基于支持样本间接式的行人再识别

发布时间：2018-05-15 13:36

本文选题：行人再识别 + 支持样本　；参考：《电子与信息学报》2017年12期

【摘要】：行人再识别就是在无重叠视域多摄像机监控系统中,识别出相同的行人。针对来自于不同摄像头行人图片存在着视角、光照和尺度变化的问题。该文提出了基于支持样本间接式匹配的行人再识别方法。该算法首先通过聚类的方法分别提取不同摄像头下的支持样本,当要对来自不同摄像头的行人进行匹配时,在距离测度的基础上利用支持样本分别判别出其所在摄像头下的行人类别,通过类别的对比判断是否为同一行人。该方法避免了不同摄像头下行人图片直接匹配,有效解决不同摄像头带来的视角、光照和尺度问题。实验结果表明该文的算法相比一些经典算法识别率有一定的提高,并且在数据集VIPe R,CAVIAR4Re ID和CUHK01上,Rank1分别达到了43.60%,41.36%,43.82%。
[Abstract]:Pedestrian recognition is to identify the same pedestrian in a multi-camera surveillance system. There are problems in view, illumination and scale change of pedestrian images from different cameras. In this paper, a pedestrian recognition method based on support sample indirect matching is proposed. The algorithm firstly extracts support samples from different cameras by clustering method, and then matches pedestrians from different cameras. On the basis of distance measure, the support samples are used to identify the pedestrian category under the camera, and to judge whether the pedestrian is the same pedestrian by the comparison of the categories. This method avoids the direct matching of downlink images of different cameras, and effectively solves the problems of angle of view, illumination and scale brought by different cameras. The experimental results show that the recognition rate of the proposed algorithm is higher than that of some classical algorithms, and the Rank1 is 43.60 and 43.82 on the data set VIPe RnCAVIAR4Re ID and CUHK01, respectively.
【作者单位】：合肥工业大学计算机与信息学院;工业安全与应急技术安徽省重点实验室;
【基金】：国家自然科学基金(61471154) 安徽省科技攻关科技强警项目(170d0802181)~~
【分类号】：TP391.41

【相似文献】