基于人体动作识别的类人机器人动作模仿

Humanoid Robot Motion Imitation Based on Human Posture Recognition

基于模仿学习以及人机交互技术,借助Kinect深度相机传感器对类人机器人上半身动作模仿问题进行了研究。首先,将改进的D-H模型应用于NAO机器人的手臂完成了手臂运动学模型的精确建立及求解,解决了传统建模方法两相邻关节平行时出现的奇异性问题。其次,提出了一种改进的基于深度图像的手势识别算法,完成了对于示教者手势的判定及模仿,与传统基于彩色图像的手势识别相比,不受光照影响的同时提升了识别准确率,改进算法的平均识别准确率达到96.2%。最后将NAO机器人作为试验平台的实验表明:NAO机器人对于示教者上半身动作的实时在线模仿运动轨迹平滑且稳定,并且在抓取实验中也显现出了较好的准确性。

Based on imitation learning and human-computer interaction technology,Kinect depth camera sensor is used to study the upper body motion imitation of humanoid robots. Firstly,the modified D-H model is applied to the arms of the NAO robot to complete the accurate establishment and solution of the kinematics model of the arms,and solve the singularity problem when two adjacent joints are parallel. Secondly,an improved gesture recognition algorithm based on depth image is proposed,which judges and imitates the gestures of the teacher. Compared with the traditional gesture recognition based on color image,it is not affected by light and improves recognition accuracy of the system and the average recognition accuracy of the improved algorithm reaches 96. 2%. Finally,the experiments using the NAO robot as a test platform show that the system enables the NAO robot to simulate the upper body movements of the teacher in real time,with smooth and stable motion trajectory,and it also shows good accuracy in the grasping experiment.