A robust registration algorithm for 3D point cloud data with scale stretches and outliers

Abstract

Abstract Generally, the point cloud data obtained by 3D scanner cannot express the overall information of objects at a single time due to the limited field of view, so registration algorithm is needed to obtain the complete information. For the registration of point cloud data sets with large scale stretches and outliers, this paper proposed a new iterative nearest point algorithm named Improved-ICP. In the case of isotropic point extension, the iterative reweighted least squares method is constructed by incorporating a weight function to the minimum error function. In calculation, the weight function is equivalent to increasing the weight the point pairs, and the weights are obtained by an M-estimation criterion. In the paper, the initial registration was optimized to get global convergence for the algorithm, and compared the Improved-ICP algorithm with the ICP algorithm and the Scale-ICP algorithm to verify the effectiveness of the algorithm. To demonstrate the robustness of the Improved-ICP algorithm, we performed several comparative experiments with different scales and different noise between the Scale-ICP and the proposed algorithm in the presence of abnormal points. Experiment results illustrate the Improved-ICP algorithm has high accuracy and strong robustness to scale and abnormal points.