open source software for optical flow

[Andrew Barton]

Hi All,

I recently heard about some open source code that is quite efficient for the problem of optical flow.
http://en.wikipedia.org/wiki/Optical_flow

Such an algorithm could be used in a low-mass system for determining the horizontal velocity of the lander during its approach to the surface.
 
Once you have the optic flow you can get the motion state. Since we will have gyros onboard, the lander knows the angular rates, and with the radar altimeter it knows the altitude. Thus it should be possible with just a single camera toback out the translational velocities.

 

The 'good features tracker' algorithm finds points of high contrast in the image then you typically use the Lucas Kanade algorithm to compute optic flow at these points here are some links:

http://www.ces.clemson.edu/~stb/klt/

http://robots.stanford.edu/cs223b04/CS%20223-B%20L9%20Optical%20Flow.ppt

 

these algorithms are both implemented in the OpenCV C++ library:

http://sourceforge.net/projects/opencvlibrary/

I'm told that this method is very low in computation (basically involves just weighted summations of optic flow measurements) but may be a bit noisier than other methods. Other methods use iterative least-squares solutions or kalman filtering to get more accurate state estimates but require more computation. If you have a computer in the range of 500MHz or so then i think these methods may be too computationally intensive.

Maybe somebody at LN is interested to experiment with this open source code?

Cheers
AB

---

Express yourself instantly with MSN Messenger! MSN Messenger

[Andrew Barton]

Following on from my previous message, I found this interesting paper
describing an approach similar to the one I had in mind.

http://articles.adsabs.harvard.edu/cgi-bin/nph-iarticle_query?2006ESASP.606E..80T&data_type=PDF_HIGH&whole_paper=YES&type=PRINTER&filetype=.pdf

On Oct 14, 12:51 pm, Andrew Barton <andy2...@hotmail.com> wrote:
> Hi All,
>

> I recently heard about some open source code that is quite efficient for the problem of optical flow.http://en.wikipedia.org/wiki/Optical_flow

>
> Such an algorithm could be used in a low-mass system for determining the horizontal velocity of the lander during its approach to the surface.
>
> Once you have
> the optic flow you can get the motion state. Since we will have gyros
> onboard, the lander knows the angular rates, and with the radar
> altimeter it knows the altitude. Thus it should be possible with just a
> single camera toback out the
> translational velocities.  
>
> The 'good features tracker' algorithm finds points
> of high contrast in the image then you typically use the Lucas Kanade algorithm

> to compute optic flow at these points here are some links:http://www.ces.clemson.edu/~stb/klt/http://robots.stanford.edu/cs223b04/CS%20223-B%20L9%20Optical%20Flow.ppt

>
> these algorithms are both implemented in the OpenCV
> C++ library:http://sourceforge.net/projects/opencvlibrary/
>
> I'm told that this method is very low in computation (basically
> involves just weighted summations of optic flow measurements) but may be a bit
> noisier than other methods. Other methods use iterative least-squares solutions
> or kalman filtering to get more accurate state estimates but require more
> computation. If you have a computer in the range of 500MHz or so then i think
> these methods may be too computationally intensive.
>
> Maybe somebody at LN is interested to experiment with this open source code?
>
> Cheers
> AB
>

> _________________________________________________________________
> Express yourself instantly with MSN Messenger! Download today it's FREE!http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/