Hello!
If you like fisheye lenses, get a coffee and sit down!
This paper by Luca Perfetti et al. describes and addresses the use of fisheye lenses for photogrammetry. More specifically, it looks at how to design a survey in narrow spaces. In this post I summarize some of their contributions and ask if you agree with their solution
Fisheye lenses are interesting in confined spaces because they offer a large field of view, allowing for a good image footprint even if the distance camera-subject is short. But the down side is that the ground sampling distance (GSD) is not constant across the image. This page of Pix4D support states that Pix4D can deal with a GSD ratio up to 1:2: https://support.pix4d.com/hc/en-us/articles/202558979
Can I use this information to define my survey plan with a fisheye lens?
When facing a planar surface with a fisheye lens (for example a building façade, or a flat terrain if we consider nadir images), the GSD varies within the image for two reasons:
- Objects that are seen in the middle of the image are physically closer to the camera than objects that are seen on the edge of the image (this is simple geometry)
- The lens distortion increases the GSD on the edge of the image (reduces the apparent size of objects)
When adding up these two effects, we see that the GSD on the edge of the image may be far more than twice the GSD in the center of the image. Let’s consider that I try to map a building façade in a narrow street, following a line pattern (“general case” described here: https://support.pix4d.com/hc/en-us/articles/202557459#label2). How do I take into account the increase of GSD on the image edge? For example, if I select a 50% side overlap, the center of the second flight line is just at the edge of the images of the first flight line. In this case 50% of image 1 is also seen on image 2, but the area of the images that have a GSD in an acceptable ratio (ratio smaller than 1:2) is less than 50%.
The solution suggested by the paper mentioned above is to crop the images in order to keep only the part that has a sufficient GSD. Would you recommend doing so? I have the feeling that it cancels the advantage of using a fisheye lens in the first place. Is it really detrimental to have regions of higher GSD in the images?
Moreover, I think that this paper doesn’t consider the increase in GSD due to the fact that objects on the edge of the image are located further away from the camera (my point a) above).
To sum up my questions:
- Is it correct that Pix4D can match objects with a GSD ratio up to 1:2 ?
- Does it make sense to take this into consideration for the acquisition plan?
- Do you recommend cropping fisheye images to keep only the central part with acceptable GSD?
Thanks for having read this long post,
Looking forward to hear your point of view on that
