MS announces new Kinect hardware for PCs with "near mode"

106 views
Skip to first unread message

judytuna

unread,
Nov 22, 2011, 8:02:19 PM11/22/11
to openk...@googlegroups.com
Did you guys see this yet?? https://blogs.msdn.com/themes/blogs/generic/post.aspx?WeblogApp=kinectforwindows&y=2011&m=11&d=22&WeblogPostName=kinect-for-windows-building-the-future&GroupKeys=

"near mode" goes in to 40-50cm.

!!! This seems particularly relevant to the recent post about hand gestures, too!

Also of note: the startup accelerator that was just announced by Microsoft. Are any of you guys going to apply?

Kyle McDonald

unread,
Nov 22, 2011, 9:45:21 PM11/22/11
to openk...@googlegroups.com
very interesting -- i asked primesense about this a few times and the
answer was always "we're not interested in the near range". so i
wonder if MS has their own engineers modifying the primesense
reference implementation?

also of note: the kinect can already see 40-50 cm in good conditions.
i'm curious if they've increased the accuracy in this range,
sacrificing far-range accuracy? this would be great news for people
working with portraiture/face biometrics/hand tracking.

kyle

drew.m...@gmail.com

unread,
Nov 22, 2011, 10:01:24 PM11/22/11
to openk...@googlegroups.com
On Tue, Nov 22, 2011 at 6:45 PM, Kyle McDonald <ky...@kylemcdonald.net> wrote:
> very interesting -- i asked primesense about this a few times and the
> answer was always "we're not interested in the near range". so i
> wonder if MS has their own engineers modifying the primesense
> reference implementation?

Possible. Or maybe the quantity of people asking Microsoft for this
more recently prompted them to contract PrimeSense to do a
modification.

> also of note: the kinect can already see 40-50 cm in good conditions.
> i'm curious if they've increased the accuracy in this range,
> sacrificing far-range accuracy? this would be great news for people
> working with portraiture/face biometrics/hand tracking.

Isn't the only real issue with using the Kinect at close range the
brightness of the laser pattern at close distances? The IR image
appears to just get washed out when you get too close - my first guess
is that they've got a way to adjust the brightness of the IR projector
(or simply replaced the laser with a less-powerful one). Or is it
that the individual dots are too large and collide with each other if
they haven't had a certain distance to spread out over? (would a
weaker laser resolve this, or would you need a different diffraction
grating?)

-Drew

Joshua Blake

unread,
Nov 22, 2011, 10:33:08 PM11/22/11
to openk...@googlegroups.com
I doubt that they are updating the sensors significantly in this revision, particularly the IR projector and IR camera. If anything, they might update the RGB sensor to be a higher quality component with lower noise, but that's it.
 
The raw Kinect data that we see in libfreenect and OpenNI provides data as close as about 400-500mm, which is the same range that that post suggests will be possible with the new Kinect. When questioned about the Kinect SDK's cut off at 800mm, the Kinect SDK team reported they are concerned about providing high quality data.
 
Drew is probably right about the nature of the problem. It might be that under 800mm, the depth-rgb calibration is less accurate, or the depth value itself becomes more and more inaccurate from the true value. In the link they describe it as a new firmware that enables this feature in "Near Mode". This would indicate they probably have a different set of exposure settings for the IR camera that reduces the washout when the IR dots are so close together.
 
I'll ask some folks on the team if they can provide any additional technical details or clarifications.
 
Thanks,
Josh

Kyle McDonald

unread,
Nov 22, 2011, 10:37:37 PM11/22/11
to openk...@googlegroups.com
more technical details would be very welcome!

in my experience, near-range sensing seems to fail for two reasons:

1 overexposure. a lower power laser would help. maybe even an ND filter.
2 defocus. this requires changes to the optics.

Jonathan Gennick

unread,
Nov 23, 2011, 7:18:51 AM11/23/11
to openk...@googlegroups.com
On Tue, Nov 22, 2011 at 10:01 PM, drew.m...@gmail.com <drew.m...@gmail.com> wrote:
Possible.  Or maybe the quantity of people asking Microsoft for this
more recently prompted them to contract PrimeSense to do a
modification.

However it's done, a near mode makes a lot of sense. Microsoft is supporting the device now for the PC, right? The logical scenario seems to be to put the device atop one's monitor, or on the desk in front of the monitor. I can see where there would be much demand for a close-in mode.
 
Jonathan Gennick
http://gennick.com/bio.html
Brighten the corner--where you are!



Carlos Roberto

unread,
Nov 23, 2011, 8:35:17 AM11/23/11
to openk...@googlegroups.com
Regarding the new firmware, is it possible that we will be able to update our current Kinect? Not good idea if we will need to buy a new hardware.
Cheers.
--
Carlos Roberto
Software Eng. Consultant @ IBM
My Blog
My LinkedIn
follow me @ twitter
My Youtube videos

drew.m...@gmail.com

unread,
Nov 23, 2011, 12:01:54 PM11/23/11
to openk...@googlegroups.com
On Wed, Nov 23, 2011 at 5:35 AM, Carlos Roberto <carl...@gmail.com> wrote:
> Regarding the new firmware, is it possible that we will be able to update
> our current Kinect? Not good idea if we will need to buy a new hardware.

Possible, depending on how Microsoft implements this. Even if it
doesn't require new hardware, we'll still have to work out the
flashing procedure - there's no guarantee that it's the same one used
by OpenNI's Sensor.

Since even USB filter drivers can't log the USB transfers from the
Kinect (it shows up as a different device type), this may require
dumps from a USB hardware analyzer. /me waits patiently for his
OpenVizsla

-Drew

Reply all
Reply to author
Forward
0 new messages