blob

Blob Tracking Plugin (Carasuelo OpenCV 0.02a) (Composition by benoitlahoz)

Author: benoitlahoz
License: Creative Commons Attribution
Date: 2012.01.23
Compatibility: 10.6, 10.7
Categories:
Required plugins:
Kineme GL

Hi,

A new plugin in the Carasuelo OpenCV bundle : Blob Tracking. I implemented the cvblob library and externalized a lot of parameters to prepare the image, and to set the tracking accuracy. It's supposed to work with 10.6 and 10.7 in 32/64.

It outputs a structure of blobs with coordinates, angle, and tracking datas (has been active for x frames, has been inactive for x frames, exists since...).

An image output is available, with the possibility to output the intermediate image. It runs quite fast on my MBP 13" even when the frame isn't resized (640x480).

The 3 example comps (Blob Tracking, Contours and Presence) are documented. I have to work again on these 2 plugins to make them fit a CVImage class I wrote to simplify OpenCV use.

There's a big issue on the image dimensions, as I have a trouble with "special" width on input images, but in most case it should be ok for everyone. I'm working on it, but if someone has an idea...

Please tell me if it works for you !

I would be glad to see what you do with it.

OpenCV Blob Tracking (built for Kinect)

boumabouma's picture

Hi everyone,

I'm pretty new here, but this website/community has helped me a lot during my last project so I figured it's time to donate back :).

During this last project I was building an installation that used Kinect to detect your silhouette and hands, with which you could then control it. Like so: http://vimeo.com/31154077

The reason we used blob tracking is because Synapse was of no use to us due to the calibration every user has to go through. Please note that despite that, Synapse is more accurate and less heavy on your CPU :)

Long story short, I've built this plugin I wanted to share with you. It uses OpenCV and CvBlob to detect the blobs and outputs their position and size as a Structure. The minimum blob size and "movement threshold" (to smooth out glitchyness from the depth image) can be configured in QC.

However, I'm not very experienced with 'cmake', 'make' and install-name, so I've got some problems with including the openCV lib in the plugin. What still needs to be done is:

  • Build (cmake/make) the opencv and cvblob libraries for all required platforms (currently only Snow Leopard 64bits is included I think)
  • change the install-name of the libs so QC can find them from the plugin bundle.

I've been fiddling around with it, but no luck so far. If anyone could give me some pointers or help out that would be swell.

For now this means the plugin wont work without some extra work: placing the openCV and cvblob libraries in /usr/local/lib I've included the libs for 64bit Snow Leopard (in the zipfile in /usr-local/lib), simply copying them and the plugin to their designated place should do the trick. Obviously, building the libraries on your system will also work and you'll know for sure the libs are compatible.

Let me know what you think.