面向翻书机器人的手势控制指令的识别算法

Gesture Control Comand Recognition Algorithm for the Book Turning Robot

所控制的翻书机器人是一种辅助特殊人群阅读的服务机器人,对其提出一种基于表观特征的手势控制指令识别算法.从Kinect(体感外设)获取的手部深度图像中利用最小深度点分割手势区域,用轨迹圆提取指尖个数及夹角等表观特征并通过决策树分类,同时利用手和肩膀的骨骼数据及其运动轨迹判定翻页方向.经过多次试验,手势控制指令识别率达到98.4%,该算法无需训练样本也能在复杂环境下快速准确地识别手势控制指令,完全可以应用于辅助特殊人群阅读时的翻书控制.

The book turning robot to be controlled is a kind of service robot to assist the disabled people to read. A gesture control command recognition algorithm based on appearance features was designed for the book turning robot. The method of the minimum depth pixel was used to separate the gesture region from depth images of hands captured by a Kinect（motion-sensor）. Then circle trajectories were drawn to extract appearance features of gestures such as numbers of fingertips and angles between fingertips, which were applied in the decision tree to classify the gestures. Meanwhile the skeleton data and trajectories of hands and shoulders were used to determine the direction of turning books. Through lots of experiments, the result shows that the recognition accuracy reaches 98. 4%. The algorithm can recognize the gesture control commands quickly and accurately without training samples in complex background environment, and it can be applied in assisting the disabled people to read in practice.