An Efficient Plane-Segmentation Method for Indoor Point Clouds Based on Countability of Saliency Directions

This paper proposes an efficient approach for the plane segmentation of indoor and corridor scenes. Specifically, proposed method first uses voxels to pre-segment scene establishes topological relationship between neighboring voxels. The voxel normal vectors are projected onto surface a Gaussian sphere based on corresponding directions achieve fast grouping using variant K-means approach. To improve integration, we propose releasing points from specified establishing second-order relationships different primitives. We then introduce global energy-optimization strategy that considers unity pairwise potentials while including high-order sequences over-segmentation problem. Three benchmark methods introduced evaluate properties by ISPRS datasets self-collected in-house. results our experiments comparisons indicate can return reliable with precision over 72% even low-cost sensor, provide best performances in terms recall rate compared methods.