基于MSCR的快速行人重检方法  

Fast person re-identification method based on MSCR

作　　者：潘明 唐宁[1,2] 

机构地区：[1]中国航天科工集团第二研究院 [2]北京航天长峰科技工业集团有限公司,北京100854

出　　处：《计算机工程与设计》2013年第12期4289-4293,共5页Computer Engineering and Design

摘　　要：针对视频监控系统数据量大的现实情况,提出一种快速行人重检方法,旨在提高检索速度。依据对称性和非对称性原则分割图像中的行人,在此基础上检测人体各部分的最大稳定色彩区域;以人体分割为依据分组,对每组的所有MSCR特征描述子分别进行K-means聚类,构建基于聚类中心的视觉词汇词典和倒排索引,以加速行人重检的匹配过程。实验结果表明,和已有方法相比,该方案在保证检索准确性的前提下具有更高的检索速度。A fast person re-identification method is presented to improve searching speed for current situation of huge data in video surveillance system. Pedestrian in the image is partitioned by adopting perceptual principles of symmetry and asymmetry. Then, maximally stable color regions are detected on each part of the human body. Grouped based on body partition, MSCR features are clustered using K-means algorithm in each group. A dictionary of visual words and inverted index are structured based on the center of clustering, aiming at speeding the matching progress of person re-identification. Experimental results show that the method proposed is more efficient than former method with the guarantee of searching accuracy.

关 键 词：行人重检 最大稳定色彩区域 词袋模型 K均值聚类 倒排索引 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]