🚀 go-pugleaf

latone@trixie.ide.com (Joe Latone) writes:


>The purpose of the system is to try to find and track a baseball
>automatically by digitizing and processing the video as fast as
>possible.

>The problem with the system is trying to deal with the various natural
>lighting conditions and backgrounds, combinations of which cause the
>white ball to get lost in the dark dirt/grass background by the time
>the video signal is digitized and processed.  I need a way to maximize
>the "brightness/whiteness" of the ball and maximize the

Sounds like a background subtraction problem to me. Just subtract a
short-time average (5 secs should be ok) from your current data.
Unless the ball is stopping on the field (and you can detect that),
this should yoield good contrast. Of course the camera would have to be
mounted. Since the ball will be more or less of the same size in
the picture, you can use a matched filter on the bckground subtracted
data. This is, for every point integrate a ball-shaped surrounding
area. The highest integral (if your ball's bright) wins.
Thirdly, maybe using only the green data is not too good an idea.
If your filed is brown, the blue data, or even more so, all data
should be better.