针对球形机器人定位问题,提出了基于立体视觉的球形机器人定位方法.通过双目相机采集环境图像序列,提取Shi-Tomasi特征点,计算尺度不变特征变换(SIFT)特征描述符,并利用欧氏距离进行立体匹配;通过KLT算法进行特征点跟踪;采用解析法求解机器人在前后帧图像之间的位姿变化量;同时采用特征点筛选、RANSAC算法和卡尔曼滤波等方法,提高运动估计的准确性和鲁棒性.实验结果验证了所提出方法的可行性. In order to solve the problem of spherical robot positioning, a spherical robot positioning method based on stereo vision is proposed. By collecting the sequence of environmental images through a binocular camera, Shi-Tomasi feature points are extracted and SIFT descriptors are calculated. The distance between them is used to make stereo matching; the feature points are tracked by KLT algorithm; the pose of the robot is solved by the analytical method; the feature points, RANSAC algorithm and Kalman filter are used to improve the accuracy of motion estimation Sex and robustness.The experimental results verify the feasibility of the proposed method.