多模態(tài)行人重識(shí)別研究綜述

打開(kāi)文本圖片集

中圖分類(lèi)號(hào)：TP391 文獻(xiàn)標(biāo)志碼：A 文章編號(hào)：1001-3695（2025）07-001-1921-09

doi：10. 19734/j. issn. 1001-3695.2024.11.0433

Abstract：Inmodernsecuritysurveillancesystems，pedestrianre-identification technologyplaysacrucialrole.Traditionalpedestrianre-identificationsystems face limitations inaccuracyandreliabilityduetochalengessuchaschanges in lighting， diferences inviewingangles，andocclusions inpedestrianimages.Toaddress thesechalenges，researchershaveintroduced multi-modalleaing methodsintotheRE-IDfield，hoping toefectivelyintegratevariousdatamodalities，suchasdepthiages，infraredimage，ndextualinformatio，toimprovetepefoanceofRE-ID.Tispaperviewedthepicationadresearch progressofmulti-modalRE-IDtechnologyinmodernsecuritysurveillancesystems.Itfirstlyintroducedthebasicconcepts of multi-modaltechnologyandthetaskofmulti-modalRE-ID，andthenoutlinedthekeydatasetsandevaluationprotocols inthisfield.Thecorepartofthearticlediscussedindetailthefusionstrategiesinmulti-modalRE-ID，ncludingfeature-level fusion andmodel-levelfusion methods.Finall，thearticlediscussd the aplication scenarios andfutureresearch directionsof multi-modal RE-ID.

KeyWords：RE-ID；multimodal fusion；data fusion；feature fusion；model fusion

0引言

隨著智能監(jiān)控技術(shù)的快速發(fā)展，RE-ID技術(shù)在實(shí)時(shí)監(jiān)控以及智能安防等多個(gè)領(lǐng)域發(fā)揮著重要作用[1]。（剩余30028字）