应用空间约束和二次相似度学习算法的行人再识别

针对空间分布的全局外观潜在变化的行人再识别问题,提出一种基于空间约束和二次相似度学习算法.通过二次相似度函数(QSF)估计每个子区域的相似度,从而形成多项式特征图,并将所有特征图融合到统一的框架中.该框架利用局部相似度和全局相似度的互补优势,结合多个视觉线索进一步提高算法的鲁棒性.实验对比3个公共数据集,结果表明:基于空间约束和二次相似度学习算法具有显著的识别性能.

Pedestrian Re-Identification Using Spatial Constraint and Quadratic Similarity Learning Algorithm

According to the pedestrian re-identification problem with potential changes in global appearance of spatial distribution, a learning algorithm based on a spatial constraint and quadratic similarity is proposed. The similarity of each sub-region is estimated by the quadratic similarity function(QSF)and the polynomial feature graph is formed and all the feature graphs are merged into a uniformed framework. The framework utilizes the complementary advantages of local similarity and global similarity to flexibly combine multiple visual cues to further improve the robustness of the algorithm. Compared with the experimental results of the three public data sets, the identification performance of proposed method is significantly improved compared with the existing pedestrian re-identification method.