摘要
针对局部线性嵌入(LLE)的非线性流形降维在步态识别率较低的问题,提出了一种加权距离测试的LLE流形降维步态识别方法.该方法依据LLE理论对训练数据重构低维特征,采用加权距离技术对测试数据重构低维特征,最后依据最小欧式距离进行识别并分析了降维过程中参数选取的问题.对单视点UCSD步态数据库和多视点步态数据库进行的实验表明,该方法的正确识别率优于LLE算法和加权LLE算法,但计算时间与LLE算法相当.
In order to improve the recognition rate of nonlinear manifold LLE (locally linear embedding) applied to gait, a new LLE gait recognition algorithm was proposed, which reduces the dimension in the manifold based on da- ta tested by weighted distance. Firstly, dimensional reduction for training data was calculated by LLE; secondly, weighted distance was utilized to reduce the dimension of the testing data; finally, minimal Euclid-distance was ap- plied to recognize the persons. At the same time, parameter selection in the dimensional reduction was also analyzed in this paper. The proposed algorithm verified by the one-view UCSD (University of California, San Diego) gait da- tabase and multi-view gait database shows that the correct recognition rate is higher than that of the LLE and weigh- ted distance LLE method, however, the calculation time is almost equal to LLE.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2013年第11期1421-1426,1444,共7页
Journal of Harbin Engineering University
基金
国家863计划资助项目(2008AA12A218-51)
华北水利水电大学高层次人才科研启动项目基金资助项目(201234)
关键词
局部线性嵌入
加权距离
步态识别
流形降维
LLE
weighted distance
gait recognition
reducing dimension in manifold