提出了一种面向类人机器人的人体动作视觉感知算法,提高了利用Kinect作为视觉输入设备捕捉到的人体动作数据的精度.首先,通过逆运动学方程将捕捉到的关节位移信息转换成角度信息.然后,以角速度和角加速度的变化为依据,将长时间的运动自动分割成独立片段,并用相关向量机原理估计出高精度的角度轨迹.最后,用角度轨迹的空间相似性、时间相似性、平滑度等指标对该算法进行了评估,并在NAO机器人平台上对算法处理后的动作进行了实验验证.实验结果表明,该算法有效提高了动作感知的时空相似性和轨迹平滑度,为高精度的动作模仿奠定了基础. This paper proposes a human action visual perception algorithm for humanoid robot and improves the accuracy of human motion data captured by Kinect as a visual input device.Firstly, the captured joint displacement information is converted into angular information by inverse kinematics equation Then, based on the changes of angular velocity and angular acceleration, the motion of long time is automatically divided into independent segments and the high accuracy angular trajectory is estimated by using the theory of correlation vector machine.Finally, the spatial similarity of time trajectory, time similarity , Smoothness and other indicators of the algorithm to evaluate the algorithm and experiments on the NAO robot platform after the action of the algorithm has been experimentally verified.The experimental results show that the algorithm effectively improves the spatial and temporal similarity of motion perception and trajectory smoothness High-precision motion simulation laid the foundation.