基于空-频域特征和线性判别分析的视频步态识别

Video human gait recognition based on space-frequency domain features and linear discriminant analysis

为提高步态识别准确率,提出了基于空-频域特征和线性判别分析的视频步态识别方法。利用离散余弦变换、Contourlet变换分别提取步态能量图的频率特征和多尺度多方向轮廓特征;融合得到空-频域特征,并通过线性判别分析映射到最佳鉴别矢量空间;根据相似性距离实现身份识别。在中科院自动化所提供的数据库中进行实验,结果表明,提出的特征提取方法优于现有常用方法。空-频域特征能够有效地区分步态中的高低频分量,并捕捉丰富的细节信息,线性判别分析在降维的同时进一步增强特征的判别能力,有助于提高识别精度。

To improve gait recognition rates,a video human gait recognition algorithm based on space-frequency domain features and linear discriminant analysis（LDA）is proposed.The frequency-domain features and the multi-direction contours are extracted by discrete cosine transform（DCT）and contourlet transform from gait energy image（GEI）,respectively.The space-frequency domain features acquired by fusing both of them are further mapped into the optimal discriminant vectors space by LDA.The identity recognition is implemented according to the shortest Euclidean distance.Experiments on the database provided by Chinese Academy of Sciences（Institute of Automation）demonstrate that the proposed feature extraction strategy is superior to other commonly existing methods.The space-frequency features can effectively distinguish different frequency components and adequately capture details of human gait.LDA can enhance the discrimination and reduce the dimensionality of features.They are both conductive to improve recognition accuracy.