Fisheye lens without EXIF

[compsci]

I tried to use a fisheye lens which lacks a digital adapter (Walimex 8mm), therefore the camera can't store EXIF data about the lens. I set the focal length manually via exiftools, however even with "Adjust Radial Distortion Parameters" and "Adjust Tangential Distortion Parameters" the reconstruction fails to produce correct results. Either the 3DF Zephyr puts all images in a plane or doesn't properly undistorts them, resulting in an odd shaped room.

What data in the exif is required and/or what settings should be used for the intrinsic calibration? Is it possible to let 3DF Zephyr figure out the radial and tangential distortion on it's own or is checkerboard camera calibration required?

[Andrea Alessi]

Hello compsci,

the exif data is used for a pre-calibration step. Since you are using an adapter, i suggest you simply strip all exif data and follow 3DF Zephyr's calibration procedure (Utilities -> Images -> Launch camera calibration).

You'll be asked to take a few photos of a pattern on your screen. Do it, then strip all the exif data (for example by converting the images to PNG) just to make sure you are not feeding wrong information to zephyr.

That should be enough. Once you are going to process your data, simply load the camera calibration in the new project wizard and you should be set.

[compsci]

Thanks for the fast reply! BTW that is a smart approach, better than having to print a checkerboard and glueing it on cardboard. The only people who'll run into trouble with this approach is when you have a curved monitor. Will try it today and report back.

[Andrea Alessi]

Curved monitor won't impact zephyr's calibration process You could actually take photos of anything and feed it back to zephyr as long as there are a lot of features

[compsci]

Unfortunately the calibration did not work as expected. I also tried another set of feature rich photos. In both the calibration was not accurate enough to aid the reconstruction. I'm not sure where the problem is, perhaps the Walimex/Samyang with stereographic projection doesn't fit with 3DF Zephyr camera model?

[Andrea Alessi]

Hi compsci,

did you try taking a few more photos? if possible, can you share the dataset with us?

[compsci]

Sorry for my late reply. I will try it again with more pictures and post the dataset in case it still doesn't work.

[compsci]

Another even later reply. It works!

The missing exif-data was manually added using exiftool (-FocalLength="8.5mm" -Lens="8.5mm" -minFocalLength="8.5" -maxFocalLength="8.5")

At first I used one set of pictures to get a rough estimation of the intrinsic parameters. This didn't really work. Different attempts resulted in greatly varying fx, fy, cx and cy parameters for some reason. The distortion coefficients were alright and relatively consistent across different attempts (although the following steps also improved their accuracy). In order to get accurate intrinsics I refined the calibration by importing the camera parameters and enabling the pre-calibration step (not sure if enabling auto-calibration actually uses the settings as a start). I manually copied the results from a cameras properties into the xml-file. This allowed to refine the estimate by using the latest parameter estimation and after two iteration with different small dataset the calibration was very accurate. While this might be a bit cumbersome, the results a quite fantastic afterwards.

So in case you have a cheap fisheye lense without a chip and thus no EXIF-data, this is not a problem. After lots of experiments and frustration I can conclude that 3DF Zephyr is also a better options than Agisoft Photoscan in speed, dense reconstruction and texture quality even though the calibration in Agisoft is easier (unless you need more than 500 images, then it's a matter of price or your skills with 3D editing software assembling the pieces).

[Roberto]

Thanks for sharing the information with everyone!