为了精确识别人手姿态用于机器手控制,文中讨论一种基于计算机视觉的手势立体姿态检测的新方法。该方法利用多目视觉之间的配合,有效地解决了检测识别时手部关键节点互相遮挡的关键问题。采用了自适应阈值分割、harris角点检测、外极线约束等几种合理、有效的点的匹配算法,对不同摄像机拍摄到的手部关键节点进行排序,这样不仅解决了特征点的位置归属问题,也解决了图像与对应点的匹配问题。通过实验表明:对手部关键节点的检测达到了满意的结果。

The detection method of hand gesture based on computer stereovision theory is discussed, the method is used to control the robot accurately. The sheltering problem of key point in the fingers is solved effectively by using the multi-camera system method. Some constraints and several reasonable and effective algorithms of matching, such as self-adapted threshold, harris corner detection and epipolar constraint, are used to sort and distinguish the key points on fingers in the different camera pictures. Because of using these methods, not only the location of the key points, but also the matching problem of them are solved. The experimental results show that detection of hand gesture are satisfactory.