基于点云数据的分割方法综述

Segmentation methods for point cloud: a survey

点云是一种能够完整表达场景信息的重要数据格式。近年来,对于点云的探索引起了越来越多研究人员的关注,并且迅速在计算机视觉、自动驾驶和机器人等许多领域得到了广泛应用。但是,由于点云独特的数据形式,使用深度神经网络处理点云时面临着独特挑战,因此基于点云的深度学习方法仍处于起步阶段。最近,利用点云处理分割任务出现了许多优秀的方法。为了激发未来对点云研究的深入发展,本文对点云深度学习方法的最新进展进行回顾,涵盖了三个主要任务,包括点云语义分割、点云实例分割以及常用的三维图像数据集,对其中处理点云的深度学习经典方法展开对比分析,提供多种方法在不同数据集上的比较结果,并且提出了一些观点和未来研究方向。

Point cloud is an important data that represents complete information of the scene. Point cloud learning has lately attracted increasing attention due to its wide applications in many areas, such as computer vision, autonomous driving, and robotics. However, due to the unique characteristic, deep learning on point clouds is still in its infancy. Recently, deep learning on point clouds has become even thriving, with numerous methods being proposed to address different problems in this area. To stimulate future research,a comprehensive overview of recent progress in deep learning methods for point clouds segmentation tasks is provided in this paper. Three major tasks are stated,including point clouds semantic segmentation, point cloud instance segmentation and 3 D image databases. It is also presented comparative results on several publicly available datasets and analyze the classic deep learning-based methods, together with insightful observations and inspiring future research directions.