基于核空间与稠密水平条带特征的行人再识别

Dense Horizontal Stripes and Kernel Space Mapping for Person Re-Identification

摘要:

行人再识别指的是在无重叠视域多摄像机监控系统中, 匹配不同摄像机视域中的行人目标。由于行人图像受到光照、视角和行人姿态等变化的影响,在视觉上容易形成很大的外观差异,针对上述问题,提出了一种基于核空间与稠密水平条带特征的行人再识别算法。该算法首先通过自顶向下的滑动水平条带提取每个水平条带的颜色特征和纹理特征,然后融合行人图像的多种特征,映射到核空间,最后在核空间里学习得到一个对背景、视角、姿势的变化具有鲁棒性的相似度函数,通过比较相似度来对行人进行再识别。在VIPeR和iLIDS两个行人再识别数据集上的实验结果表明,本文算法具有较高的识别率,其中Rank1(排名第1的搜索结果即为待查询行人的比率)分别达到48.2%和60.8%。

Abstract:

Person re-identification is to match person images observed from different camera views of non-overlapping multi-camera surveillance systems. The person images are easily affected by illumination changes, different viewpoints and varying poses, it is likely to form a lot of differences in appearance. For the above problem, this study proposed a dense horizontal stripes and kernel space mapping for person re-identification. First, the each horizontal stripe of person images is extracted from color features and a texture feature by using the top-down sliding horizontal stripe. Then, multi-features of person images fusion and kernel space mapping. Finally, the algorithm gets a similarity function which is robust to the change of background, viewpoint and posture by learning in kernel space. The proposed method is demonstrated on two public benchmark datasets including VIPeR and iLIDS, and experimental results show that the proposed method achieves excellent re-identification rates compared with other similar algorithms. Moreover, the proposed method achieves a 48.2% at rank1 (represents the correct matched pair) on VIPeR benchmark and a 60.8% at rank1 on iLIDS benchmark respectively.