Abstract: In order to solve the problems with the traditional 3D model registration methods, such as the higher requirements for the initial locations of the two point clouds and the lower accuracy of model registration sometimes, a 3D model registration method based on SURF feature extraction of the two 3D model projection images is proposed. Firstly, the range of the projection image is determined by the 3D model data, and the model grid which each projection image pixel belongs to is deduced. Thus the two-dimensional projection image can be calculated by the mapping relationship from the projected image to the texture image. Then, the SURF feature points and the eigenvalues are calculated for the two projection images respectively. The feature matching is performed according to the SURF eigenvalues. Next, the feature point pairs in the two 3D models are extracted through the mapping relations between the 3D models’ grid vertices and the projection image pixels. Finally, the two 3D model registration is implemented through the transformation matrix which is calculated from the matched feature point pairs in the two 3D models. The experimental results show that this method can improve the registration accuracy effectively and has good robustness under the condition of little change in registration time.

Key words:  3D scanning, projection image, SURF feature, image registration, 3D registration