Sportiers
Serokell participated in optimizing an IoT solution
for sports team training and games analysis.
In pursuit of better results, leading coaches seek innovative ways to analyze training techniques, leveraging advanced models and historical data. This is where Serokell came to assist with an ML solution.
The customer, a leading UK football team management, approached Serokell with a problem:
- How to highlight players' tactical errors and provide detailed analysis in real time.
- Additionally, they wanted to showcase tactics with multiple combinations that have been used in historical plays leading to victories.
This would help players gain a clear understanding of different approaches and best practices from world football, ultimately increasing the number of winning games.
To leverage a vast database of football matches and the ability to record and analyze gameplay in real time, Serokell developed an ML application.
This solution samples real-time recorded data, provides detailed analysis of player movements and ball trajectories, and helps the team understand how to apply best practices from other football teams to their games.
Our collaboration
Our task was to develop software capable of handling a vast amount of online and offline data to recognize similarities and predict possible outcomes when no such combination is available in the database. The client assigned us specific tasks that can be grouped into three main categories.
Create an ML model to develop a set of tactical combinations:
- Create simulations for offensive sets against specific defensive sets.
- Design approaches for attacking plays and corner kick defense.
- Establish methods for quick transitions between defensive and offensive plays.
Work out a more accurate overview of player performance:
- Develop tools to assess individual player metrics, such as distance covered, sprints, speed and overall team performance.
- Provide coaches with measurable performance indicators to enhance their evaluation of tactics.
Provide transparent and measurable
analysis:
- Count offensive vs defensive game proportions, analyze tactical combinations, and team performance to identify strengths and weaknesses.
- Predict scoring positions based on playing style to assist coaches in developing more productive game tactics.
The task resonated with many of our developers who are avid football fans themselves. This added an extra level of motivation and enthusiasm as we developed our strategy for this complex project.
Our process
Step 1
Taking into consideration recent developments, such as FIFA officially granting permission to use wireless sensors for tracking player positions and physiological parameters during matches, we proposed the placement of sensors and the utilization of real-time records.
This approach allowed us to work on data-driven analysis based on the positional data captured by the sensors. We were able to identify and focus on measurable parameters of the footballers' play. This, in turn, provided coaches with a tool for a more detailed analysis of their mentees’ performance.
Step 2
Our team conducted a thorough investigation of existing competitor solutions but found no suitable alternatives for this specific task. However, we came across several academic research studies that provided valuable insights. Building upon this research, we designed a new application that leverages Language Models (LLMs) and a database of football plays from the past 20 years.
In this application, we introduced key performance indicators, such as the speed of the ball, rotation, players' positioning, and kick strength. This enabled coaches to analyze and visualize the proper execution of kicks compared to real training session records.
Step 3
Another aspect of our work involved pattern-based modeling to facilitate a comparison between training sessions and historical records for various game events. This included analyzing pass patterns, corner kicks, and different attack combinations such as 4-2-2, 3-4-3, and 3-5-2 in winning games from databases like PubMed, ProQuest, Elsevier, and more. The goal was to assist coaches in explaining necessary corrections to team tactics during training sessions.
Strategic Modeling
By implementing "deep imitation learning" in our model, we were able to visualize effective responses to specific attack situations. This allows for identifying patterns for successful gameplay.
Additionally, we provided visualizations to illustrate how teams can respond to attacking combinations and create a stronger defense.
The incorporation of these pattern-based modeling techniques resulted in more productive training sessions and more successful game outcomes.
Serokell successfully developed software that can handle vast amounts of online and offline data, identifying patterns and predicting potential outcomes in football.
Results
We completed all the work within the established deadline. The final product was thoroughly tested and met the specified requirements. The solution was well-accepted by the customer, along with our suggestions for more advanced functionality.
The current version of the application, utilizing cutting-edge LLM technology, has already propelled the football team ahead of its competitors. This positions the coaches as successful innovators and establishes the team as one of the top contenders.
We are happy to have had the opportunity to collaborate with the customer and contribute to developing a one-of-a-kind software product that supports the development of sports globally.