计算机科学与探索 ›› 2018, Vol. 12 ›› Issue (1): 143-152.DOI: 10.3778/j.issn.1673-9418.1610036

• 人工智能与模式识别 • 上一篇    下一篇

步态骨骼模型的协同表示识别方法

关桂珍+,杨天奇   

  1. 暨南大学 信息科学技术学院,广州 510632
  • 出版日期:2018-01-01 发布日期:2018-01-09

Collaborative Representation Method for Gait Skeleton Model

GUAN Guizhen+, YANG Tianqi   

  1. School of Information Science and Technology, Jinan University, Guangzhou 510632, China
  • Online:2018-01-01 Published:2018-01-09

摘要: 针对目前步态识别中极易受到服饰和携带物等影响的难题,提出一种基于Kinect获取骨骼模型的步态识别新方法。对步态3D骨骼模型提取人体总质心,并与在步态周期中富有运动特征的人体四肢分质心的活动信息结合,分别得到动态与静态特征。动态特征可看作是周期信号,使用小波分解和带高斯滤波的离散傅里叶变换进行频谱处理,消除了外界干扰并增强了特征之间的差异性。通过动态时间规整算法把步态骨骼特征投影到相异空间,用协同表示进行匹配和归一化加权融合,最后根据最近邻算法进行分类识别。实验证明,该方法与稀疏表示识别算法相比得到了较为理想的识别效果,为步态识别在身份认证的应用领域提供了可靠的理论基础。

关键词: 步态识别, Kinect, 骨骼模型, 质心, 频谱分析, 协同表示

Abstract: Aiming at the problem of the influence which is easily caused by clothing and carrying object on existing gait recognition, this paper proposes a novel skeleton model-based method for gait recognition using Kinect. The center of mass of 3D gait skeleton model is obtained and combined with the activity information of the body limbs which have rich movement characteristics in the gait cycle to form the dynamic and static features respectively. Regarded as a periodic signal, the spectrum of dynamic features is analyzed by wavelet transform and discrete Fourier transform with Gaussian filter. This processing can eliminate the effect of external interference and enhance the difference among features. After projecting into dissimilarity space via dynamic time warping and matching by collaborative representation classification, the gait features are normalized, weighted and fused. Finally, the nearest neighbor algorithm is adopted to classify. Compared with sparse representation recognition, the experimental results show that the proposed method can improve recognition effect and provide reliable theoretical basis for gait recognition in the field of identity authentication.

Key words: gait recognition, Kinect, skeleton model, center of mass, spectrum analysis, collaborative representation