多通道特征级联的人体目标再识别  

Person Re-identification by Multi-Channel Feature Cascading

作　　者：曾明勇[1] 田畅[2] 吴泽民[2] 付毅[1] 揭斐然 

机构地区：[1]解放军理工大学通信工程学院研究生2队,江苏南京210007 [2]解放军理工大学通信工程学院,江苏南京210007 [3]中国航空工业集团光电控制技术重点实验室,河南洛阳417009

出　　处：《军事通信技术》2013年第3期1-6,共6页Journal of Military Communications Technology

基　　金：航空科学基金资助项目(20125186005)

摘　　要：人体目标再识别是视频监控等应用的关键问题之一,文章从特征级联和组合的角度研究了人体目标再识别问题。分析了局部特征集成(ELF)方法的缺点,研究了多通道特征级联的再识别性能,并将提出的级联特征与其他特征进行了组合,最后在公共图库上进行了广泛的验证和比较实验。实验结果表明,与ELF等其他特征相比,文章提出的多通道级联特征和组合特征可获得相当或更好的人体目标再识别性能。Person re-identification, one of the key problems in video surveillance, was discussed in view of feature cascading and combining. After a detailed analysis of the disadvantages of the ELF（Ensemble of Localized Features） method, the cascading of multi-channel features was investigated on the re-identification performance, and then combined with other features, extensive validation and comparative experiments were conducted on the public gallery. Experimental results show that the proposed excellent multi-channel feature and the combined feature can achieve comparable or better performance on person re-identification compared with other features such as ELF.

关 键 词：人体目标再识别 特征级联 特征组合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]