Exploring motion control for three-dimensional object recognition based on eye-in-hand visual servoing

Abstract

This paper proposes a visual servo control method for exploring motion of eye-in-hand robot to recognize a three-dimensional object. The CCD camera attached at the manipulator tip captures the images of the target object, and then the robot varies its own posture with referring the position transitions in the images of the feature points on the object. Our approach for optimal motion is based on the strategy in which the position estimation precision of the feature points in the captured images gets improved when the feature points move widely, and the optimal robot motion provides the widest transitions of the feature points per unit robot motion, so that the robot should be controlled to purse such motion. In the proposed method, the primary feature point is made located at the center of the image, while the other points are moved for achieving the optimal transition in the image. In addition, this paper describes the way to reduce calculation effort for the proposed method. The numerical and physical experiments have been executed with use of 3-DOF planar manipulator. The validity of our approach is confirmed in the several experimental results.