基于关节角度和DTW的太极拳视频配准方法

TaiChi Video Registration Method Based on Joint Angle and DTW

太极拳视频配准是实现太极拳线上教学的首要问题。为实现太极拳视频的自动配准,提出了一种基于关节角度和DTW算法的太极拳视频配准方法。该方法主要利用人体关节角度消除太极拳视频背景的干扰和不同太极拳视频中人体大小不同的影响,并利用动态时间规整(DTW)算法对不同时间点的视频帧进行配准。在该方法中,首先计算出练习者动作视频中关节角度的时间序列,并使用指数平滑法消除时间序列中存在的误差;然后利用上下帧之间的人体关节角度差分割时间序列;最后利用DTW算法求分割后得到的时间序列与标准动作视频中对应的时间序列之间的距离,即可得到练习者与标准动作之间的匹配度。实验结果表明:该方法中的指数平滑法对太极拳视频配准的精度有较大影响,以及如果用欧几里得距离替换DTW算法将会较大的降低配准精度。并且该方法在太极拳视频配准上与基于SIFT特征的方法相比,配准精度更高,达到81.21%。

TaiChi video registration is the primary problem in the online teaching of TaiChi.In order to realize the automatic registration of TaiChi video,a TaiChi video registration method based on joint angle and DTW algorithm is proposed.The method mainly uses the human joint angle to eliminate the interference of the TaiChi video background and the different human body size in different TaiChi videos,and uses the DTW algorithm to register the video frames at different time points.In this method,the time series of the joint angle in the exercise video of the practitioner is first calculated,and the error existing in the time series is eliminated by exponential smoothing;then the time series is segmented by the angle difference of the human joint between the upper and lower frames;Dynamic Time Warping(DTW)algorithm obtains the distance between the time series obtained after the segmentation and the corresponding time series in the standard action video,and the matching degree between the practitioner and the standard action can be obtained.The experimental results show that the exponential smoothing method in this method has a great influence on the accuracy of Tai Chi video registration,and if the DTW algorithm is replaced by Euclidean distance,the registration accuracy will be greatly reduced.Moreover,compared with the method based on SIFT feature,the method has higher registration accuracy of 81.21%in Tai Chi video registration.