RESEARCH ON IMPROVED REGION GROWING POINT CLOUD ALGORITHM

• DOI: 10.5194/isprs-archives-xlii-3-w10-153-2020
• Publication Date: 2020-02-07T20:50:21Z
• Authors: C. L. Kang; F. Wang; M. M. Zong; Y. Cheng; T. N. Lu
• Keywords: Region growing; Segmentation-based object categorization

Abstract. The effective segmentation of point clouds is a prerequisite for surface reconstruction, blind spot repair, and so on. Among them, regional growth widely used due to its simple easy implement algorithm. However, the traditional algorithm often causes problems such as over-segmentation or voiding result instability local features cloud unreasonable selection initial seed nodes. In view above shortcomings, this paper proposes an improved region growing Firstly, by calculating Gaussian curvature average data sorting setting minimum node, total number clusters reduced, quality classification improved. Secondly, criterion determined combining normal angles. Finally, according shape characteristics preliminary results, each threshold adjusted determined, optimized.The experimental results show that compared with algorithm, method can not only reduce regions, but also segment quickly effectively, solve caused method. Problems stability improve accuracy segmentation.