复杂环境下的行人重识别方法  

A Pedestrian Re-Identification Method in Complex Environments

作　　者：谭泽桓 朱文球 TAN Zehuan;ZHU Wenqiu(College of Computer,Hunan University of Technology,Zhuzhou Hunan 412007,China)

机构地区：[1]湖南工业大学计算机学院,湖南株洲412007

出　　处：《湖南工业大学学报》2025年第4期35-40,共6页Journal of Hunan University of Technology

基　　金：湖南省教育厅科学研究基金资助项目(23A0423)。

摘　　要：针对行人重识别技术在实际应用中常常因为特征遮挡而导致识别率不高的问题,提出了一种复杂环境下行人防遮挡重识别方法,其由全局和局部特征提取两部分组成。以ResNet-50网络为主干网络,首先,在全局特征提取使用特征感知注意力机制对全局特征进行提取;然后,在局部分支使用特征分割空间的方法进行局部特征提取;最后,通过多尺度双向金字塔网络进行特征融合。在常用遮挡行人数据集Market-1501等上进行了实验,验证了论文所提出方法的有效性,且提升了行人重识别效果。In view of the flaw that pedestrian re-identification technology is characterized with a low recognition rate due to feature occlusion in practical applications,a complex environment pedestrian defense occlusion re-identification method has thus been proposed,which consists of two parts:global and local feature extraction.Firstly,with ResNet-50 network as the backbone network,feature aware attention mechanism is used for a global feature extraction to extract global features;subsequently,the method of feature segmentation space is used for the local feature extraction in the local area;finally,feature fusion is performed through a multi-scale bidirectional pyramid network.Experiments are to be conducted on commonly used pedestrian occlusion datasets such as Market-1501,etc.which verifies the effectiveness of the proposed method and improves the effectiveness of pedestrian re-identification.

关 键 词：行人重识别 目标检测 特征融合 全局特征 局部特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]