Uma Revisão Sobre as Abordagens de Particionamento e de Seleção de Regiões Relevantes para Registro de Nuvens de Pontos 3D

Abstract

The registration of 3D point clouds consists of correcting the pose of two or more partial views of objects or scenes captured in different perspectives, which finds applications in robotics and autonomous navigation. In that domain, a relevant issue involves the complexity of the alignment operation, and some methods seek to reduce it by selecting subsets of relevant points from the original clouds. In order to understand the state-of-the-art in this context, a review of the methods published between 2017 and 2021 was carried out. Eleven methods were evaluated in relation to the following questions: approach, way of selecting points of interest, type of application, technical limitations and evaluation methodology of the registration. In order to support research in this domain, the protocol adopted and the results of this evaluation were carefully documented. The review revealed that the approaches found have similar purposes, such as dense cloud alignment, recording of captures by heterogeneous sensors, merging of noisy clouds and optimization of approaches from the literature.