融合腿部局部特征的步态识别方法

Gait recognition method with local feature of legs region

为提高步态识别的识别率,提出一种基于腿部轮廓区域与整体图像的特征相结合的步态识别算法。通过选取训练样本的整体散布矩阵的非负特征值对应的特征向量,组成一个维数较低的变换空间;在此空间中,为克服小样本问题,对每类样本的协方差矩阵增加一个正则项,构成新的准则函数,通过求解该优化问题,选择部分特征向量组成特征矩阵;基于此种方法,将获得的腿部轮廓区域特征和整体图像的特征进行组合,表示一个行人步态的特征。实验选取中科院步态数据库CASIA A、CASIA B和CASIA C,结合最小距离分类器进行步态识别。实验结果表明,当影响识别的因素发生改变时,例如背包、穿外套等,所提方法能够获得较高识别率。

To improve the results of gait recognition,the gait recognition method based on the feature combination of gait image and its region bounded by legs was proposed.All the eigenvectors corresponding to nonnegative eigen values of the entire scatter matrix of training samples were selected to compose a lower-dimension transform space.In this transform space,to overcome the small-sample-size problem,a regularization term was added to each sample class covariance matrix,and a new criterion function was established.By computing this optimization problem,and the eigen matrix was made up by some eigen vectors.Based on this method,the features of gait image and its region bounded by legs were combined to represent gait features.In the experiments,three gait databases CASIA A,CASIA B and CASIA C were selected and the minimum distance classifier was used to verify the effectiveness of presented method.Experimental results show that when the factors changes,such as walking with bag,the presented method gets better results.