采用Radon变换和二维主成分分析的步态识别算法

Gait recognition with Radon transform and 2-D principal component analysis

针对主成分分析算法将图像矩阵转化为向量的维数过高、求取特征向量耗时的问题,综合步态的静态和动态信息,对一个步态周期中的图像进行Radon变换,再通过模板构造,仅用一幅图像来刻画步态特征,接着用二维主成分分析(2DPCA)进行降维.为了验证所提出的算法的有效性,在CASIA步态数据库上进行实验,采用最近邻分类器来测试识别.实验结果表明在特征模板构造时选择合适的频率,采用Radon变换结合列2DPCA进行步态特征提取是有效的.

In the principal component analysis method,concatenating an image matrix often leads to a 1-D vector with high dimensionality,which makes it very difficult and time-consuming to compute the corresponding eigenvec-tors.By combining static components and dynamic information about the walking style,a novel gait representation was proposed.Gait characteristics were obtained from the Radon transform of gait sequences,where a single image could represent a person’s features by template construction.Then,two dimensional principal component analysis （ 2DPCA） was used to reduce the dimensions of training and testing data.The nearest neighbor classifier was em-ployed to distinguish the different gaits of human.We tested the proposed gait recognition method on the CASIA gait database.The experimental results demonstrated that,when frequency is chosen properly in template construc-tion,extraction of gait features using the Radon transform and column 2DPCA is very effective.