Regularity-constrained point cloud reconstruction of building models via global alignment

Recovering the shape of an object from a 3D point cloud is a challenging task due to the complex shapes and diverse requirements of different tasks. For buildings in urban scenes, maintaining geometric constraints such as parallelism, perpendicularity, symmetry, and coplanarity is important during the model fitting process. In this paper, we present a regularity-constrained point cloud reconstruction framework that comprises primitive initialization, constraint construction, and global optimization. Our approach first performs normal optimization and plane segmentation on the input point cloud. We then compute the global reference directions to set the target normal for each plane to construct the constraints. Finally, we obtain the model by globally optimizing the position and orientation of the planes while considering the constraints. Our experimental results demonstrate that our proposed algorithm not only strictly enforces geometric constraints but also closely fits the input point cloud. Furthermore, our framework outperforms state-of-the-art methods in terms of shape recovery and constraint maintenance, as demonstrated by comparative evaluations.