提出了一种保持视觉稳定性的增强现实注册算法，特征点在视频中的初匹配基于随机树检索统计得到，初步匹配的特征点采用半几何限制结合单应矩阵迭代的方法进行错误匹配排除，随后采用光流法跟踪特征点，最后根据共面POSIT算法计算出物体世界坐标系到摄像机坐标系的变换矩阵。由于采用了特征点跟踪方法，在一定程度上有效的防止了虚拟物体在视觉上的抖动。

A visual robust augmented reality (AR) register algorithm based on retrieving and tracking pre-defined image feature points is given in this paper. Feature points are retrieved from video by voting from random tree. False matching points are removed by semi local constraint paradigm with iterative homograph constraint. Optical flow is used to track the pre-defined tracking points. Rotation and translation matrix from world coordinate to camera coordinate are obtained with coplanar POSIT algorithm. For tracking method being used in our algorithm, dithering problem is effectively solved in real-time experience.