ICP algorithm is highly dependent on the initial value in frame-to-frame registration.To solve the problem,a new coarse-to-fine frame-to-frame registration algorithm is proposed.Taylor expansion algorithm is used to simplify the camera pose model.Then the coarse registration model is established using summation idea to improve the efficiency of coarse registration.Then,Singular Value Decomposition (SVD) is applied to obtain a coarse estimation of the camera pose,and ICP algorithm is utilized to improve the accuracy.Moreover,dual thresholds are preset in the stage of feature extraction and matching,by which both the coarse registration accuracy and the corresponding point amount of ICP are guaranteed.Contrast experiment was made.The result shows that:Compared with ICP and the improved algorithm RANSAC-ICP,the proposed algorithm can solve the above problem more efficiently,which can achieve equivalent accuracy as RANSAC-ICP algorithm with higher registration speed.