手—眼系统自标定技术

A SELF-CALIBRATION TECHNIQUE FOR HAND/EYE SETUP

手-眼关系的建立是校准主动视觉系统的核心内容。本文提出的自标定技术是通过摄像机在三维空间内作平移运动,实现自标定。与以往方法相比,它不要求摄像机作多组相互正交的平移运动,而是控制摄像机运动,通过所采图像数据,直接标定出手-眼几何关系;同时控制摄像机相对于初始位置作三次线性独立的平移运动,即可线性求解出摄像机内参数。理论证明,解存在且唯一。为克服所采图像数据误差,通过校正算法预先校正,数值模拟表明,该方法具有较强的鲁棒性,最后给出真实图像的实验结果。

Robotic hand/eye calibration is one of fundamental issues in computer vision. In this paper, a new hand/eye self-calibration technique is proposed. Unlike the former methods, our approach doesn't require camera of the ability to translate along two orthogonal directions many times. By controlling a camera to translate, hand/eye geometry ran be calibrated directly according to images. And by controlling a camera to translate three times and translation vectors are linear independent, the camera intrinsic parmeters can be solved through pseudo-inverse matrix theory. The analysis shows that the solution is existing and unique, and the important result has been proved. In order to overcome errors inducted from matched image datum, these matched image datum must be pre-processed by revising algorithm. Tests with synthetic date and real images indicate that the algorithm is robust and practical.