Reply To: Warsaw University try on ARSandbox

Home Forums AR Sandbox Forum Warsaw University try on ARSandbox Reply To: Warsaw University try on ARSandbox

#102042
GeoKicak
Participant

Aha!

Thank you very much for identifying what causes this stepping problem.
The kinect is suspended at approximately 2,5 meters so yes we know, it is mounted not at the recommended height.

After 15 minutes of TORCH searching all of the interwebz in an effort to find some solution to this behavior I have found only more information about the probable cause of the stepping problem.

The technology was developed by PrimeSense which is now owned by Apple. Details about that technology are not publicly available and the only legal way to get some info about it is through reviewing the patent applications.

Some time later …

So, from these documents it seems kinekt depth map is constructed by analyzing a speckle pattern of the infrared laser light which in basic is an analyzing technique of a known pattern called structured light.

This kindergarten method
(http://www.handimania.com/uploads/3d-handprint-12.jpg)
elevated by human ingenuity, machines & electronics
(http://homes.cs.washington.edu/~curless/images/hand_first_pattern.jpg)
(http://www.laserfocusworld.com/content/dam/etc/medialib/new-lib/laser-focus-world/online-articles/2011/01/93355.res/_jcr_content/renditions/pennwell.web.390.296.gif)
to obey our hearts desires was further improved in kinect, which uses the aforementioned speckle pattern (http://gmv.cast.uark.edu/wp-content/uploads/2012/07/KinectIR.png).

From my personal experience in the meaner of dots they do have a tendency to act funky and produce interference in form of moving phantom objects (eg. circles or lines at 5:03 –> https://www.youtube.com/watch?v=QAja2jp1VjE ) but these wall like interferences require 2 shifted superimposed images to be visible.

Kinect analyzes the shift of those speckles between the IR feed and the initial pattern stored in memory, then calculates the difference in the pattern dot shift at different distances and on the basis of triangulation principle calculates the depth of an object
(http://www.futurepicture.org/?p=97)
(http://www.depthbiomechanics.co.uk/?p=100).

The important part (Me thinks so) it that the fixed dot pastern is semi-random.
(http://www.francescocastaldo.org/tutorials/kinect.pdf)

Since silicates are known to reflect IR and other mineral grains in snad might absorb some of it
(http://articles.adsabs.harvard.edu//full/1994MNRAS.269.1011K/0001011.000.html) maybe it is possible that the anchor’s shift

(https://courses.engr.illinois.edu/cs498dh/fa2011/lectures/Lecture%2025%20-%20How%20the%20Kinect%20Works%20-%20CP%20Fall%202011.pdf)
can produce a Moiré pattern appearing as moving, stepping walls derived from square IR pixels.

Moire pattern

some more searching …

But then again I know nothing of optics and I am not a smart man, and so I am maybe totally wrong and this error has been reported in various places on earth for all devices that use Primesense chips (Kinect, Asus Xtion, Asus Xtion Pro, and Primesense camera) and the solution for kinect is to use libfreenect and modify SensorKinect\Data\GlobalDefaultsKinect.ini file and uncomment
;GMCMode=0
so it states
GMCMode=0
like it is written here https://github.com/OpenNI/OpenNI2/issues/81

I should definitely go sleep now,

Bye, Have a nice day 😀

  • This reply was modified 4 years, 6 months ago by GeoKicak.
  • This reply was modified 4 years, 6 months ago by GeoKicak.
  • This reply was modified 4 years, 6 months ago by GeoKicak.
  • This reply was modified 4 years, 6 months ago by GeoKicak.

Comments are closed.