多视图点云配准算法综述

The multi-view point cloud registration problem was investigated，and a comprehensive classification and summary of the research work related to multi-view point cloud registration in the past two decades were carried out． First，the concept of point cloud and multi-view point cloud registration were introduced． According to different registration tasks，multi-view point cloud registration methods were divided into two categories：multi-view coarse registration and multi-view fine registration．Specifically，multi-view coarse registration algorithms were further divided into two categories：spanning tree-based and shape growing based multi-view coarse registration． Multi-view fine registration algorithms were divided into four categories：point space-based，frame space-transformation-averaging-based，deep learning-based and optimization-based multi-view fine registration algorithms． The traits of each algorithm were properly discussed．Then，four multi-view point cloud registration benchmarks and popular evaluation metrics were introduced． Finally，a summary on the current research status of this field was given，the existing challenges were discussed，and potential future research directions were indicated.