基于空间上下文关联的车载点云聚类方法

A Spatial Context-Based Clustering Approach for Vehicle-Borne Laser Scanning Data

点云聚类是移动车载点云数据处理和信息提取的重要组成部分,同时也是实现地物自动识别的前提和关键环节。针对密度聚类中仅考虑点间空间距离聚类造成的过分割或分割不足的问题,提出了一种基于空间上下文关联的城市街景车载点云数据聚类方法,以超体素为对象,分析对象的特征及相互间的空间上下文关联,在综合多因素权值的基础上进行自适应聚类。通过两组数据的实验结果表明,该方法有效改善了车载点云数据聚类结果,提高了分类的效率和可靠性。

The clustering of laser scanning data, being the premise and key step to identify objects automatically, is an important component of vehicle-borne laser scanning data processing and information extraction. Due to the problem of over-segmentation or insufficient segmentation caused by density-based clustering only considering space distance between points in clustering, this paper presents a spatial context-based clustering approach for vehicle-bornelaser scanning data of streetview, which takes supervoxels as objects, analyses the characteristics and spatial context of the objects, and clusters by considering comprehensive multi-factor weights. The results of the two sets of laser scanning data show that this clustering method successfully improves the clustering result of vehicle-borne laser scanning data, and enhances the efficiency and reliability of the classification.