Dagstuhl: Day 3 Sessions 1 and 2

Today is the final day of the Computer Science in Sport Conference (Special Emphasis:Football) at Schloss Dagstuhl. The morning session discussed Media and Data Acquisition issues. The session was chaired by Daniel Link (TU Munchen).

Daniel presented first in this session. He reported on the Game Data Library Project for the Bundesliga.

The aims of the Project are:

  • Technical: better data validation and better IT infrastructure
  • Commercial

Daniel discussed the game observation process for the Game Data Library. This involves the acquisition of basic data that includes match information data, tracking data (at 25hz), event data, static video data that are used to create raw data and statistics. Daniel presented the architecture of this service to provide data flow.

He discussed the Game Data Model. Daniel presented an ontology of definitions in use in this project. This ontology has a data structure for: smart calculation; efficient processing and storing of data; and object orientation and the use of Unified Modelling Language (UML).

Daniel concluded his talk with a consideration of the challenges of this project for sport science. These included how to use the enormous amounts of data that will be generated and how to develop tools to analyse the data.

Roland Leser (Universitat Wien) was the second presenter in the session.

His topic was Position tracking as a challenge in game sport analysis. Roland’s abstract:

About 10 to 15 years ago only top level teams used computer assisted video annotation systems to analyze sport games and training sessions. The progress in hardware and software development made it happen that nowadays this technique is used even at amateur level. Pointing to another analysis technique we have at present similar conditions than in the situation described above. Very expensive video based position tracking systems are used by few of the top teams worldwide to analyze their game play and GPS-systems are applied to analyze training sessions of outdoor sports. Radio wave based tracking systems are currently not wide spread for performance analysis but they could dominate the future. By tendency radio wave sensors become smaller and cheaper in the next years and radio wave based tracking systems are much less service intensive than other systems. Looking forward, this kind of tracking system could be a worthwhile alternative to analyze games and training sessions of game sports for many teams. This presentation gives an overlook on the preparatory work of installing an industrial radio wave based tracking system (www.ubisense.net) for game sports analyses, outlines current results and looks ahead to future works.

In his talk Roland gave an excellent exposition of how to develop a position tracking system.

He noted systems such as Tracab and Catapult. He discussed radio wave systems too, including Ubisense (tag) and InMotio LPM (transponder).

In choosing a system for use in his research Roland identified these criteria:

Off the shelf availability of a system:

  • Price
  • Sensor size
  • Sampling rate
  • Accuracy
  • Robustness (hardware, signal, definition of player)
  • Application in training and game play
  • Opponent agreement within competition

Roland discussed the use of an Ubisense system. To date this system has not been used extensively in sport. He demonstrated the installation of the system in a sports hall. Ubisense has a 160hz facility that is not common with other Ubisense clients. The hall is calibrated and then checked for accuracy of measurement. The system allows some for data filtering (low pass and Kalman). Roland noted the development of software tools for the system to enable data visualisation (including heat maps) and performance analysis.

Roland shared an example of the recording movement with the system (small sided football).

The final presentation of the morning was by Malte Siegle. Malte looked at the accuracy of image recognition in dynamic situations. He shared the development of protocols to check the accuracy of image recognition in respect of Laveg and laser light measurement.

The field tests were conducted in a soccer stadium:

1. A linear run near the cameras with constant velocity. (Image detection worked well.)

2. Acceleration, stop, reacceleration in same direction. (Image detection issues arise with up to 1 metre error.)

3. Two players move towards each other and return after 180 degrees turn.  (An error of more than 1.5 metres.)

4. Circular run with constant velocity (Image detection worked well.)

Malte noted the variability in errors in these tests and discussed the impact of the player’s distance from the cameras.

Theses tests had identified the need for better static position detection and the clear differentiation of error sources (distractions). Malte did end with some very positive views about the protocols: good values were recorded and problems were identified. This research raised the possibilities of a new standard in the evaluation of image detection. Ultimately this will lead to the comparisons of different image detection systems.

The final session of the day was chaired by Martin Lames. This was an informal review and evaluation of the Conference. Everyone agreed that the Dagstuhl experience was outstanding and all participants hoped to have the privilege of returning.

Dagstuhl: Day 2 Session 3

Session 3 of Day 2 at the Computer Science in Sport Conference (Special Emphasis:Football) at Schloss Dagstuhl was dedicated to Coaching themes. The session was chaired by and introduced by Stuart Morgan (Australian Institute of Sport).

His introduction focussed on the ways coaches and scientists communicate.  I liked Stuart’s representation of the communication process.

Stuart explored signal to noise issues and presented these data from the 2010 Champions’ Trophy Tournament. These were all Germany’s ball movements in the attacking third of the pitch.

I presented  after Stuart in this session. I used Prezi to share my data from the first two rounds (n=16 games) of the 2011 FIFA Women’s World Cup. The Prezi is here. The Dagstuhl Abstract of the Talk is here and is:

This paper presents data from the games (n=16) in the first two rounds of the 2011 FIFA Women’s World Cup. Attention is drawn to patterns of goal scoring and the relationship with FIFA ranking (18 March 2011). The paper includes a profile of winning, losing and drawing in the 16 games (presented as averages). The paper is used as a stimulus paper for discussion of technical and tactical aspects of game play at this World Cup and in football generally.

Tim McGarry presented after me and looked at Human Information Processing: Penalty and Free Kicks. Tim shared some descriptive data from penalty kicks and shoot outs from European Championships and World Cups from 1976 to 2010 (1976 saw introduction of penalty shoot outs). He noted that in penalty shoot outs, penalty 4 appears to be the weakest link in a shoot out (of 5 penalties taken). Tim asked who should take this penalty? Should the most experienced (and successful)  penalty taker be allocated this role? Tim discussed goalkeeper actions in penalties and considered the options available in 400ms.

Tim discussed free kicks too. He presented descriptive data from the 2002 and 2006 World Cups and the 2004 European Championships and discussed the optimisation of goalkeeper reaction and response. He discussed the role of a defensive wall in these free kick situations.

Joachim Gudmundsson was the fourth presenter in this session. He discussed extracting and making sense of information from trajectories. He discussed his work in the defence services, with animal behaviour and in sport.

Johaness Uhlig (Universitat Wien) was the next presenter and discussed his work with the Austrian Under 17 Women’s team and with his club team.

Johannes described his coaching and his tactical approach based on a basic 4-4-2 formation. He emphasised three phases of play: attack, defence, switchover. He discussed the development of a tactical animation program (TAP) to support coaches and its use in practice. This is an interview with Johaness when he was the national coach.

The final presentation in the afternoon’s session was from Koen Lemmink (University of Gronigen) on Tactical Match Analysis in Soccer: New Perspectives? For some of the ideas discussed in this paper see Frencken, Lemmink and Delleman (2010).

Koen explored three different approaches to observing and analysing performance. He provided examples of each approach.

  • A Practice Model that uses frequencies of event data and player profiles. This is coach driven and has a focus on direct feedback. Example of direct feedback and direct streaming.
  • A Statistics Model that identifies performance indicators, notes statistical differences. This is a domain populated by mathematicians, statisticians and econometricians. Its focus is  pattern analysis and recognition.
  • A Theory Model that uses scientific insight to understand interactions and networks.

He emhasised the possibilities of building multidisciplinary teams that had a strong focus on explanations and shared rich data on positional play.

The session concluded with an extensive discussion that was stimulated by Max Reckers ideas about sharing data. This included consideration of the role of the embedded scientist in a sport setting.

Dagstuhl: Day 2 Session 2

 

Peter Lamb (TU Munich) was the first presenter in the second morning session of Day 2 at the Computer Science in Sport Conference (Special Emphasis:Football) at Schloss Dagstuhl.

Peter discussed Self-Organising Maps (SOMs) and presented Basketball data to exemplify an SOM approach to movement observation and analysis. He developed his discussion with an analysis of golf shots and concluded with a brief exploration of the use of SOMs in football.

Self-organizing maps are a type of artificial neural network useful for visualizing complex human movement coordination. The visualization of the network output can be enhanced by using colour or a third dimension to visualize data clusters, by adding a trajectory to highlight the time-series progression of coordination or by identifying which areas on the output map represent certain critical phases in the movement.

         

Jurgen Perl presented the second paper in this Dynamical Systems session. He discussed a neural network approach to formations in football and the development of a SOCCER analysis program.

This work is being reported in Grunz, A., Perl, J. & Memmert, D. (2011, accepted) Tactical pattern recognition in soccer games by means of special Self-Organizing Maps. Human Movement Science.

See also this 2009 paper by the same authors.

The abstract for the 2011 paper is:

Increasing amounts of data are collected in sports due to technological progress. From a typical soccer game, for instance, the positions of the 22 players and the ball can be recorded 25 times per second, resulting in approximately 135.000 datasets. Without computational assistance it is almost impossible to extract relevant information from the complete data. This contribution introduces a hierarchical architecture of artificial neural networks to find tactical patterns in those positional data. The results from the classification using the hierarchical setup were compared to the results gained by an expert manually classifying the different categories. Short and long game initiations can be detected with relative high accuracy leading to the conclusion that the hierarchical architecture is capable of recognizing different tactical patterns and variations in these patterns. Remaining problems are discussed and ideas concerning further improvements of classification are indicated.