An improved Iterative Closest Point (ICP) method based on the boundary feature points of the point cloud is proposed to improve the efficiency and accuracy of point cloud data registration in reverse engineering fields. First, an initial registration method based on the boundary feature points of point cloud is proposed. The method partitions the minimum bounding box of point cloud with grids in a 3D space, and sets up the space grid model. Then, it applies boundary seed grid recognition and growth algorithms to extract feature points from the boundary of point cloud, and works out the transformation matrix using Singular Value Decomposition (SVD) method to get the results of initial registration. Furthermore, an improved ICP accurate registration method is presented. It weighs the corresponding points of the point cloud, eliminates the points whose weight is larger than the threshold, and introduces M-estimation to the objective function to eliminate the abnormal points. Finally, the point cloud is accurately registered by the improved ICP method on the basis of initial registration. Compared with original ICP method, the improved ICP method increases the efficiency by more than 70 percent and reduces the error to 0.02 percent. The experiment results indicate that the method proposed in this paper improves the efficiency and accuracy of point cloud registration greatly.