From: Christoph on
If you mean this Function: http://www.mathworks.com/access/helpdesk/help/toolbox/images/hough.html

This is only for line-detection. I need a modified Hough-Transformation like this:
http://matwbn.icm.edu.pl/ksiazki/amc/amc18/amc1818.pdf
From: ImageAnalyst on
You still haven't said what you want to know. Let's say that you find
the centroid and diameter of all the disks. Then what?
From: Christoph on
I need the coorinates of the center of the 4 circles.
The 4 circles are in reality 4 spheres, which are captured by a camera, for tracking.
This should be able in realtime, but this is not the problem yet.
I just need the center-coordinates, not a complete solution of the tracking-problem.
From: ImageAnalyst on
If you're tracking, what will you do if there is substantial overlap?
If you split apart blobs that aren't perfectly circular you could find
the centroids but it will be off a little bit due to the overlap. But
if there's overlap you're going to have a tough time anyway even using
Hough methods. I was wondering how you're going to handle that. How
much accuracy do you actually need anyway? What you think you need
may not be possible for all cases (e.g. substantial overlap).
From: Christoph on
I need as much accuracy as possible.
The detection of the centroids when there is no overlap is simple.
Just for the overlap situation I'm thinking for a good solution.

I thought about movement estimation with vector (like in MPEG4) to predict the movement before and in the overlap. But this is too heavy for me.
So I decided to try it with the Hough-Trafo, because I found good examples in the internet where it worked good in overlap situations.
A complete overlap is the worst case, that can happen. I haven't made any thoughts about it. If there are 3 centroids of the 4 detected, then the controids of two would be approximately the same. For this I will need the position of the centrodis in the previous frame.
But at now I'm still at one frame not at a video...