基于广义主成分分析的步态识别算法研究

步态识别是根据人行走方式的不同对人的身份进行识别的.通过背景减除实现人体检测,运用形态学操作和图形几何变换实现了图像的标准中心化.在特征提取阶段使用步态能量图(GEI)来描述每个步态序列,分别使用主成分分析、二维主成分分析、完全的二维主成分分析以及加权完全的二维主成分分析对特征进行降维,最后采用最近邻分类器来测试识别结果作对比研究.实验结果表明权衡计算量和识别率,二维主成分分析对于GEI的步态识别比较有效,识别率可达95.43%.

Research on a gait recognition algorithm based on generalized principal component analysis

Gait recognition makes use of human walking patterns for biometric recognition and is a novel topic in the field. Effective segmentation using a simple method to extract silhouettes of walking figures from the background was the first step of our method; this played a key role in gait recognition. Then, morphology was used and a standardized and centralized image was obtained by geometric transformation. Afterwards gait energy image (GEI) was used as a feature extraction method--describing gait characteristics obtained using periodic sequence images according to their cyclical divisions. Following this, feature dimensionality was reduced through principal component analysis (PCA), two-dimensional principal component analysis (2DPCA), complete two-dimensional principal component analysis (C2DPCA) and weighted complete two-dimensional principal component analysis (WC2DPCA) respectively. The nearest neighbor classifier was then used to distinguish different human gaits. By balancing calculation and recognition rates, experimental results demonstrated that 2DPCA based on GEI has encouraging recognition performance with a recognition rate of about 95.43%.