A hybrid top-down, bottom-up approach for 3D space parsing using dense RGB point clouds

Nowadays, despite the advanced developments in engineering, automatic large-scale point cloud processing is still one of the challenging topics in many applications. In this regard, segmentation of the indoor point clouds into partitioned spaces is highly demanded in building information modeling (BIM) and robotic society. This paper proposes a novel automatic hybrid top-down, bottom-up approach for the 3D space parsing in the building environment and inferring relations between spaces. The proposed method is based on applying a deep convolutional neural network (CNN) for semantic segmentation of main elements and the use of existing knowledge in the construction of buildings. Unlike the existing methods, the proposed approach does not require pre-knowledge about the space layout. The results of evaluating the proposed method on two datasets with different designs highlight the capability of the proposed approach in 3D space parsing, extracting wall footprints, and particularly finding the topological relation between them.