under water camera

[iam_msy]

Hi. I am currently trying to create a model from video taken using underwater camera. I didn't know the calibration values that's why after import I just run the autocalibrated images. In the process of creating sparse data, the object that I created was noticeably bent. Unfortunately I don't have any known control point and known distances to scale my objects.

I just want to ask whether the camera calibration manager in the utilities would help me in improving my data and if yes, may I know how? During import it says that it internal camera parameters will be automatically extracted, does it also support underwater camera?



Thanks.

[Andrea Alessi]

Hi,

yes, you can also zephyr in underwater scenarios!

if the object is simply bent it is possibly due to error accumulation (i.e. a planar patch of land looks bent like a "hill"). if this is the case, you can either put "fake" control points on the first run or you can use the workspace -> optimize camera parameters using terrain assumption to "flatten it".

The camera calibration manager can help in the sense that once you have done a correct reconstruction, you should export your camera calibration and use it for future projects. Zephyr is autocalibrated but in some cases it may fail, in which case, you should run your own calibration or export a calibration from a successful reconstruction.

[iam_msy]

Hi Andrea,

Sorry for the late reply. I tried to put control points but those are the controls that I've also see in the camera for the top view so that later I could stitch them together. Is that what you are asking me to do? If not, could you please explain to me further? Should I export the controls and assign coordinates on them?

Also, could you please elaborate more on how does the optimize camera parameters using terrain work? When I tried it, the default settings it still a little bent, I made the constraints confidence weight more than 50, it usually suggested to remove some photos. Other settings were done successfully but it obviously lessen the sparse point cloud.


The image below is when I tried the optimize camera parameters only ( the first option) but there still a little bend in the data.

One thing that I've noticed is that the images that I've extracted from the video have slight deformation as if they were taken with fish eye lens.I don't have any camera details or calibration details and I'm hoping that this could still be sorted by automatic calibration of zephyr?

Thanks again for your help.

[Andrea Alessi]

Hi,

- yes, you can drive the reconstruction towards the correct solution using control points. In this case, you can add as many as you need. Obviously, this is very prone to human error so you need to be very careful. Add control points, run a bundle adjustment and force their height accordingly. Again, be very careful when doing this.

- unfortunately of course, by increasing that value you are telling zephyr that after a certain threshold the automatically computed points are invalid. You can't have it both ways -- either the constraints win or they don't - so it's expected behaviour.

- it's normal that you see them that way. Zephyr is autocalibrated, so as long as you didn't post-process the images (e.g. by cropping them) you are fine. However in certain cases you may be forced to create your own calibration (e.g. by running a working dataset with the same lense) to obtain best results