- Would like to try out the Hough algorithm to see how it compares in customisation, robustness to outliers
- Fix the area estimation. Possibilities:
- pcd triangulation for a 2d polygon in 3d space
- Using vertical planes to clip roof planes
- Planes that are on similar enough should be grouped if the are within a certain area (saw the algorithm split one flat section into two). Probably is a way to avoid this happening in the core algorithm though
- Tweaking the figures on the algorithm to match the specific use case is important. i.e. Ransac threshold to within the flatness of a roof + lidar error margins.
- Preprocessing the point cloud seems like a very powerful tool. Gives the main algorithm the best chance to shine
- PLY - Polygon file format - format to store 3D data
- OBJ - format to store 3D data, supports lines/trimeshes geometry, materials, textures
- gltf/glb - supports animation, scene graph, textures
- Open3d - library for 3D data processing
- Trimesh - library for maniplating tri based meshes
- PDAL - library for manipulating point cloud data
- KD-Tree statiscical outlier removal (https://www.mdpi.com/1424-8220/21/11/3703/htm)
- least squares
- RANSAC
- Hough
- Principal component analysis
- Singular value decomposition
- Covariance matrix