中文摘要：

构建了一种新型的满足服务机器人视觉任务需要的双目视觉系统，该系统是由一个云台（Pan／Tilt／Zoom，PTZ）摄像机和一个手眼系统组成的变基线双目配置，具有更大的视觉空间和灵活性。对该双目配置下实现目标定位的关键参数——双目结构参数的实时自标定方法进行了研究，提出了将大部分繁琐的坐标变换关系转移到离线环节标定的离线／在线二步标定算法。仿真实验表明，此方法可有效降低在线处理时对图像匹配点的数目要求，同时该双目系统对图像噪声和机械运动误差具有一定鲁棒性。

英文摘要：

A novel binocular vision system was established to meet the demands of vision tasks of service robots. In this vision system a Pan/Tilt/Zoom （FYZ） camera and a hand-eye system were integrated to construct a baseline-variable configuration and to obtain larger vision space and better flexibility. The method for real time self-calibration of the binocular structure parameters crucial to object location under this novel binocular configuration was studied. Then a two-step offline/online algorithm was proposed to calibrate the most part of fussy coordinate transformation at the offline stage. The simulation experiments showed that the proposed method could effectively reduce the number of matching points essential to online calculation, and the binocular vision system was robust enough to resist image noise and mechanical movement noise.