Scanning sculptures of a facade

[Mfranquelo]

Dear all,

Im going to do a little project next week which involves scanning the back of sculptures on a facade that cannot be reached using structured light scanners (or maybe they can, but its a much slower process) I dont need extremely high resolution.

Ive thought of mounting two 5d's 50mm lens with two flashes on an horizontal rail. (Synced) acting as a stereo pair.

The sculptures dont have any visible features, at least NOT many, to my eye. So i may need to use projected noise pattern? In this case, i guess i can project any pattern. Is just a standard noise pattern good enough ? or are there better pattern for feature recogniziton ? how big should the pattern be in relation to the object?

Are there any benefits of using a stereo pair set up ? Its easier for math operations knowing the distance between each camera's i guess ? whats the recommended distance from camera to camera ? Ive read 6,37cm somewhere.. but on the OpenCV book i read that 12cm is recommended.


Manuel.

[bisenberger]

Hi Mfranquelo,
Instead of thinking in terms of distance between cameras, which would also be affected by focal lenth and distance from target. It might be better to think in terms of overlap.

Here is an image that shows photo overlap for legacy aerial photography:

[Mfranquelo]

Hello bisenberger

Thank you for your answer.
I know i need a high amount of overlap, cameras would be facing to a central point P so that each image is in 90% overlap (maximun) with its pair one. That is not critically affected by the distance between each cameras but the angle.

Any thoughts on the noise projection ?

[James]

There was a great post on this a little while back http://www.agisoft.ru/forum/index.php?topic=1542.0

It would probably work best if you used this with a single off camera flash synced with both cameras. If you could keep the flash stationary then you could take several stereo pairs without the projected image having moved. If you can also keep the stereo pairs stationary then you can do a pair with the projection and one without for later texture mapping.

I haven't tried it but can see many cases where it would be useful, even on a single camera setup.

Re camera spacing, i can't find it anywhere but am sure i saw somewhere the recommendation that the baseline distance is ideally around 10% of the distance to the subject, rather than any absolute value.

The agisoft wiki here http://www.agisoft.ru/wiki/Photogrammetry#Relief_displacement also recommends convergent camera angles to maximise overlap assuming a known distance to your subject, and maximise reconstruction precision in the z/depth axis.

[Mfranquelo]

Very nice post James, thank you for pointing it out.

Im not sure what advantages i could take of that system however (maybe i need to think a little bit more), -> I could fire 2 strobes of light, one without the transparent film and the other one with the film to project the noise pattern, each of them syncronized with the camera.

I cannot think of other advantages of using it and not using the conventional projector. Everything would be in perfect sync using flashes. even the projected pattern.

I?ll read a little bit more about the distance. thank you.

[Kiesel]

I see three advantages to project a pattern with a flash in comparison to a projector:

1. higher portability (no power cable, smaller, lighter, ...)
2. brighter
3. higher resolution of slides -> higher resolution of the calculated geometry

The ratio between camera stereo base and distance to the object should be in the range of 1:5 to 1:10 for good stereo model.

Regards,

Karsten