DFL Introduces AI-Driven Automated Event Detection Built on 3D Player Tracking.

Roger Hampel

Image: DFL Deutsche Fußball Liga

DFL Deutsche Fußball Liga has advanced its data automation strategy with the implementation of AI-driven automated event detection (AED), built on its new 3D tracking system introduced at the start of the 2025/26 season.

The initiative represents a structural shift in how match events are captured, processed and distributed across Bundesliga data products.

From Manual Logging to Machine Learning DFL

Since the beginning of the 2025/26 campaign, the DFL has deployed 3D player tracking across all matches. The system captures movement data through multiple camera feeds, translating each player’s body into 21 defined skeletal tracking points — including head, shoulders, hips, joints and feet.

This data enables full virtualisation of player movement and creates the foundation for automating event recognition.

Historically, match events such as passes, duels, throw-ins and shots were manually logged live by multiple operators. While accurate, the process is resource-intensive and subject to human interpretation variance. The new AI model, developed by Sportec Solutions AG with infrastructure support from Amazon Web Services, uses machine learning to classify match events in real time.

Confidence Scores and Event Classification

The model was trained using hundreds of annotated video clips for each event type. During live matches, the system assigns a probability value — referred to as a “confidence score” — to each detected event, indicating the likelihood that a specific sequence corresponds to a defined category.

Currently, the system can reliably detect ball-centric events including:

• passes and receptions

• shots

• set pieces such as throw-ins, corners, kick-offs, goal kicks, free kicks and penalties

Following a proof-of-concept phase in early summer 2025, the technology entered live testing at the beginning of the 2025/26 season. According to DFL representatives, extensive validation is ongoing to ensure product-level quality standards.

Dr. Hendrik Weber, Head of Sports Technology & Innovation at the DFL, added:

“We continuously strive to further develop data collection in a forward-looking way. At the same time, the DFL maintains very high quality standards for its products and services. These extensive tests are therefore essential. Through software-supported correlation and processing of tracking and event data, we will be able to generate entirely new data products for various stakeholders, including clubs and fans. Semi-automated event data collection is a very strong first use case within our ambitious 3D programme.”

Expanding the Scope Beyond Ball Events

Development continues to expand the system’s capability toward more complex scenarios, including dribbles, duels and off-ball movements such as scanning behaviour prior to ball reception.

Holzer described the process as comparable to human learning: while approximately 80% of defined events can be detected with relative simplicity, the remaining edge cases require iterative machine learning refinement.

The objective is not to eliminate human operators but to reduce manual workload and increase data depth. Operators will continue to validate AI outputs and manage quality assurance.

Commercial and Product Implications

Automated event detection significantly increases the volume and granularity of data that can be generated during live matches. According to Holzer, the system enables a scale of capture comparable to assigning an individual operator to each player.

For the DFL, the strategic upside extends beyond operational efficiency. Enhanced tracking and event correlation can support:

• advanced club performance analytics

• enriched broadcast graphics

• fan-facing data products

• new commercial data packages

The automated event detection system is currently running in parallel with manual processes. Full operational deployment is expected following validation completion.

The initiative forms part of the DFL’s broader 3D data programme, launched in the previous season.