The Semi-dense ICP Algorithm Based on the SIFT Feature Points Neighborhood

Abstract

Abstract The ICP algorithm based on RGB-D images is one of the most widely used and the most effective point cloud registration algorithms. How to improve the registration accuracy, registration speed and robustness of the ICP algorithm has become an important research hotspot at present. However, existing algorithms have low computational efficiency under large-scale point cloud data, and when the overlap between point clouds is small, the registration accuracy is low. To deal with these problems, this paper proposes a semi-dense ICP algorithm based on SIFT feature points neighbourhood. First, the algorithm selects the SIFT feature point matching algorithm as the rough registration method of ICP, because the SIFT feature points have the advantage of rotation invariance, illumination invariance, and scale invariance; then based on the SIFT feature points, the algorithm selecting its neighbourhood as the matching range. The experimental result shows that this algorithm has the better registration accuracy and computational efficiency.