查看文章

Automatic building information extraction is an active research field in photogrammetry and remote sensing. However, most methods are proposed for supervised segmentation of point clouds or images, which can only capture limited building texture or geometric information, resulting in the obtained buildings being often fragmented. Therefore, we propose a bidirectionally greedy framework to extract spatial-continuous, geometry-complete, fine-textured 3D building models from large-scale 3D meshes captured by airborne in an unsupervised manner. The framework consists of two key steps in opposite directions, namely greedy culling and greedy recovery. Greedy culling will maximize the removal of non-building primitives based on geometric and textural features. Greedy recovery is designed to maximize the detection of building primitives that are mistakenly removed by the greedy culling, by utilizing topological …