Room partitioning in complex environments by supervoxel segmentation and anchor pixel linking

Abstract

This article presents an innovative method to automatically partition rooms in cluttered environments, which is a critical task for indoor reconstruction, navigation, as well as scene understanding. It involves two main phases, starting with the operations of morphological erosion and feature analysis on the projected supervoxels to highlight gaps and remove narrow passages. This is followed by a refinement phase, where the continuous and clean wall boundaries are connected in the occupancy evidence map under the constraint of orientation information. Finally, the individualization of rooms is achieved by inversely propagating the segmented result in image back to point cloud. Experimental results demonstrate that the proposed method outperforms mainstream approaches, particularly in challenging scenarios characterized by heavy occlusion, curved walls, multiple ceiling heights, or long corridors.