局部约束组稀疏表示的步态识别方法

Locality constrained group sparse representation for gait recognition

提出一种局部约束组稀疏表示的步态识别方法。通过预处理提取人体二值化侧影图,计算步态周期并利用HS(Horn-Schunck)算法生成步态光流图,经降维后利用局部约束组稀疏表示的方法进行分类识别。在标准稀疏表示分类方法的基础上,引入了组稀疏约束和局部平滑稀疏约束,使其最小重构误差的非零重构系数分散在与测试样本相邻的同一训练类别组内。在CASIA Dataset B数据库上的实验结果表明,该方法有较高的识别率。

A locality constrained group sparse representation for human gait recognition was presented.First a preprocess tech-nique was used to segment the human silhouette from the walking videos,then gait period was calculated and gait optical flow image was generated by HS algorithm,after dimension reduction,the GFI was classified using locality constrained group sparse representation.The method introduced group sparsity constraint and local smooth sparsity constraint based on standard sparse representation classification algorithm.Experiments with CASIA Dataset B showed that the method outperformed several other gait recognition methods.