a stable pedestrian ID.

关 键 词：无人车 视觉 近距离行人检测 自适应双重识别技术 重识别网络 

分 类 号：U461.99[机械工程—车辆工程]