pictures and more descriptions to come. Stay tuned! Note that this is simply an intermediary document while we work on a real documentation format. See/contribute to #516 for documentation progress.
Resize
coming soon...
OpenSfM
OpenSfM is a Mapillary Structure from Motion library that performs feature extraction, matching, and point cloud generation. This example from Mapillary shows some of the results of OpenSfM.
coming soon...
Matching
coming soon...
Point Cloud Generation
Once the sparse point cloud of tie-points is generated, ODM then uses a process called Multi-View Stereo. The algorithm is described in this paper:
Shen, S. “Accurate Multiple View 3D Reconstruction Using Patch-Based Stereo for Large-Scale Scenes.” IEEE Transactions on Image Processing 22, no. 5 (May 2013): 1901–14. doi:10.1109/TIP.2013.2237921.
This process creates depth maps for each image then merges them into a single, dense point cloud.
Meshing
The meshing algorithm is a Poisson Surface Reconstruction
Texturing
We use Multi-View Stereo for textuing the mesh output. Their website with plenty of information can be found here.
Georeferencing
Up until this point none of the outputs have any reference to the real world. In this process we extract the EXIF or GCP data into UTM coordinates and apply those to the 3D products (point cloud, mesh, textured mesh).
Orthophoto Generation
coming soon...