3D Scanning from photos 0.8 beta release notification

I have just uploaded my 3D scanning from photos software, now called Carapace Copier, to SourceForge.

Zip files are available for download at [sourceforge.net] and I have created a folder in the Reprap SVN repository under trunk/reprap/miscellaneous/ called CarapaceCopier to store the files.

It is a Java program that just needs JRE 1.5 to run.

It will take as input a calibration sheet, simply a white background with black circles on it, information about the dimensions of the paper this was printed on to, and a series of images of the object to be 'scanned' sitting on this calibration sheet and it will output an ASCII STL file.

Currently this STL file is the maximal volume estimate for this object with a surface made up of small boxes (voxels). With all the default settings and the calibration sheet printed out on A4 or US Letter paper these boxes are a little over 2mm on a side. This shape is found by producing a silhouette out of each image using the assumption that if the known calibration sheet can be seen from the camera viewpoint, then there is no object in the way. This is an intermediate stage as I am hoping to be able to do better than this and have colour/texture matching between images to bring out finer details but it is a start.

The software comes with a small document with lots of screen-shots following through the 5 step process on an example calibration sheet and example images of a juggling ball on this sheet. Note that the user is only involved in the first 2 steps and the other 3 are just logically separate automatic steps.

There is also a 40 page Software Technical Report, that is hopefully going to be useful to those who wish to modify the software and/or add it to their software tool-chain.

There are a few known bugs already documented in the ReadMe but I would like to get some feedback (and even possibly help) in finding and fixing any other issues that arise.

Hi Reece,

... this sounds really interesting!

I'll download and test it, but can't await when it's ready.

My aim is micro-scanning with series of images from my microscope-cameras or a special setup with two cameras as stereo-pair with a FOW-area of around 5x5x5mm and resolutions of 720x560 pixels per sensor ...

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]

Thank you for doing this, looks like a great deal of effort on your part, look forward to trying it out

---

Random Precision

Hi Reece,

... i tested with the atached image-sequence - it's not exactly a representation of the ball

How is the best strategy to get a better fitting?

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]

Seems to me like the DAVID-laserscanner is a better way to go.

You only need:

• A camera (e.g. web cam)
• A hand-held line laser
• Two plain boards in the background
• A Windows PC
• The free software DAVID-laserscanner

---

Bob Morrison
Wörth am Rhein, Germany
"Luke, use the source!"
BLOG - PHOTOS - Thingiverse

Hi Bob,

... i'm urgently searching a photogrammetry scanner for scanning finer details than possible with laserpointers.

I've tested David LS, it works for details down to 0.8mm if you have a fine line, and with a 'wire-shadow' i can get by with 0.4mm, but i need much finer details ... and Recce's programm combined with video-microscopy should do the job

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]

@Viktor: Of course if you are looking at microscopic objects that you are taking with a microscope camera then his method is the better choice. But for general purpose scanning of objects that you want to print with a RepRap the DAVID-laserscanner (or other similar projects) are the way to go.

---

Bob Morrison
Wörth am Rhein, Germany
"Luke, use the source!"
BLOG - PHOTOS - Thingiverse

Be hard to use David on say a model of T-Rex or say a model of a Mastodon basically anything bigger than a sheet of paper.

Great work Reece. I'll be happy to use it if I ever achieve printing. I'm to cheap to buy the parts and in the past 5 years I spent far more making various DIY contraptions that just didn't quite work out.

Quote
Arvin
Be hard to use David on say a model of T-Rex or say a model of a Mastodon basically anything bigger than a sheet of paper.

Difficult but not impossible!
But the calibration sheet would have to be wallpaper sized and the laser line very powerful...

---

Bob Morrison
Wörth am Rhein, Germany
"Luke, use the source!"
BLOG - PHOTOS - Thingiverse

Reece - great stuff. But can you re-arrange things on Sourceforge so it doesn't appear as the main RepRap download?

---

best wishes

Adrian

[reprap.org]
[reprapltd.com]

@Adrian Oops. Sorry about that.

In reference to the David LaserScanner, yes it is better and is likely to remain so for a significant period of time but I went for the low tech route and try to solve the problems in software so that once they are solved it is easy to spread them around.

As far as the issues with the test juggling ball looking not particularly ball like. Yeah it sucks and I was hoping to have something better by now. Its in an intermediate state but its really time it should come out and see the light of day, so here it is. Not quite "release early, release often" but better than nothing.

I've put a few suggestions in the tutorial as to how to get it better. In short, more photos, from more angles, with less shadows.

The resolution of the outline could be trivially made more fine with the use of octree-voxel subdivision but the eventual idea is that this outline is just used as the maximal volume for a better processing technique to process. The current idea is to get better recognition of shadows to cut down on some of the issues by use of local contrast filtering etc. and then to find/invent a texture/feature/colour matching scheme for the subset of the image we know the object resides in. At the moment all colour information is thrown away and just the greyscale values used in processing to make it more simple. This probably needs to be added back in at this point.

Viktor,
Have you looked at imageJ, if I understand correctly you want to construct a model from image stack?, also mimics is a commercial software for the purpose but very expensive.

---

Random Precision

Hi John,

... i'm searching a free/open-source software for calculating 3D-surfaces and complete objects from stereoscopic images or series of images taken by turning the object or moving the camera(s) around.

I developed some video-microscopes with a much higher FOW - look here for some images we made with the optics ... the last 4 images are red-cyan anaglyphs made with the stereoscopic setup.

I have some videos too, where the objects turns and i can calculate 3D-sets from any pair of frames ...

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]

Sorry Viktor, I miss understood, some plugins in imageJ use the narrow depth of focus of microscopes to construct a model,
I have been searching for free or affordable software for some years to construct models from stereo pairs, but have not found anything good enough, photosculpt is good at some things and I was impressed with some results, but is only good at shallow depths, I did play with autopanosift some time ago,it can extract points from multiple images, so may be possible to create some sort of point cloud?, but if you are in a hurry I am sorry but know of nothing off the shelf that is affordable.

---

Random Precision

... Photomodeler Scanner is capable of generating point clouds and 3D-meshes, but something pricey

I was testing some 3D-applications with stereo-cameras and neuronal networks detecting 3D-points 'on the fly' - this was aimed for high-speed stereoscopy for preparing the airbags in the case of an car-accident - but with more than ten thousand Euros this wasn't the right tool ... but it made really good 3D-meshes

Another very accurate high-speed 3D-scanner was working with a beamer projecting different patterns on a vibrating surface - this type works like DavidLS, but with much more sophisticated calculations, so it wasn't given away as open-source

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]

This may help if you have not already seen it, strucured light source code and binary

---

Random Precision

Hi John,

i was there long before and downloaded the source, but for programing you need the aditional toolboxes/packages and Matlab, so i put it away until i'll have some more time for reprogramming with my toolchain

---

Viktor
--------
Aufruf zum Projekt "Müll-freie Meere" - [reprap.org] -- Deutsche Facebook-Gruppe - [www.facebook.com]

Call for the project "garbage-free seas" - [reprap.org]