Automatic 3D reconstruction of indoor building point clouds under occlusion based on gestalt rules

The occlusion of complex building layouts and the high presence of structural elements are the key challenges for indoor building point clouds reconstruction. This paper presents a new Gestalt rules-based indoor building point clouds reconstruction method. In this method, different Gestalt rules are first constructed and then missing point clouds of different structural elements are gradually reconstructed. Our reconstruction work is carried out according to the process of ‘overall framework – structural surface – surface details’, including indoor wall reconstruction, surface hole filling, and door and window reconstruction. First, a method, combining Gestalt closure rule and multilevel hierarchical labeling, is proposed to reconstruct missing walls with arbitrary directions. Then, holes in the structural surface are filled using an image inpainting algorithm under a spatial relation constraint. Finally, doors of different types, states, and components, as well as windows of different types, are reconstructed using a multi-feature hierarchical combination method. To validate the effectiveness of our method, we selected several typical scenes from different datasets and compared them with the existing methods. Experiments show that our proposed method is effective in different types of structural components reconstruction in different indoor scenes, despite the presence of significant clutter and occlusion.