Tracking Soccer Players
Abstract
In this talk I'm going to present the problem of tracking soccer players
during a game, which consist in finding the position of each player in
the field at each instance of time. This is a challenging problem because of
different factors, such as, quality of image sequence, occlusion of the
players, the need to use many cameras in order to cover the whole field,
change of illumination, problem with shadows, camera movement, etc. Tracking
soccer players is of the great interest, because the information obtained
maybe used for tactical variation that a team can assume during a game,
it can contribute for improving the training of players of different positions,
the measurement variables such as total distance and velocity may be
associated to the physiological variables and to technical and tactical
information. The information may also be useful for annotation and for
computer animation, giving the possibility to generate virtual cameras.
We attack this problems using at least four cameras in order to cover the
whole field and to have a reasonable size of players in the image sequence.
The problem we divide in two parts: The first is the segmentation, which
consist in extracting the players from the image. The segmentation gives
as result a set of blobs in each image. In the second part we use this
information for constructing a graph, where the node are the blobs and the
edges are the distance between the blobs of two consecutives frames. Each
node contains all the information such as the color, area, size, position,
etc. For tracking players we use this graph and search the minimal path to
find the trajectory. The edges are weighted using the color similarity
and the direction of the trajectory. I'm going to present some results that
we get using this approach.