Hello everyone.
I am having a hard time finding generic computer vision forums online, so I'll post this problem here knowing that this is not a specific openCV related question.
I'm trying to detect buildings from aerial images on a forest environment.
The images are like this one: http://postimg.org/image/erxow7c2p/
The method should be robust enough to detect a building in most frames. One advantage is that the route is always the same, so the buildings are actually always the same ones, but can appear in different perspectives.
I've already eliminated the sky from the images and most deep forest areas, and more can be done through simple color segmentation, but I don't think that blob analyses will be robust enough for this application.
I've tried template matching but it doesn't work at all. I was thinking maybe a machine learning approach, but I really need this method to be fast (ie 10 fps).
Any thoughts?