基于图像切块的多特征融合行人再识别模型  

MULTI-FEATURE FUSION PERSON RE-IDENTIFICATION MODEL BASEDON IMAGE PARTITION

作　　者：张羽 朱玉全[1] Zhang Yu;Zhu Yuquan(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013,Jiangsu,China)

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《计算机应用与软件》2023年第8期207-213,共7页Computer Applications and Software

摘　　要：针对目前由于监控场景背景复杂、拍摄视角差异引起的外貌变化,需要更多细节的特征来区分不同行人的问题,提出一种基于图像切块的多特征融合行人再识别模型。该模型包含3个特征提取分支,分支1和2负责提取行人不同层级的全局特征,分支3将图像水平分为上下两个部分并分别进行局部特征提取,将三个分支的特征进行深度融合。针对全局特征和局部特征的差异联合三种损失函数进行监督训练。在主流的数据集上进行了验证,结果证明该模型可以显著提高行人再识别的准确率。Person has significant changes in appearance due to differences in shooting angles and complex background of monitoring scenes,which needs more detailed features to distinguish different pedestrians.Aimed at this problem,a multi-feature fusion person re-identification model based on image partition is proposed.This model contained 3 feature extraction branches.Branch 1 and Branch 2 were responsible for extracting global features of pedestrians at different levels.Branch 3 divided the image horizontally into upper and lower parts and extracted local features respectively.The features of the three branches were deeply fused.Aimed at the difference between global features and local features,three types of loss functions were combined for supervised training.The proposed model was verified on mainstream datasets.The results prove that the model can significantly improve the accuracy of person re-identification.

关 键 词：行人再识别 图像切块 多层级特征 全局特征 局部特征 特征融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]