New OpenNI/NITE/Sensor release!

[Ziv Hendel]

Hi Everyone!

The new release will come out very shortly, it's being uploaded to the
servers at the moment... :-)

We're very sorry that it took so long and for all the lack of
communication with the community and we're doing our best to improve
it all in the very near future!

Here are some of the new features:
* The license key is no longer required for NITE! (But for now it was
only removed in windows, we'll fix the linux based versions soon as
well...)
* Added a brand new JAVA wrapper and samples.
* Many many bug fixes/memory leaks/speed improvements in both NITE &
OpenNI.
* NITE calibration is now much faster! It takes about half a second
and we're working toward completely removing it in the future...
* Added new callbacks in NITE:
1) Focus gesture
- GestureIntermediateStageCompleted event is called for the wave
gesture after each wave. For the click gesture, it is called every
10cm along the Z-axis.
- GestureReadyForNextIntermediateStage event is called for the
wave gesture when the user can change direction. For the click
gesture, it is called when the user moved his hand enough and can
start retracting.
2) Hand Tracking
- HandTouchingFOV is called when the hand is close to the FOV.
3) Scene Analysis
- Fixed bug when scene is occluded (no depth in most of the scene
due to occlusion) - now users keep being tracked.
4) User
- UserExit event is called when a user is occluded (number of
pixels fall beneath a threshold) and touch the side.
- UserReEnter event is called when a user that exited before can
now be seen.
5) Skeleton capability:
- CalibrationInProgress. Called each frame when calibration is in
progress, with a current status.
6) Pose Detection capability:
- PoseDetectionInProgress. Called each frame when pose detection
is in progress, with a current status.
- PoseDetected. Replaces the deprecated PoseStart.
- OutOfPose. Replaces the deprecated PoseEnd.
* The skeleton and hand tracking algorithms are much more robust and
many problems were fixed.
* Initial support for the Android & ARM platforms.
* OpenNI API change: Running XML scripts is now done via ScriptNode
object, and this object is responsible for the lifetime of created
objects.
* OpenNI API change: From now on, context object works with reference
count:
1) InitFromXml() was marked deprecated. Now has another overload
with takes the script node as output.
2) RunXmlScript() was marked deprecated. Now has another overload
with takes the script node as output.
3) RunXmlScriptFromFile() was marked deprecated. Now has another
overload with takes the script node as output.
4) Shutdown() was marked deprecated. Release() should be used
instead (or let dtor do its stuff)
* Fixed a crash in shutdown of .NET & Unity programs.

The full change log will be out when the sources are updated on
GitHub...

Ziv.

[Ziv Hendel]

Okay, the new version is available to download from www.openni.org...

Enjoy!!

We'd love to get your feedback about all the new features or if you
have any problems...

Thanks,
Ziv.

[Andreas Binder]

Great news, I was looking for new the android stuff, is it only available on github? have not seen the new bits yet ...

--
You received this message because you are subscribed to the Google Groups "OpenNI" group.
To post to this group, send email to openn...@googlegroups.com.
To unsubscribe from this group, send email to openni-dev+...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/openni-dev?hl=en.

[Kunal shah]

Hello i am new to asus xtion i need to develop one motion sensing technology with .net c# can you help me to start...


--
" Monitor alone uses more than half of computer's total power. I choose to switch off the monitor every time I take a break."

Please consider your environmental responsibility before Printing.

Save paper, save water
Say no to Plastic

[Jinhan Lee]

Thanks for updating and taking a good step.

I have just had a quick test and checked if the segmentation fault
problem is gone. My program just crashed.
I think that the problem is still existing. Would fixing this bug be
in the next release?

Thanks you.

Program received signal SIGSEGV, Segmentation fault.
0x00007fffe7e01862 in Segmentation::checkOcclusion(int, int, int, int)
() from /usr/lib/libXnVFeatures_1_4_0.so
(gdb) where
#0 0x00007fffe7e01862 in Segmentation::checkOcclusion(int, int, int,
int) () from /usr/lib/libXnVFeatures_1_4_0.so
#1 0x00007fffe7e09174 in Segmentation::detectionByFg() () from
/usr/lib/libXnVFeatures_1_4_0.so
#2 0x00007fffe7e12b37 in Segmentation::update(Array2D<unsigned short>
const&, MotionDetectorByEdges const&, Floor const&, Farfield&,
std::vector<Vector2D<int>, std::allocator<Vector2D<int> > > const*) ()
from /usr/lib/libXnVFeatures_1_4_0.so
#3 0x00007fffe7df84a9 in SceneAnalyzer::Update(xn::DepthGenerator
const&, std::vector<Vector2D<int>, std::allocator<Vector2D<int> > >
const*) ()
from /usr/lib/libXnVFeatures_1_4_0.so
#4 0x00007fffe7ddd541 in XnVSceneAnalyzer::UpdateData() () from
/usr/lib/libXnVFeatures_1_4_0.so
#5 0x00007fffe7debeae in XnVSkeletonGenerator::UpdateData() () from
/usr/lib/libXnVFeatures_1_4_0.so
#6 0x00007ffff7b953f8 in xnUpdateDataImpl(XnInternalNodeData*) ()
from /usr/lib/libOpenNI.so
#7 0x00007ffff7b97a0d in xnUpdateTreeImpl(XnProductionNodesSet*,
XnNodeInfo const*) () from /usr/lib/libOpenNI.so
#8 0x00007ffff7b97843 in xnUpdateTreeImpl(XnProductionNodesSet*,
XnNodeInfo const*) () from /usr/lib/libOpenNI.so
#9 0x00007ffff7b98b66 in xnWaitNoneUpdateAll () from /usr/lib/libOpenNI.so
#10 0x00000000004973c7 in xn::Context::WaitAndUpdateAll() ()
#11 0x0000000000495427 in CVNIGrabber::update() ()
#12 0x000000000048961d in main ()

[Ziv Hendel]

GitHub sources are now fully updated as well...

Ziv.

[Ziv Hendel]

It's still very preliminary and a fully working version will only be
out in the future...

Ziv.

[Ziv Hendel]

Is there an easy way for us to recreate this issue?
I'd like to take a look at it.

Thanks,
Ziv.

[Jinhan Lee]

Usually, how I tested is repeating two things: getting into and
getting out of the scene. It is kind of a stress test.
Sometimes it happens in one trial.

Jinhan

[Zeraphil]

Can we read the calibration data now?

[James Walker]

On 7/13/2011 12:50 PM, Ziv Hendel wrote:
> GitHub sources are now fully updated as well...

The pull requests are still open. Have they just been ignored?

--
James W. Walker, Innoventive Software LLC
<http://www.frameforge3d.com/>

[MichaelK]

Anyone tried it with the Kinect? Or do we have to wait, till a new
kinect-sensor-driver will be released?

[pod]

MichaelK, seems to work fine with the current avin driver :D

On 14 jul, 10:32, MichaelK <email.kuz...@googlemail.com> wrote:
> Anyone tried it with the Kinect? Or do we have to wait, till a new
> kinect-sensor-driver will be released?

Ziv, one thing I noticed, when using LoadCalibrationDataFromFile in an
user that has his arms occluded (like behind his body), for instance
in NiUserTracker example, the algorithm "imagines" his/her arms are
horizontal, in a "+" shape. When finally the hands are in the FOV of
the sensor, then the recognized skeleton slowly (takes like a second
or so) adapts to the correct position, even if I set the skeleton
smoothing factor to 0 (does it default to something else now?).
Previously it was pretty much instantaneous.

[MichaelK]

What are the following dlls for?
OpenNIFilter.dll
Policy.1.3.OpenNI.Net.dll
Policy.1.4.XnVNITE.Net.dll

And why are the callbacks "userExit" and "UserReEnter" not available
in the documentation?

[pod]

They are, in the UserGenerator page. For instance,
xn::UserGenerator::RegisterToUserExit

[Ziv Hendel]

Thanks Jinhan!

We have all the info we need and we're going to try to reproduce and
fix it on our Linux64 test machines.

Ziv.

[Ziv Hendel]

Hi James!

I promise we're going to go all over the requests on GitHub real soon
and integrate the ones we can :-)

Sorry that it took so long,

Ziv.

[Ziv Hendel]

Hi Pod!

We've just fixed this bug and a new version will come out in Sunday...

There's a quick fix for it here:
http://groups.google.com/group/openni-dev/browse_thread/thread/912685f92ccb61a8/25392ec91d3c80ac#25392ec91d3c80ac

It will degrade the skeleton quality by a bit but it's only until the
proper fix is out in Sunday :-)

Thanks again,
Ziv.

[Ziv Hendel]

Hi Michael!

OpenNIFilter.dll is a DirectShow plugin that is still a work-in-
progress. It makes the sensor appear as a webcam in skype, messenger,
etc. It only allows video for now and audio will be added in the
future.

All the Policy*.dll files are needed to maintain backward
compatibility with previous .NET wrappers so apps that were compiled
with older OpenNI,NITE versions will still work.You can read about it
here: http://msdn.microsoft.com/en-us/library/7wd6ex19.aspx

Ziv.

[MichaelK]

Cool. Is there a guide how to use the Kinect with Skype? My Skype
could not find a "webcam".

[timmb]

Thanks for all of your hard work in the new version!

The calibration is an awful lot better (it's quite apparent when
playing recordings made with the previous version as we seem to be
hanging around in the psi pose for a long time after the skeleton has
already been found).

However, I'm finding that the confidence ratings per joint to be a lot
less reliable in this version - a lot of joints are incorrect and
being reported as confident whereas in the previous version they were
being reported as unconfident.

Do you know if there are any plans to improve this? Or any settings
similar to the above that can overcome this in the meantime (NB I'm
not using XML for initialization but presumably any XML options can be
replicated in code).

Tim

[Ziv Hendel]

You need to run "regsvr32 OpenNIFilter.dll" to register it in the
direct show filter list, but like I said, it's not very useful without
audio :-)
(If you want to uninstall do: regsvr32 /u OpenNIFilter.dll)

Ziv.

[Ziv Hendel]

Hi Tim!

Do you have any recordings that shows the degraded quality and
confidence rating problem in the skeleton joints?
That would be very useful for us!

Thanks,
Ziv.

[Olivier]

Hello Ziv,

> * NITE calibration is now much faster! It takes about half a second
> and we're working toward completely removing it in the future...

Does this mean that you aim at getting completely rid of the -
currently mandatory- psy pose from the user at the beginning ?
In other words, having quite a similar behavior as MS SDK in its
current state ?

This issue, I guess, is one of the most important feature that people
are now expecting from the OpenNI SDK.
For us, this is an important information since we switched to MS for
the only reason of this initialization.
Would you have a roadmap in mind for this ? (Rough approximation is
enough: is it a mater of 1, 2, 3 quarters ?)

Many thanks,

Olivier

[Amir Hirsch]

Hi Olivier,

There are bindings for the KinectSDK to OpenNI available here:

https://www.assembla.com/code/kinect-mssdk-openni-bridge/git/nodes/

You'll need to modify modules.xml and comment out NITE or modify the SampleConfig.xml and add a Query as specified in the INTALL notes.

I've tested it with OpenNI's UserTracker and it works without calibration.

Amir

[Olivier]

Hello Amir,

I'm not sure I understand: can I use the KinectSDK bindings for initialization of the skeleton and then use the tracking of OpenNI ?
I think OpenNI tracking is much better, and I still want to keep using this one.

For the future, I plan to use the Asus Xtion sensor because of its wireless capacity.
It is important for me to stick to the OpenNI and get rid of the psi pose initalization (therefore my question about the roadmap to Ziv).

Thanks,

Olivier

[timmb]

Hi Ziv

I do but it's 1.76 gigs. Do you have a dropbox folder I can put it in?

The reason I'm getting the issue is because the people in the video
are standing sideways. Obviously you can't expect a skeleton for what
you can't see but if you compare between this version and the previous
you'll find the new version is reporting incorrect joints that are
sometimes flickering between two different positions but the
confidence rating is always 1. In the previous version the positions
may not have been any more accurate but it dropped the confidence
level down a lot more dependably.

The confidence levels are pretty important for my application as I'm
merging multiple skeletons from different cameras.

Is there any way I can change this setting from within the code rather
than XML:
<Node type="User">
<Configuration>
<Property type="int" name="SkeletonHeuristics"
value="0"/>
</Configuration>
</Node>

I've tried using SetIntProperty on the user generator but it does not
recognize the property name.

Also, is there a reference somewhere that has all of the properties
that can be set?

Cheers!

Tim

[Ziv Hendel]

Hi Oliver!

The answer is yes, we are working hard to completely get rid of the
calibration.

I can not tell you the exact timeline but it will take some time (a
couple of months) until the feature is fully operational.

Ziv.

[Ziv Hendel]

Thanks Tim!

We have fixed the bad 1.0 confidence bug and it will be released in
the version tomorrow...

Ziv.

[Ziv Hendel]

Hi Jinhan!

Thanks for all your help, we can now easily reproduce this crash on
linux 64-bit machines are are working on a fix..

Thanks,
Ziv.

[Jinhan Lee]

I am glad to hear that. Thanks for working on this bug.

Jinhan

[Olivier]

Awesome !!!
Thanks for taking the time to answer..... and for the great work !

Olivier

[MichaelK]

Hi, will the new version be released in the next 8 hours?

[Ziv Hendel]

Yes it will come out today :-)

Something like 8-12 hours but hopefully sooner.

Ziv.

[Ziv Hendel]

Hi Tim!
We just released a new version that fixes this issue... If you have
the time can you possibly test and confirm that it's all good now?
Thanks,

Ziv.

On Jul 16, 6:37 pm, timmb <tim.murraybro...@gmail.com> wrote:

[MichaelK]

Hi Ziv,

the OpenNIFilter.dll is missing in the 64Bit package. Will it be
included in future releases?
I tried it with the 32Bit package, but it did not work, because I am
using a 64Bit OS.

Thanks.