The basic idea of our approach is to separate static and dynamic
parts of the scene and process them each using specific algorithms
[7]. Our method works for wide field shots and exploits assumptions
about the image content. The overview of the pipeline is shown Figure
1. First, each input image is segmented into static background
and moving players. Then, a background panorama is constructed
from the whole shot using a classical mosaicking approach, assuming
a fixed rotating camera. From this, a depth map is created for the
whole panorama using assumptions about the planar structure of the
field, and a heuristic, but sufficiently accurate model for the background.