光流分量分解的步态识别-计算机科学.pdfVIP

第 43 卷第9 期 计算机科学 Vo1. 43 No.9 2016 年 9 月 Computer Science Sep 2016 光流分量分解的步态识别 罗正平刘延钩杨天奇 (暨南大学信息科学技术学院 广州 510632) 摘 要 步态识别因其远距离和难于伪装等特点在生物识别技术中颇受关注。针对目前特征提取方法信息量不足的 现状，提出一种基于光流分量分解的步态识别方法，对步态光流图中横向分量和纵向分量为正的部分按行、列进行分 解，求出每行和每列的光流横、纵向分量，得到 4 个特征向量。根据训练得出的每个特征向量在识别过程中的权重进 行特征融合。将主成分分析和线性判别分析相结合，用动态时间规整算法进行匹配，最后采用最近邻分类算法分类。 在 CASIA Database B 和 C 上的实验表明，该方法在正常、背包和穿大衣的条件下分别得到了 97% ， 90%和 64%的识 别率，在慢速和快速行走的条件下分别得到了 88%和 87%的识别率。 关键词 步态识别，光流，主成分分析，线性判别分析，动态时间规整，特征融合 中图法分类号 TP39 1. 41 文献标识码 A 001 10. 11896/j. issn. 1002-137叉 2016.9.059 Gait Recognition ßased on Decomposition of Optical Flow Components LUO Zheng-ping LIU Yan-jun YANG Tian-qi (College of Information Science and Technology ,Jinan University ,Guangzhou 510632 ,China) Abstract Gait recognition has gained tremendous attention for its characteristics of long distance and hard-tσdisguise. Aiming at the problem of insufficient information of the existing feature extraction method ,this paper proposed a novel gait recognition method based on the decomposition of optical flow components. The positive transverse or longitudinal components in gait optical flow image are decomposed by rows and columns ,then the transverse and longitudinal com- ponents of optical flow for each row or colum are calculat吐， and four feature vectors are obtained. The four vectors are fused according to their weight in recognitio n. Principle components analysis and linear discriminant analysis techniques- are combined ,and dynamic time warping algorithm is used to match. Finally ,K-nearest neighbor algorithm