How to use images with different zoom levels for incremental reconstruction?

mapillary / OpenSfM

Open source Structure-from-Motion pipeline

https://www.opensfm.org/

BSD 2-Clause "Simplified" License

3.41k stars 861 forks source link

How to use images with different zoom levels for incremental reconstruction? #881

Open n-shubham opened 2 years ago

n-shubham commented 2 years ago

Hi! I wanted to know how images with different (varying) zoom levels can be used to create a reconstruction? Using incremental reconstruction, it ends up reconstructing a curved surface for flat land. Is there some parameter that can be used to fix this?

Additionally, could you point me to references that I can read to understand how both incremental and triangulation reconstruction algorithms works in some detail? The documentation is superficial.

YanNoun commented 2 years ago

Hi @n-shubham,

Yes, you can use different zoom level, as long as the focal length is properly stored in the image's EXIF. If you end up with a "bowl effect", it might be :

Combination of bad geometry (flat land + nadir capture)
Incorrect focal length or projection in the .exif files or camera_models.json files (fisheye versus perspective)

Let me know if that helps,

Yann