To perform rectification, the
yardlines, sidelines, and hashmark lines are found using the Canny edge detection
algorithm and line-finding software from the INRIA research lab in France.
Intersection
points between lines are identified and matched with the field model at each frame. The
intersection points can be recovered regardless of the state of the players on the field.
Even when player is directly occluding an intersection point, as indicated by the arrow,
the intersection point will still be recovered with sub-pixel accuracy.
The field model is initialized to the first frame of the video sequence (by manually
marking four corresponding points), after which the intersections are tracked
automatically.
Once corresponding intersection points are recovered, the rectification is performed using
a simple homographic transform (perspective four-point transform).
|