Analysis of team sports

Trajectory-Based Analysis of Team Sports

Game segmentation

A Gaussian mixture model is used to model the state of the team centroid for each individual phase of the game. To obtain the parameters of the model, we apply the Expectation Maximization algorithm on the manually segmented data (See [3] for details).



Figure 1: Team centroid calculated from positions of active players.
lkhdsfi
	(Move the mouse over the images)
(a)	(b)	(c)

Figure 2: Manually annotated trajectories which were used to build the game model and automatically obtained statistical model (move mouse over the images). Every game phase is represented by a two-component Gaussian mixture model. (a) offense (b) defense (c) time outs half-times and other major interruptions.

Segmentation result

Using the procedure described in [3], we were able to segment different team games with a very high degree of accuracy. The obtained segmentation accuracy varies between 85 % and 95 % in basketball and 90 % to 97 % in handball. The movie bellow shows a short sequence from a real basketball game played at the Slovenian championship finals. The markings above the court represent the annotations which were performed by the basketball expert (manual annotation) and the annotations that were obtained by using our segmentation approach (automatic annotation).


We have upgraded our segmentation approach so that it can also handle well different unusual game situations such us tracking mistakes, player exclusions, substitutions and injuries (See [4] for details).

The bellow examples show how the the presented method can handle different unusual situation in basketball and handball. (Videos are displayed at 60fps)




	Annotation coloring: *blue: blue team in defense green: green team in defense black*: time out
previous page \| next page

Selected publications:

(1) M. Perše, M. Kristan, J. Perš, S. Kovačič."Automatic Evaluation of Organized Basketball Activity". In: Michael Grabner, Helmut Grabner (eds.), Computer Vision Winter Workshop 2007, St. Lambrecht, Austria, pp.11-18. February 2007.

(2) M. Kristan, J. Perš, M Perše, S. Kovačič."Towards fast and efficient methods for tracking players in sports". In: Proceedings of the ECCV Workshop on Computer Vision Based Analysis in Sport Environments, pp.14-25. May 2006.

(3) M. Perse, M. Kristan, J. Pers, S. Kovacic."A Template-Based Multi-Player Action Recognition of the Basketball Game". In: Janez Pers, Derek R. Magee (eds.), Proceedings of the ECCV Workshop on Computer Vision Based Analysis in Sport Environments, Graz, Austria, pp.71-82. May 2006.
(4) M. Perše, M. Kristan, J. Perš, G. Vučkovič, S. Kovačič. "Temporal segmentation of group motion using Gaussian Mixture Models". In: Janez Perš (ed.), Proceedings of Computer Vision Winter Workshop 2008 (CVWW08), Moravske Toplice, Slovenia pp.47-54. 4-6 February 2008.