Best method for detecting multiple white ovals on black background

asked 2013-06-20 15:18:20 -0600

updated 2013-06-20 16:29:08 -0600

I am working on a project where I would like to, each frame, get an array that contains the central coordinates of white ovalish blobs moving around on a black background, and I am curious what is the best filter to do this job? It is worth noting that I am doing this on an iPhone via the GPUImage framework (as openCV was too detrimental to the framerate), and thus do not have access to a general blob detection filter. Would something like a corner detection filter work, with proper parameters? Thanks.

Edit: Here is a youTube video showing the current detection system in progress:
http://www.youtube.com/watch?v=OxdPZ23v43Q

Note: What you are seeing here is already after I have run a number of filters on the raw feed (high pass filter for motion, luminance to increase contrast, etc). Also, this is a SO post with a little more information: http://stackoverflow.com/questions/17056571/gpuimage-detecting-average-luminosity-of-multiple-rectangular-sub-regions

Comments

You could show the example images?

BeS ( 2013-06-20 15:43:13 -0600 )edit

Thanks for updating the question. But is it possible to give us some example images like BeS suggested?

Prasanna ( 2013-06-20 23:07:29 -0600 )edit

add a comment

answered 2013-06-20 15:39:21 -0600

Prasanna

236 ●2 ●10

There are actually way too many ways to do such a thing and on top of that choosing one out of those can only be done by either knowing the problem clearly, its conditions, best case scenario, worst case scenario and various other parameters. I would consider this information not sufficient and would like to have a bit more info about the problem and what you have tried till now, what has worked, what has not, etc.

Just for the sake of argument, Let me consider a perfect video having perfect frames. Lo and Behold! there are contours and you can use their moments for description of their shape (As you say they are ovals but you never mentioned their variability in terms of parameters - eccentricity, etc.). And guess what it doesn't end here. Hough Transform, Wavelets, Phase Congruency, Derivative Filters, Connected Components, Active Contours, etc. for detecting and then Motion Templates, Phase correlation, Optical Flow, Cam-shift, Template Matching, etc. for tracking them. These are just a few. Even you can train cascades and make it a classification problem.

Now you can see how difficult it is to even begin answering your question. And these are just a few. Try giving more info about the problem and please don't wait for a solution to be provided. You are your own best teacher and you have more drive in you than all of us to work on that problem. Try searching a bit more and you will find something better as you know a lot more about the problem than us

Now don't mistake me. I am just telling you that how helpless I am or others reading your post are.

Please update this thread soon with your work. Or else I am afraid that it might be closed.

I am really sorry that I am not able to provide you with any viable solutions for your question. Those mentioned are just a few. Some might be stupid, some might be overkill for the problem in your hand. Maybe you can start exploring on these lines but dont limit yourselves to this small list.

Regards,

Prasanna S

edit flag offensive delete link

add a comment

Best method for detecting multiple white ovals on black background

Comments

1 answer

Links

Question Tools

Stats

Related questions

Best method for detecting multiple white ovals on black background edit

Comments

1 answer

Links

Question Tools

Stats

Related questions

Best method for detecting multiple white ovals on black background