A new approach to self-calibration of hand-eye vision systems

This paper presents a new approach to self-calibration of the robotics hand-eye relations. The setup we use is the wrist-mounted camera for a robot manipulator. The camera motion can be controlled with the motion of the manipulator. According to the special designed motion of the camera, the whole calibration process can be divided into two steps. First, let the end of the manipulator do four linearly independent translational motions. After three of them, the rotational relation of the hand-eye system can be solved. After the manipulator completes the translations, the camera intrinsic parameters and the rotational matrix can be solved as an entity. The depth information of the point in the scene can be obtained as well. Then, the translational vector of the hand-eye systems can be solved after a pure rotational motion of the manipulator and an imaginary rotation of the camera frame. This method requires no reference object and only needs two characteristic points in the scene. Besides, it can accomplish the self-calibration of the camera intrinsic parameters and the rotational matrix as a whole, which reduce the computation errors and cost.