A 3D Point Cloud Segmentation Method Based on Local Convexity and Dimension Features

Segmentation of 3D point clouds is an essential part of automatic tasks, such as object classification, recognition, and localization. The segmentation results pose a direct impact on the further processing. In this paper, we present an improved region-growing algorithm based on local convexity and dimension features for 3D point clouds segmentation. The point clouds on tabletop is removed from the original dataset by using RANSAC algorithm. Then the seed point and growing rules arc set according to the local convexity and dimension features. Our method can reduce the uncorrect segmentation to some extent, and reduce the impact from the selection of seed points on the segmentation results. Expoiments are provided to demonstrate that the proposed algorithm outperforms the traditional region-growing one from the perspective of segmenting the adjacent objects.