Orthomosaic generation and photo stitching are different methods for creating orthophotos, but the advantages of orthorectification make it the preferred method.
The photo stitching method glues images together and requires a low number of matches/keypoints (less than 100). It works well only if the terrain is perfectly flat. If the terrain is not flat, it can lead to artifacts where objects visible in several pictures do not align well. This kind of errors are accumulated over the whole dataset and therefore it is recommended to use photo stitching only for small datasets. Most distances are not preserved, which leads to inaccurate measurements.
Orthomosaic (true orthophoto) generation
The orthomosaic (true orthophoto) generation is based on orthorectification. This method removes the perspective distortions from the images using the DSM. A high number of matches/keypoints (more than 1000) is required to generate the model. This method handles all types of terrain, as well as large datasets. Distances are preserved, and therefore, the orthomosaic can be used for measurements.
The following steps are performed to generate the orthomosaic:
- Input: Images with perspective (for example, facades are visible, roofs do not have the correct size as the scale is not preserved).
- Calibrate images and compute 3D and 2.5D model.
- Project images on the 2.5D model to generate the orthomosaic.
- Output: Orthomosaic (similar to satellite imagery, facades are not visible, roofs have the correct size).
The orthomosaic corrects the following:
- The perspective of the camera.
- Different scales based on the distance that each point of the object/ground has from the camera.