基于人体动作姿态识别的机器人仿人运动*

以关节式机器人为对象，进行机器人仿人运动研究。从人体动作姿态识别、人-机动作映射、机器人运动控制等方面，详细阐述机器人仿人运动算法。提出人体动作姿态识别方法，利用Kinect传感器捕获人体运作的关节点位置信息，在建立人体基准坐标系的基础上，为了得到描述肩、肘运动的动作信息，计算人体手臂动作的关节角度，实现人体动作姿态的识别。在分析人体肩、肘等关节和机器人机构差异性的基础上，建立人体手臂与四自由度机械手臂的人-机动作映射规则。针对机器人自由度较少，无法完全复现人体运动的情形，分析、比较不同控制策略的优缺点和适用性，寻求适合机器人操作的复现控制策略。关节式机器人接收运动控制指令，执行相应的关节运动，从而实现机器人仿人运动。相关试验验证了人体动作姿态识别和机器人仿人运动控制算法的有效性。研究成果对于提高机器人控制和操作的简单易用性、提高人机交互能力具有借鉴意义，对于扩展机器人应用领域具有实践意义。

Humanoid Motion of Manipulator that Based on Human-posture Recognition

Humanoid motion of manipulator is studied. The problem of human-posture recognition, human-robot motion mapping and humanoid motion control are summarized. A Kinect is used to capture the human body movement. In the coordinate system on the human body, the joint angles of the human arm are calculated to recognize the motion of shoulder joint and elbow joint. By analyzing the difference between the motion of a human arm and that of a four-degree manipulator, the human-robot motion mapping is developed. To deal with the problem of limitation in motion imitation, caused by robot lower-mobility, three method of humanoid control is comparatively analyzed. And the method to maximize the range of motion is selected to control the manipulator. The manipulator would execute the action, according to the received motion instruction, to realize the humanoid motion. The validity of this approach is proved by several experiments. The method proposed will help to make the manipulator control simple and easy, help to improve the ability of human-machine interaction, and help to expand the application of the manipulators.