基于距离匹配的行人再识别技术综述

Survey of Pedestrian Re- identification With Metric Learning

行人再识别是指给定一张行人图像,在已有的可能来源于非交叠摄像机视场的行人图像库中,识别出与此人相同的图像。研究该问题有着非常重要的现实意义,同时也面临许多挑战。它是智能视频监控中的一项关键任务,是近年来计算机视觉领域中一直备受关注的研究热点,适用于安防以及公共场所寻人等技术领域。距离度量是行人再识别技术中存在的核心问题之一。对现有的基于距离匹配的行人再识别方法进行评述,并分析其中具有代表性方法的优缺点,介绍了常用行人数据库的特点,然后总结现阶段行人再识别研究所面临的挑战,最后对行人再识别技术的未来发展方向进行了展望。

Pedestrian re- identification,identifying the same person＇s images in an existing database from non- overlapping camera views,as a key task in intelligent video surveillance and a major concern in the field of computer vision in recent years,is valuable but challengeable. And the technique can be applied in different important applications,e. g. security and finding someone in the public place.Distance measure is one of the core issues of pedestrian re- identification technique,and we review the existing methods based on metric learning. According to the research,the advantages and disadvantages of different typical methods are discussed,and the characteristics of popular database are described,then,summary the technique challenges faced by the institute at this stage. Finally,the development of pedestrian re- identification in the future is proposed.