Setup for scanning branches

Hello all,

I will be scanning tree branches in the next few days as a method trial for comparison against LiDAR scans with a Riegl VZ400.  I will cut the branches from trees and scan then in the lab/studio.  My goal is to produce as good a point cloud as possible (not so worried about the mesh, though they go hand in hand), and definitely not worried about color.  I am mainly concerned about the reproduction of structure to the finest level possible.

I have 2 studio lights with diffusers (Elincom D-lite 2) + fluorescent ceiling lights, a rotating platform, and a green screen.  My camera is a Nikon D7000 (will take RAW and convert to jpg).  The branches are relatively large - maybe up to 2 meters in length.  There seem to be a few options i have to choose from, and advice would be much appreciated (I’m new to photogrammetry):

  1. Move around the branch with the camera, or rotate the branch with the camera fixed?

The only downside to rotating the branch would be that the lighting relative to the branch changes with each photo.  How much of a problem might this be, given that the light isn’t completely even?

  1. Use a green screen or not?

Rotating the branch would allow the use of the green screen in back, which could potentially help with the recognition of finer branch sections, but I’m not sure how it would interact with realitycapture’s algorithms.

  1. Use targets?

I’ve heard some mention of using targets in realitycapture, but not much.  Is it possible?

Thanks very much for your attention!


Couple questions first.

Which lens(es) are you using with your Nikon D7000?

What is the diameter of the branches? (on average)

Can you put up a photo of the setup? i.e. Area/place where you plan to scan the branches and with the branches in place




Hi Johnathon,

Thanks for your reply.  I have the Nikkor 18-140 VR lens on the D7000, and figure it’s best to fix the zoom while taking the majority of pics.  Probably somewhere around 35mm.

The base diameter of the branch will usually be from 2-3 cm.  The finest elements of the branch are usually around 2-3mm in width, though some may go down to 1mm in width.  (I would be surprised if those very small features will be able to be captured).

See photos of the space below (it’s a square classroom approx 10m on a side), but I won’t be able to actually set things up until just prior to the shoot on monday morning.




Don’t underestimate the level of detail you’ll get.

Below is a quick scan of a 20 eurocent coin I just did with my smartphone as an example.
This is an optimized preview quality reconstruction from source material that is quite noisy.

With proper lighting and using raw images the result would be a lot better.


Regarding your setup:

Your best bet would probably be to mount a “turntable” on the ceiling and hang the branch from it.

Then you can have the camera set up on a tripod in a fixed location and take photos as you rotate the branch.

Take between 12 and 36 photos per revolution, start at the top and move the camera down making sure you have plenty of overlap.

That way you can have the green screen as background and set up your lights on the left and right side of the camera.

This has the added benefit that leaves or twigs that are connected to the branch likely won’t move or be otherwise deformed.

Thanks ShadowTail.  So, you don’t think the change in lighting (relative to the branch) will be much of a problem?  I’m not sure how realitycapture will handle matching pixels when they are lit differently in different pictures…

The differences should be minimal with that setup as the light will always come from behind the camera and thus nicely illuminate the faces the camera can see.

Depending on where you set things up you could even abuse the projectors to project a white screen at the branch (If you can turn them) to get more light :slight_smile:

Great, nice ideas, thanks.  I figure I should leave the ceiling lights on as well, to add a uniform field there as well… agreed?

Also, any thoughts about the use of targets?  Alternatively, I could add some colored tape to portions of the branch to assist the algorithms…

RC doesn’t understand targets like some of the competitor software does.

If the bark has enough detail you should not need colored tape. Take a few test shots and do a quick reconstruction and then decide if you will add tape.

What you probably want to do is take a few shots from further away / zoomed out (gradually zoom out) until you have most of the branch in a few photos. That will help with alignment. With this it is important to have a transition from close up to further away so RC can establish the links between the pictures.

as for reconstruction:

There is a setting that tells RC to NOT use the sparse point cloud from alignment for the preview quality reconstruction.
It will then take a few minutes to complete but the result will be much improved and better for a first assessment of the reconstruction quality and a glimpse at the level of detail to expect from a proper reconstruction.

Ah, I hadn’t considered changing the distance to the object much.  So… I can take one round of shots that incorporate the entire branch from, say, two elevations (maybe 15 pictures in each round, so a total of 30 pics from afar).  Then I can take a set of pics from closer…  and maybe a couple transition photos in between the two distances?  I’m not sure how the transition from far to near works in RC…  Do I have to do full rotations in the transition, or can i just take individual photos in the transition?

This is a bit difficult to explain… the illustration below might help?


Hi Alexander,

nice chart!  :slight_smile:

You don’t need to do full zoom-rounds but I would place them at more sides, 3-4 at least.

Also, consider different height levels - the more angles, the better.

The number of shots in each cirlce depends on how dense your object is - the denser, the more images per circle.

The light will be no problem whatsoever - RC is really good about dark areas, so as long you can see something and there is no bad noise, you should be fine. The only problem could be the texture, but even that is handled well usually. And since you don’t care about it, even better!

Since you say you want to get the best possible result, I would rather suggest a fixed lens and also strongly recommend to use tiff instead of jpeg. Jpeg is good for many needs, but it can make a small difference. If you have drivers installed so your raws can be displayed by windows, RC can use them directly.

The level of detail is only limited by the resolution of the images. In my experience the smallest detail should have at least two dozens of pixels across the width to get anything, more if the texture is tricky. That’s just a rough guideline of course. But you can’t expect anything if your thinnest twigs are only a few pixels wide…

And really make sure that there is absolutely no movement of the branches whatsoever - though from what you described of the size that seems almost impossible to me. Hanging it like ShadowTail suggested is a clever idea, but how will you prevent it from turning slightly?

About the green screen: If it is such a big room, you might be able to set the DOF so that the background is so blurry that you don’t need the screen.

What you mean by coded targets are called Ground Control Points in RC. It does work really well although not automatically recognized. It will only work if the targets stay in exactly the same position in relation to the object, so it won’t work with the haning scenario.

Great, thanks Götz.

Is there a reference guide anywhere for how to combine photos from afar that get general structure with closeups for detail?  Just trying to figure out what the most efficient way for me to do this would be.  I’m not sure there’s a reason to do the afar shots at all, now that I think about it…

Light: I’m not so worried about dark areas.  Rather, I’m concerned that, as the branch turns but the lights do not, a darker patch in one photo will be lighter in the next, etc.  One option might be to set up top and bottom lighting, as the shadowing wouldn’t change much in that case as the branch rotates on the Z axis.  I could bounce the flash off the ceiling, with a fill flash off the floor, and hope that might keep it even as the branch rotates.

I don’t own a fixed length lens, unfortunately.  I’ll be careful about keeping it fixed.  If we end up using this method for our full study (this is a trial), I’ll pick up better kit, such as a fixed lens.  Getting a 2nd camera might be cool too, so we only have to do one rotation rather than two.

Cool that RC can handle RAW - didn’t realize that.

For the GCP’s, are there particular patterns you suggest?  Or, is it just a matter of having a colored or reflective material fixed relative to the branch?  I can attach them to branches if they’re small - i don’t think they’ll move if I make a good effort.

The movement of the branches will be an issue with the thinnest ones, but most are relatively fixed and rebound to their postitions…  The alternative would be to walk around the branch rather than rotating it.  I’ll do a trial of those.

In the end, I think I’m going to mount the branch on a rotating platform.  Hanging them gets pretty difficult and isn’t quite as portable…



Hi Alexander,

yes there is a reason for different distances, I totally agree with ShadowTail on that! If it makes sense in your workflow or not is up to you to decide though…  :slight_smile:

You said that you are not concerned about the colour, hence there should be no problem from dark to light representation of the same area. But agreed, constant light is in theory a tiny bit more ideal, since nobbly parts will not cast different shadows. That again is only a problem if there is strong (as in direct) light from a flat angle to the surface.

The suggestion about the lens was only in respect to your aim to “produce as good a point cloud as possible”. Better lens, better result. Fixed lens, most often better lens. That has nothing to do with keeping the same focal length using a zoom, which is not a dogma either. I myself use zoom lenses, but I do not have the same ambition as you. If you are satisfied with 97-99% of the possible, then go ahead!  :wink:

I would think that in your case 4 GCPS on top and 4 on the bottom are enough, a bit wider apart than the diameter of the branch. But only if you are genuinely concerned about absolute measurements, as in down to the millimeter. If not, then it is absolutely fine if you measure any recognizable length in the model and scale that. RC does a really good job with objects that are encircled by images. Only if you cannot do that, you need to make sure to use GCPs all around the fringes.

Again, if you want the best possible accuracy, you NEED to be concerned about the movement of ALL parts of the object. If it only moves by ONE pixel, it can skrew up your alignment. Not bad, but it will not be the “best possible”. I personally think for natural objects like your branch, a super high accuracy is not neccessary. But you might think different, which was implied by you…     …ok, I’ll stop beating that dead horse now.  :smiley:

Yes, the best way to go ahead is to try it out for yourself  - I’m always in favour of that! I would say that between 10 photogrammeters, there are at least 12 different workflows…  :wink:

Good luck!

Haha, cool, thanks Götz.

For the GCPS, I’m still confused about what exactly to use.  There are those chili tags perhaps?

Understood about the “best possible”.  We’re using this for a scientific project, quantifying the branching of trees.  So, basically, the better we can do, the better.  I think your rule of thumb of a dozen pixels of the finest feature is a good one.

Finally, re: distances, I assume the reason for the differing distances is to 1) build the overall structural model that ties everything together at a large distance with lots of overlap between each photo, and 2) capturing detail at closer distances, but not necessarily covering the entire structure… is that about right?

Thanks for your feedback…


couple more things

Can you put up an image of these branches?

Do you have access to any other lenses? if not, are you willing you purchase a different lens for your project? i.e. up to $700



Hi Alexander,

yes, the overall shots provide a coherent framework for the whole structure and the closer shots provide the fine detail. You need to experiment to find the best workflow. I think you have plenty of prior input now…   :wink:

Don’t worry too much about the targets. You can just make your own, if needed. You can also use pencil crosses on a sticker. Never mind what, the only important aspect is the precision, whith which you measure those targets, and how well they can be identified on the images. The Chilitags (there are also Apriltags) you linked to are only important, if you want automated identification, for which you will need separate software.