A coarse registration method of range image based on SIFT

Abstract

A novel method for the coarse registration of range images is proposed. This approach is based on texture-feature recognition. As the development of optical digitizing technique, it is now able to acquire the range images and associated texture images sequentially or simultaneously. It's possible to identify the range feature points through texture feature points. Scale Invariant Feature Transform (SIFT) is an efficient method for texture feature generation. SIFT transforms texture image into a large collection of local feature vectors, each of which is invariant to image scaling, translation, and rotation. The mismatched correspondence pairs can be discarded using random sample consensus algorithm based on epipolar geometry constraint. We select more than three well-registered texture-feature pairs, with which we could find the associated range-feature pairs of the range images. Initial pose estimation of the two involved range images can be computed by these range pairs, and the fine registration is implemented using iterative closest point (ICP) algorithm. Our approach utilizes the texture information to register the range images, leading to a technique that can be automatically performed while the influence of 3D noise can be avoided. The experiment results demonstrate that the proposed approach is efficient and robust for the registration of multiple range images.