Basketball 3 points line recognition

I ’m doing a project and I’m trying to detect the 3 points line in a basketball court. The frames are tv frames, taken from mid height and lateral view.

I’ve calculated an homography. Over the image after applying the homographic transformation I’ve run canny and circular hough, and tried to find an ellipse that will catch the upper part of the court, but I can’t get anything that is good enough. Does anyone have a suggestion to what could I do?

To build training data? Or are you looking for an architecture or network idea to implement?