Having had much success with a lowly camera on building exteriors, getting nearly all photos to register in a single very sharp component first time using standard default one-click, am realsing that interiors are a different matter.
A camera with bigger sensor, manual settings to force small aperture for depth of field and low ISO, thus tripod essential, maybe remote trigger, and/or artificial lighting. Also when photographing a window from inside.
So, a couple of questions arise so far -
does it matter if some photos in a series are taken with the camera upside down, close to floor level? Is it OK to turn them 180o in Windows Explorer?
as lighting needs to be diffused, is a freestanding flash with diffuser OK? The more flash light the better, as long as well diffused?
does it matter if some photos in a series are taken with the camera upside down, close to floor level? Is it OK to turn them 180o in Windows Explorer?
No rotation in postprocess. direct use as they are. and DISABLE the AUTO IMAGE ROTATION in camera ( if the setting is present ) and therefore no problem taking images with rotated camera…
as lighting needs to be diffused, is a freestanding flash with diffuser OK? The more flash light the better, as long as well diffused?
Yes, not a problem this way.
Another Q - with flat featureles interior surfaces (wall plaster) I had understood that each photo must contain at least one distinctive feature so that RC can correctly locate photos which include featureless surfaces.
But seeing extensive candy-floss results, I’m wondering if in fact featureless surfaces should be somehow eliminated completely from any photo?
Even if photos are correctly located, can RC extract meaningfull match-points from featureless surfaces within them?
I’ve had some experience with RC alignment of large interior structures, and can offer the following:
I’m happy to say that you will likely have better luck than you imagine with plastered walls, as long as you shoot with plenty of overlap and frame tightly enough to capture the mesostructure present. That is, while from 15’ away a white plastered wall is effectively featureless, at ~5’ there’s actually enough detail in the stippling to participate in the overall reconstruction. If shiny black surfaces are the worst, shiny white surfaces are a close second – but plaster without a top coat is very much Lambertian, so as long as the surface isn’t glossy you should be OK. Do resist the urge to cross-lighting the surface to help accentuate the ‘tooth’ of the wall, given my second point below:
Your instinct to provide more light and better exposure is a good one, but try not to change the lighting angle dramatically between light positions. While all SfM feature recognition tries to be light invariant, that’s a tall order. If you’re able to treat your scene as a connected series of distinct zones and then arrange lighting for each that shifts only a bit between poses, you’ll be happiest in the processing phase.
Thanks Gotz - and Kevin, that sounds very good advice - both get in close and don’t vary the lighting within discrete sections - thank you.
Sustematically photographing walls @ 5ft range is going to mean many showing nothing but flat plaster!
When you say ‘not change lighting angle’, I would already be doing everything to arrange max-diffuse (directionless) lighting - would your advice then still apply? I’m thinking of a pair of freestanding flash guns with proper diffusers but ‘pointing’ in 2 directions, on a stand that’s independent of the camera tripod - and remote camera release so my own shadow won’t get in the way.
I guess adjacent sections (with different lighting) would rely on prominent distinctive features to bind them together.
Having found that simple hand-held camera shots are v inadequate, am thinking of lots of equipment, like above - is that going to be necessary?
The problem with changing light is simply the change in appearence of features - imagine a bump on the plaster, it will cast a shadow in a certain direction. So in one image the black shadow will be e.g. top left, on the next bottom right, on the last no shadow at all. Ergo RC will be confused. If it stays within a certain limit, it might still work well enough, depending on your needs for accuracy. Just try it out for yourself, there is no way around it!
I personally have good experience with a flash on the camera and no diffusor at all. If it is only a plain surface, there shouldn’t be much of a problem if your lens does not throw a shadow. The vignetting can be counteracted somewhat in RAW developement. If you can’t avoid shadows, make sure you have enough images of each area so the shadowy bits will be outnumbered by images where the area is properly lit.
Gotz, I understand that shadows varying between adjacent photos might confuse - that’s why I asked (to Kevin):
“When you say ‘not change lighting angle’, I would already be doing everything to arrange max-diffuse (directionless) lighting - would your advice then still apply?”
Shadows would hopefully be non existent or weak - so could the light source be moved with the camera?
Of course the answer is to try it out - but would appreciate pointers as I get hold of equipment to try.
yes, there should be no problem with a moving lightsource, at least in terms of alignment and geometry. That’s what I was trying to tell you with my flash example - it is also a moving light: it moves with the camera! The only issue might be that you get different colours due to different brightness which can affect the texture.
My advice about trying it out is an honest and well meaning one. I am not trying to get around a proper answer, otherwise I would not post at all…
The point is that there are so many variables, that it is impossible to pick the perfect equipment right away. I would argue that there is no perfect equipment at all, and I am sure that everybody does it in a different way. I started with a cheap powerzoom camera and a tiny sensor (very much not ideal) and not much of a clue with a rather large project. Still the results were not too bad, only I spent way too much time fixing everything. Soon after I was able to produce quite good scans, but they were also not perfect. It’s a process. I think that using a flash on the camera has some advantages, which is speed. I could imagine I would be through with one room when you start to reposition your lights for the second wall…
I’ve had good results shooting with fixed ‘soft box’ lighting in interiors; try to set up lighting for each part of your scene, using the natural segmentation suggested by your environment. Of course, you do need to move the lights after shooting a part, but if you’re using bounce light as fill, the feature detector should be able to pick up correspondences across parts.
As Götz pointed out, a strobe mounted on the camera can work, too. It will make feature detection more reliant on shooting with higher overlap, but that’s not a bad thing. The problem with a strobe fixed to the camera is that, for flat ‘featureless’ walls, changing the camera+strobe angle relative to the wall usually radically changes the features detected from shot to shot. SIFT and other feature detectors try to be lighting invariant, but as I noted, correlating features for varying light on a flat wall is a poorly posed problem. If you can make it easier on yourself with static lighting setups, you’ll likely find you’re getting much better feature correlation between views.
I’d love to see some examples of little trials as you go! Break a leg!
in which respect do you think the features change radically when changing the angle? As I pointed out in terms of difference of brightness? Because on flat or almost flat surfaces, the shadows are negligible.
The thing is that in most cases there is no time (or patience) for tests like you suggested. I have anb example where I used an old superzoom camera with a 1/2,33’’ sensor and a strobe and got results that could not be better for my intentions. And the objects (see example) are on the opposite end of a flat wall…
There is no smoothing - the stone looks exactly like that. And now look at the details of stone tooling and the plaster. What more do you want? The thing is if you, as you said, have enough overlap and change angles only slightly, then it will still manage to tie everything together. If you are not working at the limits in terms of resolution or project size, I find that it is quite efficient to take a lot more photos than theoretically neccessary.
And that ties into the trying out again: it very much depends on the object and the intentions you have…
