[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Toon-members] tag patches - kalman filter build fix, minor fixes
From: |
Gerhard Reitmayr |
Subject: |
Re: [Toon-members] tag patches - kalman filter build fix, minor fixes |
Date: |
Thu, 6 May 2010 22:56:42 +0200 |
the equivalent to CameraMeasurement is the IncrementalPose in measurements.h
I updated the documentation to reflect this, thanks for the heads up.
Incremental pose is used for an incremental measurement to the current pose.
often with camera tracking that is a simple way to incorporate a measurement.
the model is transformed into camera coordinates anyway for
prediction/projection and then optimization can work directly in the camera
coordinate frame. the resulting small update can then be used with an
IncrementalPose.
cheers,
Gerhard
On 6 May 2010, at 22:15, Ryan Pavlik wrote:
> I was just using constantvelocity.h: we used the library to create a SCAAT
> filter implementation for a camera-based tracker, so we had to produce our
> own measurement model class. Unfortunately, we're a bit stuck right now
> since the Jacobian we computed using Matlab and our measurement model gives
> us NaN all over the place when rotation is zero. Our general impression at
> the moment is that we might need another way to handle the measurement
> non-linearity rather than the Jacobian, but we haven't gone any further than
> that yet. We did look at the measurement.h code as an example of how to
> structure our own, and will probably use it to process some accelerometer
> data, but haven't implemented that yet.
>
> In the examples, a camerameasurement.h file was referenced, but there's no
> such file in the actual distributed code. Is this just a suggestion of how
> to structure things, or is there an actual implementation of that class
> somewhere that might be committed?
>
> Ryan
>
> On Thu, May 6, 2010 at 7:07 AM, Gerhard Reitmayr <address@hidden> wrote:
> Hi Ryan,
>
> Thanks for the update. I added the two small ones already to the repository.
>
> For the transformCovariance one, I have another implementation that I believe
> is better using cholesky to calculate the inverse. I will check this in, once
> I tested that it produces the same results. In this one the
> transformCovariance function is not needed, so we might not need the fixes in
> TooN.
>
> Also, I saw that all the other headers (constantvelocity, measurements.h)
> etc. are stale and do not work. I am in the process of fixing them, but just
> wanted to ask, if you are using those as well ?
>
> cheers,
> Gerhard
>
> On 6 May 2010, at 01:28, Ryan Pavlik wrote:
>
> > All,
> >
> > Here are the changes to tag, split up into logical chunks. The only major
> > one is the build fix to the kalman filter class: the other two are minor
> > details.
> >
> > Thanks!
> >
> > Ryan
> >
> > --
> > Ryan Pavlik
> > Human-Computer Interaction Graduate Student
> > Virtual Reality Applications Center
> > Iowa State University
> >
> > address@hidden
> > http://academic.cleardefinition.com/
> >
> > <0001-Compile-fixes-to-kalman-filter.patch><0002-Fix-an-unused-variable-warning.patch><0003-Fix-spelling-error-in-comment.patch>_______________________________________________
> > Toon-members mailing list
> > address@hidden
> > http://lists.nongnu.org/mailman/listinfo/toon-members
>
> --
> Gerhard Reitmayr
> Institute for Computer Graphics and Vision
> http://www.icg.tugraz.at/Members/gerhard
> tel: ++43 316 873 5082
>
>
>
>
>
>
> --
> Ryan Pavlik
> HCI Graduate Student
> Virtual Reality Applications Center
> Iowa State University
>
> address@hidden
> http://academic.cleardefinition.com
> Internal VRAC/HCI Site: http://tinyurl.com/rpavlik
--
Gerhard Reitmayr
Institute for Computer Graphics and Vision
http://www.icg.tugraz.at/Members/gerhard
tel: ++43 316 873 5082