三维点云孔洞修复方法综述

Review of three-dimensional point cloud completion methods

针对点云数据缺失问题,该文综合国内外大量点云修复技术研究。三维模型构建在自动驾驶、逆向工程领域中发挥越来越大的作用,三维点云数据是其中的重要数据源。利用三维激光扫描设备,可以高效、准确、实时的获取被测物体表面三维空间坐标。但是由于模型物体遮挡或者环境等原因,不可避免的会出现点云缺失的状况,这会对物体三维重建等后续处理造成一定的影响。然而,在三维点云孔洞修复方面还缺少比较系统完善的综述。本文从基于几何、基于模型检索、基于深度学习3个方面对当前主流的对修复技术进行了综合分析。文章对3种修复方法进行了概括,总结现有各种技术修复方法的优劣,同时展望了未来的发展趋势。

Aiming at the problem of point cloud losing,this article delivered a comprehensive review of a large mount of both domestic and aboard researches of point cloud completion.Structuring the 3Dmodel was playing a more and more important role in the field of autonomous driving and reverse engineering,which the point cloud was one of the most important data sources.Using the 3DLiDAR scanner,the 3Dcoordinates of the objects’surface could be obtained effectively,accurately and in real time.However,due to the occlusion,environmental elements and so on,point cloud losing occured inevitably,which influenced the result of 3D reconstruction and other following data processing process.Also,there was a lack of comprehensive and systemic review of repairing 3Dpoint cloud losing.Aiming at point cloud completion,the state-of-art completion methods were analyzed comprehensively base on three aspects of geometry-based,alignment-based and Learning-based approaches.In this work,we summarized the advantages and shortages of the 3kinds of repairing methods upon.At the same time,we also forecasted the tendency of future works.