基于条件生成对抗网络的人体步态生成

Gait generation of human based on the conditional generative adversarial networks

针对人体个性化步态生成研究中的生成目标单一、个性化特征刻画不全面等问题,提出一种基于条件生成对抗网络的人体个性化步态生成方法。首先,将全身共51个关节角作为生成目标;其次,根据个体参数、行走速度、关节构成和协同关系等行走特征,建立数据标签并构成条件信息;再次,利用条件生成对抗网络模拟人体步态形成过程;最后,通过调整条件信息生成具有不同行走特征的个性化步态。经实验分析,该方法生成的个性化步态与真实行走数据的相关系数高于0.98,平均绝对偏差小于0.08 rad,阈值绝对偏差在5%以下,且步态稳定性判据结果均处于稳定区间内。实验结果表明,该方法能够有效生成对应不同行走特征的个性化步态,同相似研究相比,对行走特征的刻画更全面,生成步态具有更好的整体性。

To solve the problems of single target generation and incomplete characterization of personalized features in the study of human personalized gait generation, this paper proposes a method of human personalized gait generation based on the conditional generative adversarial networks. Firstly, a total of 51 joint angles of the whole body are set as pre-processing targets. Secondly, according to the walking parameters such as individual parameters, walking speed, joint composition and synergy relationship, data are labelled and condition information is constructed. Then, the human gait formation process is simulated by the conditional generation confrontation networks. Finally, the personalized gait with different walking characteristics is generated by adjusting the condition information. Through experimental analysis, the correlation coefficient between the personalized gait generated by the method and the real personalized walking data is larger than 0.98, the average absolute deviation is less than 0.08 rad, the absolute deviation of the threshold is below 5%, and the gait stability criterion results are within the stability interval. Experimental results show that the method can effectively generate personalized gait corresponding to different walking characteristics. Compared with similar researches, the walking features are more comprehensive and have better integrity.