Similarity Registration Algorithm of 3D Point Clouds Based on Improved ICP

For the similarity registration problem of 3D point clouds data with a large amount of noise and outliers, this paper proposes a similarity registration algorithm of 3D point clouds based on improved iterative closest point (ICP). Firstly, a similarity registration optimization model of 3D point clouds is established by introducing the pseudo-Huber loss function, to obtain accuracy similarity registration of 3D point clouds. Secondly, the initial scale factor is solved according to the scale consistency criterion. Thus, the similarity registration problem is transformed into the corresponding rigid registration problem. Finally, the rigid transformation is solved for each scale factor according to the scale factor interval. By using alternating iterations, the optimal rigid transformation and scale factor can be obtained. Experimental results show that the proposed algorithm can achieve high similarity registration accuracy by efficiently suppress the noise and outliers.