Vision-based motion recognition of the hexapod for autonomous assistance

In a cooperation style in which a human operates one robot and the other autonomous robots assist that robot, the autonomous robots must be able to recognize the motion of the human operated robot in real time. Vision is the most useful sensor for this purpose. By using vision, the assistant robot recognizes the current target object and actions of the human operated robot to the target. For the working robots, we use two hexapod robots which are able to utilize two forelegs as arms in order to manipulate objects. Joint motion of each leg and occlusion by another leg makes it difficult to recognize and track the motion of the hexapod robot. We propose a method for the recognition and tracking of the 3D position and posture of the hexapod. Through the use of 2D image matching, the body of the hexapod, several legs, and toes of the legs can be recognized in turn. The positions of the body and toes are measured by binocular-stereo, and the angles of the legs can then be obtained, which are used for motion recognition. We propose a motion recognition mechanism using an eigenspace method which can reduce the dimensions necessary for matching. The effectiveness of this method is tested through an experiment using two hexapod robots