多尺度降噪自编码器的遮挡行人重识别研究与应用

打开文本图片集

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）07-040-2220-07

doi：10.19734/j.issn.1001-3695.2024.09.0372

Abstract：Toaddress theisueofocclusioninpersonre-identification（ReID）andalleviate theimpactof insufficientocclusion datasets，this research focusedonoccudedReIand proposedamultiscaledenoising autoencoder-based method.The method usedknowledgedistllationlearninginastudent-teachermodelforjointtraining，enablingthetransferofknowledgefromthe teachermodel tothestudentmodel.Usingartificiallyccludedimages totraintheautoencoder，compressedtheinputdatainto alatentspacefeaturerepresentation，decoded toreconstructdatathatcloselyresemblestheoriginalinput，achievingdenoising reconstruction.Basedonthetrainedautoencoder，further trainingwithealocludedimagesandincorporatinganatentionmo duletodiferentiatebetweenthefeaturerepresentationsofoludedimagesandholisticimages，enhancedthemodel’srobustnesand recognition performance foroccluded images.Experimentsdemonstrate thattheproposed methodachievessuperior performance onthe Occude-Duke，Occluded-ReID，and Partial-ReIDdatasets compared tocurrentlyadvancedoccluded pedestrianre-identification approaches.

Key words：person re-identification；occlusion；denoising autoencoder；knowledge distillation

0 引言

行人重识别（ReID）是计算机视觉领域的一个重要研究方向，旨在解决如何在不同监控摄像头中识别同一行人的问题。（剩余21145字）