more detail and less noise

[km206]

Hello, this is my first post.
I used to Photoscan standard version.

I did some tests for clay model, then I did some improved scan.
my purpose is to get mesh for fine detail, scratch texture and any other clay texture.
but it getting geometry has some problem.
I tryed take a picture in light box. this test has become less noise, but I think it not enough.
And the shape model doesn't have detail.
I attached the original photo and shading mmodel picture.

it made by set to "high" and 0.3 million faces.
Please give me some advise.

[RalfH]

The lack of detail in some areas or the model could be due to several reasons:
(a) incomplte coverage / insufficient number of photographs: every point of the object should be visible in at least 3 (better more) photographs
(b) poor image acquisition geometry: the angle under which the camera looks at the object's surface should generally be steeper than 45° for optimum results (it can be shallow in some images, but for every object point there should alwas be at least 3 images taken at relatively steep angles)
(c) poor camera calibration: if anything moves in your lens/camera system between images (due to zoom, focussing, image stabilisation), use "separate" (instead of "group") camera calibration settings
(d) poor texture: if there is insufficient texture on the object surface, 3D reconstruction will be less detailled or contain noise
(e) blurred images: same as for (d)

[km206]

Thank you RalfH,
I'll try to check camera callibration.

[hsmith]

RalfH,

Thanks for a very helpful list of how to get better scanning results.
Just a clarification for item (b):  the angle between the camera and the subject should be as close to normal (90 degrees) as possible?

Harold

[RalfH]

Harold,

it's not necessary to be close to 90°, but shallow angles can be problematic for several reasons (mainly view obstruction, poor matching point recognition and increased depth error). Anything steeper than 45° can usually be considered good, and if every object point is seen by at least 3 cameras under good angles, 3D reconstruction should work well.

Of course there is some room for experimentation, and the algoritms implemented in Photoscan appear to be very robust. For example, I have successfully generated a model for which all cameras saw the object (in this case, the ground surface) at shallow angles (less than 30°), but as a result geometric resolution was quite poor, there were some holes in the model, and many small surface features were quite distorted. It was a usable model, but very far from what you can get with a better set of images.

[km206]

my test had get better scan model.
I changed set callibrate cams "separate". (thank you RalfH).
Then, change "smooth" to "sharp" geometry reconstruct option setting .
Please look at my result. this getting better. (but i think this is not enough.)
Thank you,

[Andrew]

>use "separate" (instead of "group") camera calibration settings

This sounds interesting, I never tried this - are you referring to 'Split Groups' in left pane of 'Camera Calibration' dialog window? I usually use same focal lenght but sometimes resort to using OIS and wonder if this could improve precision of my reconstructions?

Thanks,
Andrew

[RalfH]

Andrew,

yes, this is what I mean. In my experience, image alignment is better when using separate camera calibration. For example, when using grouped calibration I sometimes get artefacts (similar to interference patterns) in the 3D model; this happens much more rarely when using separate calibration (and then only if there is sub-optimal overlap).

The advvantage of separate camera calibration is a logical consequence of what happens in a lens system with moving parts - even if focussing repeatedly goes back to the same focus setting (e.g., infinity), as soon as something is moving within the system you can't expect that all lens elements will return to exactly the same position. With OIS, the image principal point is shifted between photos, and this is quite a crucial point for camera calibration.