基于视觉的动态手势识别及其在仿人机器人交互中的应用

手势识别是人和机器人交互中的重要组成部分，本文针对双目视觉系统SFBinoeye实 现了基于光流PCA(主分量分析)和DTW(动态时间规整)的命令手势识别，用以控制仿人机器人 SFHR的手臂运动．利用块相关算法计算光流，并通过主分量分析得到降维的连续投影系数， 与手掌区域的质心位置组合为混合特征向量．针对DTW定义了新的加权距离测度，并用它对 手势进行匹配识别．针对9个手势训练和识别，识别率达到92.4%，并成功地应用于机器人的 手臂控制中．

VISION BASED DYNAMIC GESTURE RECOGNTION ANDITS APPLICATION IN HUMAN-HUMANOID ROBOT INTERACTION

Gesture recognition is an important component in human robot interaction. This paper realized a command gesture recognition architecture based on optical flow PCA and DTW for a stereo vision system SFBinoeye. The recognized gesture command was used to control humanoid robot SFHR's arm to move correspondingly. Optical flow is computed using block relative algorithm, which is then handled with PCA to get continuous projection coefficients with lower dimension. The projection coefficients are combined with the center position of hand as complex feature vector. A new weighted distance is defined in DTW, which is used for gesture pattern match. The proposed gesture recognition architecture is tested on 9 gestures with of recognition 92.4% rate. This architecture is successfully applied in the control of robot arm.