How best to capture fine scale branching architecture

Hello sketchfab community!

We have been scanning trees with a terrestrial LiDAR for a couple of years but are now attempting to capture branches. See here and here for a couple of examples.

Capturing the 3D architectural morphology of branches is pretty challenging! They can have fine features (~1 mm diameter), irregular branching structures that are often sparse but also occlude one another, and they can also be difficult to work with e.g. trying to keep them still.

We have attempted a couple of methods for achieving this (laser scanning and photogrammetry in the above example) but I was wondering if any of you have experience scanning fine featured objects (not necessarily branches) or know of other methods (e.g. structured light etc.) or other teams that have attempted similar?

The methods we have tried so far have been successful but require expensive equipment (we have been using a RIEGL VZ-400 – see pic – or large photo room). In order to make this approach adoptable to the wider science community it’d be great to hear of low cost options (and the trade offs) of e.g. using mobile phone technology.

I’d be very interested to hear your responses!

2 Likes

Interesting problem…I have yet to experiment with this, but will try running some tests this week. :slight_smile:

1 Like

In this situation, you might have better luck trying photogrammetry within a light-box and rotating the objects. This would remove background noise and allow the algorithms to focus on the main geometries. I’ve seen that laser and LiDAR pick up thinner objects better than photogrammetry does (power lines for example), so have you tried a handheld scanner? There are many out there that are used for manufacturing/reverse engineering.

1 Like

to clarify @kungphil:

  • have you been masking your photogrammetry image sets?
  • have you tested the same image set in different photogrammetry software?

Thanks for your reply, yes we have been using a light-box with a green screen and rotating the branches on a spinning table. This approach is slow though!

We have also tried hand-held lasers (ZEB-REVO) but the beam divergence is large and beam footprint is elliptical which has resulted in quite a noisy point cloud.

We haven’t tested the hand-held structured light solutions and wonder if this is a good approach. Their promotional videos make them seem like a very rapid solution but I’m not sure if minimum feature size would be an issue?

I am not the photogrammetry person, that’s @ashenkin. But I think yes we have been masking the images using a green screen and 2D bar codes to reference position (here is a vid of some trials).

We have been using Reality Capture to process image sets.

Ah interesting video!

@ashenkin Was the result you share of the hawthorn tree captured just one angle? Was the height of the camera and tilt adjusted to capture images of several rotations?

It might not hurt to grab a free trial of Agisoft Metashape and/or 3DF Zephyr and run the same images through these software.

One final thought re. photogrammetry is to check that depth of field is not an issue and adjusting aperture or consider focus stacking if it is.

Hello,

Nice and very unusual project. I like your setup with the branch fixed to the rotation table, I wonder how many “what are you doing in there” you got so far!

My experience is with very low quality data and really not optimum conditions, so I guess it can be easily improved.

  • Tiny branches (lavender)
    I used a low quality USB microscope, one angle to the base and 40-50 images. The “setup” picture is in the commentary section. The light was very bad (from USB LED) and no background control expect for a piece of paper (lot of reflective stuff behind that paper…). The scanning is less than a minute and processing less than 10. The image are like 2MP so it goes fast. The quality or even the focus did not impact much the result, it does impact the texture. But there are enough feature at the end to get camera positions.

  • Video of penzai
    This other test of small tree was recorded in a garden in Suzhou, China. It was in summer so very sunny weather. The “white dot” on top of the point cloud are actually saturation of the sensor due to the light, not snow. Shot in 1080p while walking around the object (20-30s). Position of the branch came out ok but way too much saturation and lot of leaves “hide” the trunk. Again very low resolution image, very fast processing and it works “on site”.

In terms of “personal opinion on expensive technique that I don’t have”. I think handling scanner or LIDAR might not be practical as you will have to “reach out” to get all the details, looking all side of the branch. These are good instrument, but not meant for such small things. Plus once you get out of the lab, if that’s a consideration, you will be stuck.

Depending the level of accuracy you need, you might consider a customize setup that fit well “branches”, like a circle of cameras or something. SFM is indeed slower, but the cameras give you more flexibility in the setup.
Also you can record other data from simple camera. There are package for structured light (at least for research) if you want to try.
You can also use “linear laser”. It’s closer to “structured light” than “lidar” as it records how the laser beam in your image moves according to the distance between camera and object. Hence resolution depend greatly on camera accuracy. But it costs nothing.

@kungphil hey - have you found what you need yet?