robustly finding a region in a photo using fiducial
Hello, I'm new to feature detection, but have some limited experience with cv2 and Python.
I want to detect if a set of rods are protruding from a surface in a machine. The rods will move and I'll take a photo, then do the analysis.
I think the way to do this is to stick on some fiducial markers and use opencv to detect them, work out the camera pose, then extrapolate where the rods should be. I can detect the rods by looking for their shadows (dark spot if a rod is up).
I've tested the idea by calibrating my camera with a chessboard, then sticking the chess board next to the rods. I can find the chessboard, then extract the ROI and detect the shadows pretty well.
However, I would have thought a fiducial on each side of the row of rods would give better results, especially as I expect to have more rods to detect in the future.
I don't really know how to go about detecting multiple fiducials and getting opencv to yield me the pose.
Can anyone give me some pointers?
Thanks, Matt
OK, I've worked out how to track a pair of fiducials with opencv, see here: https://github.com/mattvenn/fiducial However, I don't know how to find the center of the fiducial in order to use as a reference point. It seems to me that the way the feature matching works will result in a set of features whose center point isn't guaranteed to be the center of the printed fiducial...
OK, I couldn't work out a way of combining the information I got back from the fiducials into one homography to then do the transformation of the rod positions. So I made one larger fiducial that has an empty space in the middle for the rods. That way I can do it all from one transformation. I'd still like to know how to use multiple, separately tracked fiducials to create a uniform geometry.
See my blog post for more details http://www.mattvenn.net/2014/02/08/using-a-computer-to-read-braille/