复杂场景点云数据分割研究综述

Review of Segmentation of Point Clouds in Complex Scenes

点云分割是三维场景理解的关键内容,在自动驾驶、机器人等人工智能领域有重要应用。首先,总结点云数据类型和数据特点。其次,追溯点云分割的发展历程,总结并评述了复杂场景中点云分割的代表性方法。点云分割按照分割结果是否含有语义信息可分为传统分割和语义分割。点云语义分割普遍采用了机器学习算法,根据特征提取方式可分为基于人工特征的方法和基于自动特征的方法。基于自动特征的点云语义分割方法普遍采用了深度学习框架,是当前研究的热点。最后,总结当前研究现状,展望复杂场景点云分割未来的发展趋势。

Point cloud segmentation is a key stepin understanding 3 D scenes,which widely used in automatic driving,robots and other artificial intelligence fields.Firstly,the data types and characteristics of point cloud are summarized.Secondly,it traces the development of point cloud segmentation,reviews the representative methods of point cloud segmentation in complex scenes.According to whether the segmentation results contain semantic information,point cloud segmentation methods can be divided into traditional segmentation methods and semantic ones.Semantic segmentation generally uses machine learning,and can be divided into artificial feature extraction methods and automatic feature extraction methods which based on deep learning,the hotspot of current research.Finally,it summarized the limitations of the current researches,and forecast the trend of point cloud segmentation in complex scenes.