关键词: 局部跟踪, 加速稳健特征（SURF）特征提取, 光流法, 单应性矩阵, 三维注册

Abstract: A tracking and registration method based on SURF and optical flow in the AR is proposed. The moving object region is tracked by the optical flow method. This paper extracts and matches the features of the region with SURF because the algorithm maintains advantages of affine invariance, scale invariance and fast computing speed. The final registration matrix is calculated using the homographies between adjacent frames. This method reduces the computational time deeply while the accuracy of the registration unchanges. Convinced by experimental results, this approach can work at almost real time and achieves accurate registration, and can keep a good robustness in some environment changes.

Key words: region-based tracking, Speeded Up Robust Features（SURF） feature extraction, optical flow method, homography, 3D-registration