Abstract: Person re-identification is among the key issues in multi-camera surveillance system. According to the influence factors and the identification process of?pedestrian?human visual system, this paper presents a person re-identification method based on visual perceptual model. Two descriptors named local weighted CIELAB histogram and salient region features are chosen to build person appearance statistical characteristics. The histogram is with a vertical symmetry axis of pedestrian torso and legs as the center according to the local symmetry, and the salient feature is detected under the Bayesian framework according to local statistical feature. Then different distance measure methods are used and through the adaptive weights method to realize linear combination. Comparisons and analysis experiments based on VIPeR database verify the proposed method performance.

Key words: visual perceptual model, CIELAB color space, salient feature, person re-identification

摘要： 行人再识别是多摄像机协同监控系统中需要解决的关键问题之一。针对行人再识别问题的影响因素，根据人类视觉系统对行人进行识别的过程，提出一种基于视觉感知模型的行人再识别方法。该方法根据行人的局部对称性将行人分为头部、躯干和腿部，分别以行人的躯干和腿部的垂直对称轴为中心建立基于感知均匀颜色空间CIELAB的局部加权空间直方图，结合贝叶斯框架下基于局部统计特征的显著区域检测方法描述行人外观特征。两种特征分别采用不同的距离测度计算相似度，并通过自适应选取权值的方法进行线性融合。基于VIPeR数据库的实验比较和分析验证了该方法的行人再识别性能。

关键词: 视觉感知模型, CIELAB颜色空间, 显著特征, 行人再识别