Is anyone doing multi Kinect scans?

[Fred Kahl]

Hi All,  

I'm curious if anyone is having any luck with scanning using multiple Kinects? Currently I have only 1 Kinect, but wonder about improving the quality of scanning people using 2 or more Kinects at once. 

Does anyone know what the limit to the number of Kinects that can be connected to the scanning computer is? I'm wondering about creating a scanning booth to scan people full body, for the purpose of 3D printing them. I imagine that 3 Kinects could nicely triangulate to scan a person. I remember having seen kinect powered scanning booths for fashion/virtual clothes try ons, and think a similar set up world work well with ReconstructMe. 

One concern I have is with my hardware- I am currently running the software on a MacBook Pro under Windows 7 on Bootcamp. This setup's Nvideo GeForce 330M seems to barely have the power to be able to do a single sensor capture, so I'm not sure if the double sensor will work with it or not. I'm also curious what hardware other people are running ReconstructMe on? I would be curious what people would recommend as a decent powered PC laptop to scan with. 

best,

fred

[Mark Schafer]

I use two sensors. Someone else (see threads or blog) used a mirror to capture the back side at the same time as the front. Seems ideal for full body scan ideas.
We are expecting an update that allows two scanners (more?) to simultaneously add to the same volume. That will make things simpler as will the upcoming multi-volume reconstruction which joins adjacent regions into one.

You can try two sensors at the same time by runnig them in two shell windows at athe same time.
You wil need a separate USB controller for each camera as they use all the bandwidth. This is likely to be the controlling factor for number of devices connected (as well as the shear amount of GPU memory that would be required - unless offline reconstructing).

I have not heard of a laptop with multiple USB controllers - they all tend to have only one. By adding USB cards into a desktop taking up PCI slots - you can get more sensors connected.

For one sensor - any laptop with a GPU that has a openCL driver and > 512MB of memory (but really 1GB is better) will probably work. I believe a table is being constructed of successful cards. Its all in the forum... search away

[Fred Kahl]

Interesting, So you need a desktop setup with multiple USB cards... makes sense but not as convenient. 

I was looking online and saw a couple examples of people doing scanning booths like this one: 

http://www.youtube.com/watch?v=9MwS_nk9n2A

The whole thing of having people stand still to be scanned while you rotate them or move around them is kind of interesting. Kind of a throwback to photography 100+ years ago. I was just looking at finding the easiest/quickest way to scan people now. This system claims to take 3 seconds to capture someone, and they are using Kinects:

http://www.styku.com/

Well I am looking forward to the next update... stitching seems exciting! I want to scan full bodies and print them out! 

fred

[Christoph Heindl]

Am Dienstag, 10. April 2012 05:00:05 UTC+2 schrieb Fred Kahl:

Hi All,  

I'm curious if anyone is having any luck with scanning using multiple Kinects? Currently I have only 1 Kinect, but wonder about improving the quality of scanning people using 2 or more Kinects at once. 

Fred, ReconstructMe does not yet support multiple sensor per single ReconstructMe instance. What you can do and what has been done, is to start ReconstructMe twice and directing each instance to a different sensor.  

Note however that the last version of ReconstructMe has a bug, that will not allow you to do that (use version 193 for testing). We've been playing around with multiple sensor support lately (labs), but there is nothing to expect soon.

Best,

Christoph

[Charles Dunk]

Great way to remove interference between multiple Kinects, just shake them! Microsoft Research has released the following video showing how it works:

http://www.i-programmer.info/news/194-kinect/3869-shake-n-sense-makes-kinects-work-together.html 

[donnagalea]

I saw styku exhibiting their more updated version about 2 months ago. They used the Microsoft Kinect developer package to write an algorithm to find closest points (there's a correct term for this, but I forgot), which you can do with xbox OpenNI. They were doing calculations of pants sizes - not whole body scanning - they used 4 kinects, 2 kinects on each side of a person, turned on 2 at a time to avoid interference. 

I have used 2 kinects 180 degrees from each other with Reconstructme running on 2 shell windows but the interference left a gap where the two scans are supposed to intersect. 

My current solution is to use 1 kinect, have my subject slowly rotate, once to scan their torso, once to scan their abdomen, and once to scan their feet. They have to have no protruding limbs, and their feet has to be together. I can get a near solid scan in this way. 

I will then add planes to block the hole at the top of the head and bottom of the feet, then import the scan to zbrush to be remeshed. I'm planning on posting this process on youtube when I get a chance, but it works pretty well. 

Attachment 1 is a toy dinosaur I scanned using 2 Kinects on 2 separate Reconstructme windows, then I manually pieced them together. 

A2 - scanning with 1 kinect just by having the subject turn 3 times. I deleted the floor area, otherwise the scan is untouched. 

[racin...@gmail.com]

Check out these guys, software is easy to use also.
http://manual.kscan3d.com/1.0/index.php/Creating_A_Multi-Sensor_Setup_For_Full_Body_Scanning

[csmal...@gmail.com]

That is a hell of a lot of manual work for a pretty mediocre result. You would be better off just using the new version of reconstructme with texture and a single kinect or try taking photos using Agisoft Photoscan.

[jamesm...@gmail.com]

I'm going to try this multi-sensor setup. I already have two USB controllers (one for each hub front and back;3-3.0's, 3-2.0's: PCI express is nice; not sure on my model) and an iBox Nano coming in March 2015. Anyone know of any of the other 3D Scanning Software that works well with this method?

>WINDOWS 7(64-bit)
>16gb RAM
>7550 GTX graphics card(I think, not at my PC now)
>AMD A6 processor

I'm hoping that I have enough power to process the images on two monitors, my graphics card is basically split if I'm running two programs on separate displays.