VAR solutions for football

To accurately detect offside in football, numerous factors related to the geometry and topology of the football pitch must be considered. In this article, we will examine the main challenges faced by Virtual Offside Line (VOL) systems.

The perfect football pitch

An ideal football pitch should comply with FIFA recommendations regarding dimensions, markings, and surface parameters, and should be perfectly flat.

However, deviations are allowed, which may affect the performance of the VOL system.

Deviations in line markings

Some football teams may adjust their line markings at the home pitch. Minor changes in line markings become a tactical tool for the home team and may affect the correct rendering of offside lines by the VOL system.

In our practice, we have also encountered situations where pitch markings were asymmetrical relative to the half-way line. This may lead to errors when drawing the offside lines. In reality the deviations may be 10–20 cm in either direction – a significant difference for offside decisions.

Moreover, we've met cases where the main pitch marking rectangle (touch lines and goal lines) forms a rhombus or trapezoid instead of a rectangle. With a pitch length of 100 m, the error may amount to about 0.1%. However, when the VOL system is expected to detect offside with 1 cm accuracy, this error becomes significant and affects decision making.

Topological features of the football pitch

A football pitch is a complex structure and is not always perfectly flat. The pitch’s drainage system ensures water flows towards the edges to prevent stagnation, resulting in bulges in the center and slopes at the edges.

For the VOL system to operate correctly, three-dimensional pitch parameters must be considered. Any irregularities complicate the drawing of a straight line due to potential distortions, which may affect the final offside detection.

VOL system and cameras

When the VOL system uses cameras positioned close to the pitch edge, errors may occur due to incorrect height determination in the model. For instance, if a camera is mounted 10 m from the pitch edge at human height, and players are at the opposite end of the pitch (up to 50 m away), the offside line drawing error may reach 50–100 cm if the VOL system does not account for local height differences.

It is also necessary to consider image distortion when drawing offside lines. Shooting with cameras of different focal lengths leads to distortions where straight lines appear curved. In the VOL system, wide angle shots require offside lines to be drawn curved accounting for this distortion.

Improving VOL system performance

When working with a regular, non-ideal football pitch, we face numerous complexities that must be considered. To correctly draw the offside line, pitch and line marking measurements are required, along with constructing a topological model of the entire pitch, including all three-dimensional irregularities. Additionally, it’s worth considering to use cameras that reduce the likelihood of VOL system errors.

Creating a model of the pitch

Let's consider a regular football pitch with known topological coordinates for standard points (31–35 points) which are uploaded to the VOL system. If a disputed incident occurs near standard points within 1 m, the VOL system operates with high accuracy.

However, an incident may also occur at a considerable distance from known points. Given that depressions and bulges are distributed across the pitch, a more accurate three-dimensional topological pitch model must be created to ensure acceptable VOL system accuracy. This requires a grid with a larger number of points.

In our opinion, an acceptable grid for normal VOL system operation has a cell size of 2×2 m. The optimal grid has 1×1 m cells. This is a substantial pitch measurement task, but it enables accurate line drawing.

Of course, performing topology scans before each match is costly and impractical. For each pitch where matches may be held, a topological map must be created – a 3D object onto which the current actual pitch dimensions and line markings can be overlaid. This allows offside lines to be drawn as accurately as possible on a three-dimensional surface, as relative coordinates and necessary measurement heights will be known.

📄 Our guide to collecting field topology data for VOL technology

Optimizing cameras for offside detection

Another challenge is the need to capture potential offside incidents with a camera providing sufficient detail for the VOL system. When using broadcasting cameras, it may turn out that no suitable camera was available for offside detection in a specific situation.

This is addressed by installing fixed cameras that are always capable of capturing complex incidents. For example, the Premier League uses 30 cameras for precise fixed offside detection. They allow accurate determination of player positions in three-dimensional space.

Installing a large number of cameras requires substantial stadium equipment investment but guarantees high accuracy in offside detection and resolution of disputed incidents.

Our recommendations

For maximum efficiency within reasonable budgets, the following steps can be taken to improve VOL system accuracy:

• Perform topological pitch measurements twice: at the start of the season and in the mid-season. Also, perform pitch topology scans whenever pitch parameters change – e.g., when line markings are significantly altered or the turf is replaced. (📄 Instructions for collecting field data for using VOL technology).
• Whenever possible, conduct topological scanning of main pitch points (31–35 points) during line marking. Estimated task duration: 2 hours.
• Use special 16 m cameras positioned on each side of the pitch to capture the maximum number of disputed incidents.
• Ensure fixed cameras are synchronized with broadcasting cameras; otherwise, errors may reach up to 10 cm when capturing at 50i/50p.
• It is not enough to rely only on high-speed cameras (100-120 frames per second). Using four dedicated offside cameras and one overview camera can significantly reduce coordinate detection errors.