I think that I might want to try out Canny’s edge detection algorithm to identify the goal posts on the negative of the flood filled image (meaning remove the grass and ball).  I think the main problem will be determining which lines are the goal posts.  Mainly because the robots don’t always see the entire goal.  Sometimes it is just one side or the top.  Also, the image can be very noisy and lines may be jaggy since the head is moving.  Also, the algorithm needs to be very fast.  So, if the Canny method can reliably find edges around the goal posts the main problem will be to isolate them.

Some details on Open CV Canny.

I found this document that a robocup spl team published on their goal detection.  They used a Hough transform.  They made clear that if the background was similar in color to the goal post that it wouldn’t work.  So, I’m not sure if it will work.  Especially at a distance the line due to the curve of the post will essentially blur into the background.  That is why I’m thinking that maybe a Gabor filter that will detect the gray/black goal netting will be a better approach?  However, this at a distance may still pose a problem… Really we just need to try it out!