基于轮廓分析的广义步态识别算法研究

Generalized Gait Recognition Based on Silhouette Analysis

在步态识别中,最关键的是获取步态特征之后如何选择最佳投影方向,且计算复杂度较小.因此,根据对现有算法的分析,提出一种基于轮廓特征的的广义步态识别算法.在传统的线性判别分析方法基础上,通过重新定义样本类间离散矩阵寻找最佳投影方向,使不同的目标映射到同一低维空间中,在保留同类结构信息的同时最大化不同类的间距.首先对每个序列进行运动轮廓提取,根据轮廓解卷绕方法将二维轮廓形状转换为一维距离信号,并通过广义线性判别分析方法(Generalized Linear Discriminative Analysis,GLDA)得到最佳投影空间,最终利用支持向量机(Support Vector M achine,SVM)完成分类识别.实验结果表明,该算法简单有效,具有更高的识别率,并且计算代价及处理速度明显优于其他现有算法.

The key of gait recognition is to search the best projection direction after obtaining the gait features with lower computational complexity. A generalized Linear Discriminative Analysis based on Euclidean norm（ GLDA） for gait recognition is proposed in this paper. Based on the classical algorithm of LDA,GLDA seeks a mapping to project human gait sequences collected from different people into a low-dimensional feature subspace by redefining a more standard scatter matrix between classes,such that intraclass geometrical structures are preserved and interclass distances of gait sequences are maximized simultaneously. Firstly,with the counterclockwise unwrapping,2D silhouette image are transformed into 1D normalized distance signals,and the distances of a walking person are chosen as the basic image feature. Secondly,the best projection space is got by GLDA. Finally,the recognition task is completed according to Support Vector Machine（ SVM）. Experimental results showthat the proposed algorithm is simple and effective and outperforms other existing approaches in terms of recognition accuracy,computational complexity and processing speed.