This is a gstreamer plugin for recognition
of specific images in a video stream.

The video stream can be a video file 
or a video camera capture
and the collection of images 
can be changed dynamically 
by entering the new directory at the prompt
when executing a script.

The plugin uses OpenCV for the recognition of images.

To use it on Ubuntu, first install the dependencies (on maverick) :
> sudo apt-get install gstreamer-0.10 libgstreamer0.10-dev libgstreamer-plugins-base0.10-0 libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev gstreamer0.10-plugins-good gstreamer0.10-plugins-bad gstreamer0.10-plugins-ugly-multiverse gstreamer0.10-plugins-ugly gstreamer0.10-pulseaudio gstreamer0.10-ffmpeg libcv-dev libcvaux-dev libhighgui-dev python python-gst0.10 python-gtk2 nodejs ffmpeg v4l-conf mplayer

Then compile the plugins with :

> sudo ./autogen.sh
> sudo ./configure --prefix=/usr
> sudo make
> sudo make install

Then go to the script directory and launch the analysis of a video with :

> cd scripts
> python videofeed.py ../data/slides.mp4 ../data/images

or the analysis from a video camera (v4l2) with :

> cd scripts
> python v4l2feed.py /dev/video0 ../data/images

The scripts take two additional optional arguments,
that are optional because they have default values :

* algorithm [ 'match' or 'surf' ( default ) ] : algorithm used for detection
* minscore [ 30 by default ] : minimum score in the comparison with an image to detect a true positive.

The minscore represents different things according to the algorithm used :

* for the 'match' algorithm, it's the correlation factor between two images
and should be set to a value in [0,1],
a good default value would be 0.90.

* for the 'surf' algorithm, it represents the number of common SURF points
between two images and should be set to a value like 30,
the default used here as 'surf' is the default algorithm
( the only one that is not sensible to changes in luminosity,
distance from the screen, orientation, ... ).

For any more information or support, contact ydegoyon@gmail.com.
Developped for Kognate, inc.