> I am developing a markerless gait analysis system using image processing
> techniques to locate the joint centres for tracking. I am hoping to achieve
> this by fitting a simplified model of the leg (literally just four lines in
> the 2D plane) to the actual footage of the patient's leg. I have managed to
> successfully pick out the patient from the background of the film clips but
> I am having trouble actually fitting the model to the footage without manual
> intervention. I was wondering if anyone has tried this technique before and
> how they managed this particular problem. I am really looking for source
> code or algorithms at this stage as I have read several papers on the
> theory.

I'm struggling with the same problem even if applied to the whole body.
I can provide you some links I've found useful.

Hough transform

Adrew Blake
Introduction to Active contours and Visual Dynamics

Judit Verestóy and Dmitry Chetverikov
Fearture Point Tracking Algorithms

Image Processing Fundamentals

The Condensation Algorithm Home Page

Jaewon Shin
CONDENSATION for Visual Tracking

I think that the techniques that better fits your problem is Condensation.
Anyway, as you can see some papers are from a site called CVonline. In the
same site I found a short paper about gait analysi, really interesting
albeit short. It says that also the computing of Optical Flow is an
interesting technique to collect info about gait.
Try this other link too:
Stan Birchfield
KLT An implementation of the Kanade-Lucas-Tomasi feature tracker version
Then on the Intel site you can find a really huge library for image
processing, completely free and really powerful.
Hope all this is useful to you.

Riccardo Trocca
Laboratory of Musical Informatics
University of Genoa

To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l