Monday 12 March 2012

Kinect accuracy and precision with RGBDemo

The presentations that impressed me the most during the workshop "Low cost 3D: sensori, algoritmi e applicazioni" (Trento, 8-9 March 2012) were the those of FBK's researchers and especially:  "Tecnologia TOF per 3D imaging – low-cost o no?" (L. Gasparini, D. Stoppa) and "Caratterizzazione di sensori attivi e passivi low-cost" (F. Menna). One of the topic of these contributes was Kinect and the level of detail,  accuracy and precision this instrument can reach. In particular it was said that using Kinect with a distance of 50 cm it is possible to record 3D scenes with an error of more or less 2 mm.
Alessandro and me already tested Kinect in ArcheOS (on Ale's laptop) during our teaching experience in Lund University, but in that occasion we did not check the level of accuracy. For this reason Alessandro compiled RGBDemo on my laptop during the workshop and, thanks to the coffee break, we tested it with some researchers reaching an accuracy of 1 mm.  
This evening I repeated the experiment at home and here is the video of RGBDemo Reconstructor during data acquisition:


The yellow square you see in the movie is a normal folding meter and its side is 20 cm long. Here is a more detailed picture of it:


In this other movie I checked the accuracy I reached within MeshLab. When I measured the side of the squared the result was 0.198 m (better seen in full screen...).


I think this level of accuracy is more then enough for archaeological applications. Anyway I will go on with other test in the next months, hopefully with something more professional than my living-room :). 

7 comments:

  1. very interesting!! thank you guys!!

    ReplyDelete
  2. Amazing results. Kinect is looking more and more an affordable tool and less a toy.

    With an accuracy of 1 mm you can use the sensor to quickly model large amounts of artifact fragments (i.e. pottery).

    Probably, in a near future, extensive artifact scanning would be more used. In this way, fragments can be automatically remounted saving researchers time

    See:

    http://formaurbis.stanford.edu/docs/FURproject.html

    2 mm are more than enough to field work or even big targets, with irregular surfaces (i.e. granite statues, rock art). Unfortunately remains the limitations of the infrared laser when working under direct sunlight.

    rgaidao

    ReplyDelete
  3. Yes,
    by now we are looking to Kinect more as an indoor tool (especially in low light conditions).

    ReplyDelete
  4. Pretty cool!

    http://www.pgrammetry.com/forum/viewtopic.php?f=30&t=429

    rgaidao

    ReplyDelete
    Replies
    1. Hi Ricardo,
      thanks for the report. The video is really nice, They do amazing things (especially considering the rel-time software response to the Kinect input); I just fear that what they do is not yet possible in archaeology: it looks like that in the video Kinect is really close to the sandbox (low errors) and that the scene is captured indoor (no direct sun)... Like you already noticed, we should test Kinect outdoor...

      Delete
  5. Hello Luca. I've done some tests with the Kinect, to document caves. Use Skanect software. The results seem interesting. In this link you can see a sample.
    https://www.box.com/s/8843aa9e1348bce14a86

    ReplyDelete
  6. Hi,
    It looks very good! I did not know Skanect software; it seems to be developed by the same team of RGBDemo (http://manctl.com/projects.html). I really think that caves and underground environments are a good field of application for Kinect and this kind of software.

    ReplyDelete

BlogItalia - La directory italiana dei blog Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.