Weighted Symmetric ICP Point Cloud Alignment Algorithm Based on Local Feature Description

Abstract

Point cloud alignment based on local shape features is widely adopted due to its effectiveness and robustness, and most current local feature-based alignment algorithms are difficult to meet satisfactory performance in terms of alignment accuracy, robustness and time efficiency at the same time. In this paper, we propose a new weighted symmetric Iteration Nearest Point (ICP) point cloud alignment algorithm based on local feature description, which firstly encodes the spatial and deviation angle information of point clouds based on robust reference axes to form an efficient and robust feature descriptor, and then establishes the correspondence between point clouds based on local features, and combines with a 1-point sample consensus algorithm to obtain the initial alignment results. A weight function is constructed based on the residual distance between corresponding points, and the weighted symmetric ICP algorithm is designed using the symmetric point-to-plane distance as the optimization criterion to refine the initial alignment results. Experiments on public datasets show that our method has better alignment accuracy, robustness, and higher alignment efficiency than the currently popular local feature-based point cloud alignment algorithms.

Keywords: point cloud, local feature descriptors, alignment, transformation estimation, ICP