RESULTS
The algorithm was implemented using the OpenCV library in a P.C using Windows Vista, Different tests were made using distinct planes for the reference image: Luminance, Hue and Cb-Cr. A very white saturated gray version of the input image is superposed with the birds. The thresholds for the generation of the binary reference map were manually adjusted, the following videos show different configurations of the system.
Birds driven by luminance:
In this configuration the black and white image is taken by thresholding the luminance plane, so darker areas appear as black and birds tend to go to these regions. In this configurations birds form recognizable shapes, but they look dark most of the time.
Video 1:
Video 2 (With smaller birds)
Birds Driven by hue:
When the hue plane is used as reference to generate the BW image, the manual threshold can be selected in order to differentiate skin colors from the rest of the scene. With this setup colorful birds can be seen in clusters over skin regions . (the results using Cb-Cr were very similar .)
