1) Use LiDAR point cloud to calculate three geometric features: flatness, normal vector distribution and elevation texture consistency.
2) At the same time, use the aerial orthophoto to calculate the color features——Normalized Difference Vegetation Index (NDVI)
3) Then the two types of features are combined to construct the energy function data item, and the digital surface model (DSM) and NDVI are integrated to construct the smoothing item, and the graph cut algorithm is used to optimize the initial building area.
4) Using the orthophoto color information within a certain range of the initial building edge, the idea of foreground and background segmentation is used to further optimize the building edge
Elevation on buildings (DSM)
The normal vector of the local point cloud of the building area will be concentrated in one (local flatness) or several directions
Gray level co-occurrence matrix consistency to describe this texture feature. The smaller the fTH, the worse the consistency, that is, the richer the texture, indicating that the grid is more likely to belong to the vegetation area