Exploring motion planning for 3D object recognition

In this paper, a new control scheme on image-based visual servoing is described. The proposed approach consists of two level strategies, one is in the motion planning and the other in the task planning. The method in the motion planning is based on the eigenvector derived from the product of the extended Jacobian. When the camera moves in the direction along the vector, the largest shift of feature points in the image is obtained so that the most effective motion of the camera is achieved, which contribute toward the accurate three-dimensional position detection. In the method in the task planning, the accurate three-dimensional position and orientation are detected, based on the structure analysis. In the analysis, the indices where enough surveillance has been achieved in the motion and so on are applied. The motion which consists of the motion planning and the task planning is called exploring motion, in which the features are recognized by traveling around all the target object.