1. colmap coordinate system:
- a. The coordinates of the upper left corner are (0.5, 0.5) for bilinear interpolation
- b. The directions xyz of the coordinate system are red, green, blue, lower right and front
- c. The pose of colmap is of word2camera , which is different from sensor2world, which is commonly used in slam, so sometimes it needs to be converted :
– point P, camera is Pc, then: Pc = RP+t
– and for camera2world: P = RPc + C, C = -(R transpose)t - d. The origin of the world coordinate system is in the first frame , but not necessarily (0,0,0), due to the deviation of BA optimization
-e. The relative coordinates are the coordinates of the next frame relative to the previous frame, and the absolute coordinates are relative at the first frame of - f. Regarding the GPS coordinate system, WGS84 (GPS standard format, geocentric coordinate system, BLH latitude, longitude and earth height), UTM coordinate system (6-degree zone, zone = int(lon/6)+31), WGS84 to UTM :
– use epsg
– use proj4
– refer to OpenMVG
This part refers to official documents and online courses
2. Camera model
Look directly at the source code and recommend options . Generally speaking, a pinhole camera model that removes radial distortion is sufficient.
3. For the overall framework, refer to the official documentation :
Among them, SparseReconstruction contains three output files, and we use this to convert formats according to our own needs (such as accessing openMVS, the ps interface already exists). cameras.txt, images.txt, and points3D.txt:
