步态识别特征的提取和重要性排序

Extraction and importance ranking of features for gait recognition

提出一种基于骨骼关键点检测技术的步态识别方法,并讨论常见的步态特征在障碍诊断时的重要性排序,以期为步态障碍的及时检查和识别提供参考。首先,根据不同的步态模式设立试验组和对照组,受试者按照要求分别完成规定的范式动作,经由深度相机实时获取受试者骨骼关键点的运动轨迹数据;然后,从获取的数据中提取步态识别特征;最后,采用以信息增益为依据的模糊二元对比决策方法对步态识别特征的重要度进行排序。通过对每个特征的重要性进行排序,为步态障碍的诊断和自动识别提供参考。研究结果表明3种步态模式下所有步态识别特征的平均值之间均有显著差异,特征的优先排序为步频、膝关节最大屈曲角度、足偏角、髋关节最大屈曲角度、步幅、支撑相、步速、髋关节最大伸展角度、质心左右移动、膝关节最大伸展角度、质心上下移动、步长、步宽。

Agait recognition method based on skeleton keypoint detection technology is proposed,and the importance of common gait features in the diagnosis of gait disorders is discussed,so as to provide a reference for the timely examination and recognition of gait disorders.Firstly,an experimental group and two control groups are set up according to different gait patterns.All the subjects complete the prescribed normal motions according to the requirements,and the real-time movement trajectory data of skeleton keypoints of the subjects are obtained via depth camera.Then,gait recognition features are extracted from the obtained data.Finally,the importance of gait recognition features is ordered by fuzzy binary comparison decision method based on information gain,thereby providing a reference for the diagnosis and automatic recognition of gait disorders.Significant differences are found in all features for gait recognition among 3 different gait patterns.The priority order of gait features is as follows: stride frequency,maximum flextion angle of knee,toe-out angle,maximum flextion angle of hip,stride,stance phase, gait speed,maximum extension angle of hip,L/R distance of the center of mass,maximum extenstion angle of knee,U/D distance of the center of mass,step length and stride width.