A New Method for Planar Semantic Segmentation of Point Clouds in Indoor Scenes

The development of laser measurement technology has shifted the focus of 3D reconstruction from outdoor to indoor environments, leading to an increasing demand for creating precise indoor 3D semantic models. Semantic segmentation of indoor point cloud planes plays a crucial role in the process of indoor scene reconstruction. In this paper, we propose a method to build a plane model with vertical spatial coordinate axes to determine whether a point belongs to an inner group. We fit the indoor point cloud, remove a large number of clutter points, and perform an initial segmentation of the point cloud planes. We then utilize region growing algorithms and least squares method to extract parameters for ceiling, floor, and wall planes, and finally conduct semantic labeling. By comparing our method with existing approaches, we demonstrate that our approach effectively avoids the impact of indoor clutter while retaining the original point cloud features, leading to improved efficiency in plane segmentation and avoiding false positives.