MSR结合MCCA的多视角鲁棒步态识别方法

On Multi-View Robust Gait Recognition Method Based on Fusion of MCCA and MSR

针对许多传统的步态识别算法过分依赖视角的问题,提出了一种基于多视角子空间表示结合边际典型相关分析的步态识别方法.首先,使用简单有效的特征步态能量图表示每个序列,并利用多视角子空间表示方法测量样本之间的差异;然后,利用MCCA提取子空间的判别性信息;最后,利用最大隶属度原则和最近邻分类器完成识别.在公开的多视角步态数据库OU-ISIR-D和CASIA-B上的实验结果显示,方法在CASIA-B,OU-ISIR-D两个数据库上的识别精度可分别高达99.8%,99.1%,相比几种较为先进的步态识别方法,该方法取得了更好的识别性能且能有效处理对象内变化和缺失数据.

For the issue that many traditional gait recognition algorithms is relied too much on angle of view,agait recognition method based on fusion of multi-view subspace representation（MSR）and marginal canonical correlation analysis（MCCA）has been proposed.Firstly,a sample effective feature gait energy diagram has been used to represent each sequence,and MSR method used to measure the variances among samples.Then,MCCA has been used to exploit the discriminative information from these subspaces for recognition better.Finally,maximum membership degree principle and nearest neighbor classifier has been used to finish recognition.Experimental results on a widely used multi-view gait database CASIA-B and OU-ISIR-D show that the recognition accuracy of proposed method can achieve 99.8%and 99.1% on CASIA-B and OU-ISIR-D respectively.It has better recognition performance than several advanced gait recognition methods,and it is effective in handling intra-subject variations and missing data.