 |
 |
 |
 |
 |
 |
 |
Trajectory-Based Analysis of Team Sports
|
Introduction and overview
The idea of an advanced and objective analysis system that would enable the sport experts to prepare good team tactics is not new. In fact, several different approaches to basketball analysis, from data-mining to manual anotation and analysis of player postions have been presented in last decade.
With the development of advanced video and other technologies, it is now possible to obtain large quatities of game data (Figure 1). This data should be preocessed as fast as possible to give the coach or a team expert an instananeous insight about the performance of his/hers team
 |
 |
|
(a) |
(b) |
|
 |
||
(c) |
||
Figure 1: Game trajectories. (a) Basketball, (b) Tennis, (c) Handball |
||
Game structure
We have developped a multi-step approach towards the automatic trajectory-based activity recognition that considers the structure of the team sports (Figure 2). The proposed approach reduces the time which is needed to analyze the game since the game is analyized in several steps and in each step a more complex model is used in the analysis procedure.
Figure 2: Structure of the basketball game.
The game is analyzed in three steps:
- First, the continuous game is SEGMENTED into several different phases (Offense, defense and inactive game)
- Afterwards, every segmented section of the game is RECOGNIZED using the activity templates
- Finnaly, once the tipe of the segment is established a very complex statistical model is used to EVALUATE the overall tema performance and the performace of individual players.