摘要
行人重识别技术是嫌疑目标跨摄像头的完整活动路径分析系统的核心技术,但是在被用于实际应用时面临计算复杂度高和存储开销大等问题.针对这些问题,以视频侦查应用为背景,提出一种基于空间颜色特征的行人重识别方法,建立一种在计算复杂度和性能上较均衡的行人外貌特征描述符,对外貌中直观和重要的颜色特征采用对光照具有不变性的颜色描述符进行描述,一定程度上降低了光照变化对颜色特征稳定性的影响.为了弥补颜色直方图缺少空间信息的缺陷,并提高颜色直方图对光照和姿态变化的鲁棒性,把行人团块按语义分割成躯干和腿两部分,再把每一部分细分成若干子块,然后分别计算颜色描述符,并在计算相似度时引入位置信息.实验结果显示本方法在取得接近最好性能的同时具有低计算复杂度的优点.
As one of the key technologies,pedestrian re-identification plays an important role in complete trajectory analysis of suspect objects cross multiple surveillance cameras,but it faces high computational complexity and large storage cost problems for practical application.Aiming at these problems,this paper provides a pedestrian re-identification method based on spatial color feature for video investigation application to construct appearance descriptor with a tradeoff of computational complexity and performance.The visual features of appearance are described by illumination invariant color descriptor,to a certain extent,which reduces the influence of illumination change on color stability.In order to compensate for the lack of spatial information of direct color histogram and improve appearance descriptor′s robustness to illumination and pose changes,pedestrian blob is divided into two semantic parts:the torso and legs.Then,each part is subdivided into several particles and color descriptors are computed separately,and location information is introduced in similarity calculation.Experiment results show this method gets performance close to the best,and has advantages of low computational complexity.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2013年第S2期209-212,217,共5页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
关键词
视频检索
视频监控系统
行人重识别
外貌描述
相似性度量
video retrieval
video surveillance system
pedestrian re-identification
appearance de-scription
similarity measure
