Sports Analysis with Computer Vision: Pressing Intensity

I recently felt a bit nostalgic and decided to rewatch the incredible 2022 World Cup final between Argentina and France.

What a match!

As the game ended, certain moments stood out – incredible saves, brilliant goals, but also relentless defensive work.

Afterwards, I fell down the rabbit hole of post-match analysis. Pundits kept showing stats and graphs.

Honestly, I felt a bit overwhelmed by the sheer volume of data – expected goals, heat maps, pass completion percentages... it was a lot.

But one term kept popping up, something the analysts stressed as crucial, especially when talking about strong defenses: pressing.

They talked about high presses, counter-presses, pressing traps. It sounded important, but how do you actually measure something like that?

It seemed less tangible than a simple count of goals or passes. How can you put a number on a team’s collective effort to make it difficult to score for the opponent?

Curious, I started reading a few articles and research papers, trying to make sense of it. While on it I found this interesting paper proposing a new way to look at this.

First Off, What Exactly is "Pressing" in Sports?

Before we get into measuring it, let's clarify what pressing means. Think of it as organized, intense defending with a clear goal: win the ball back quickly.

It’s not just randomly chasing the player with the ball. It's a coordinated team effort. Players work together to close down space, cut off passing options, and force the player with the ball into making a mistake or a turnover.

Famous analysis blog, Spielverlagerung describes it as creating "tension with the intention of getting the ball back." Every player's movement contributes to this pressure somewhere on the field.

When a team is pressing effectively, it feels like the field is shrinking for the team in possession. They have less time, less space, and fewer easy options.

Pressing requires fitness, tactical understanding, communication, and aggression. Successful pressing can lead directly to scoring chances by winning the ball close to the opponent's goal.

However, if done poorly, it can leave gaps in your own defense for the opponent to exploit. It’s a high-risk, high-reward strategy, fundamental to many modern teams' tactics in sports like soccer, basketball, and hockey.

The Old Ways of Measuring Pressure

Coaches and analysts have always known pressing is important, so naturally, people tried to measure it. But the early methods had significant limitations:

1. Passes Allowed Per Defensive Action (PPDA): 

This is one of the most well-known older metrics.

It looks at how many passes the attacking team makes in their own half (or a specific zone) divided by the number of defensive actions (like tackles, interceptions, fouls, challenges) the defending team makes in that same zone.

A lower PPDA number suggests more intense pressing (fewer passes allowed before a defensive action occurs).

The Problem: PPDA uses event data (things that happen to the ball). It tells you nothing about the positioning or movement of players off the ball.

A defender might be perfectly positioned to cut off a pass but doesn't make a tackle – PPDA misses this entirely. It also doesn't account for player speed or direction.

2. Rule-Based Pressure Events (e.g., StatsBomb) 

Some data providers created simpler rules. For example, StatsBomb counts a "pressure event" whenever a defender gets within 4 or 5 yards of the player carrying the ball.

The Problem: This is better than previous, but still very basic. It's just about proximity. Is the defender actually moving towards the ball carrier with intent, or just jogging nearby?

Is the defender sprinting or standing still? Are they positioned effectively between the ball carrier and the goal?

This simple radius rule misses a lot of crucial context, especially player speed and direction.

3. Other Approaches

Researchers tried other things too, like defining expert rules based on proximity and movement towards the ball, or trying to identify pressing strategies from video analysis.

Some even created formulas considering relative position to the ball and the goal.

However, many of these methods lacked a key ingredient: fully incorporating the dynamic nature of player movement, especially speed and direction, using the detailed positional data now available.

They often felt incomplete or not intuitive enough for coaches to use easily.

Essentially, the old methods gave a partial picture. They couldn't fully capture the dynamic pressure created by all players' movements across the field at every single moment.

"Pressing Intensity" Using Tracking Data

This is handled in the paper "Pressing Intensity". Its main goal is to create a more precise, dynamic, and intuitive measure of pressing by fully utilizing positional tracking data.

Modern soccer leagues track the exact (x, y) coordinates of every player (and the ball) multiple times per second. This paper uses that rich data.

The core idea borrows concepts from existing models used to calculate "Pitch Control" (which estimates which team controls which areas of the field), but adapts them specifically for measuring pressure.

Instead of asking "who controls this space?", it asks "how quickly can each defender intercept an attacker or the ball?"

The key improvements this paper introduces are:

• Uses Positional Data for All Players: It considers the position, speed, and direction of every defender relative to every attacker (and the ball) at each frame of the tracking data.
• Incorporates Speed and Direction: It's not just about distance. How fast a defender is moving and which way they are heading is crucial for determining the actual pressure they apply.
• Dynamic Calculation: Pressure isn't static. It changes constantly. This method calculates pressing intensity at the frame level (e.g., 25 times per second), giving a continuous, dynamic view.
• Intuitive Output: The final metric is presented as a probability – the chance a defender can intercept an attacker or the ball within a short time window (e.g., 1.5 seconds).
  
  This is easier for coaches and analysts to understand and use than abstract indices.

By doing this, the paper aims to provide a much richer, more accurate picture of how pressure ebbs and flows during a match, influenced by every player's actions.

How It Works?

Let's peek under the hood without getting lost in complex math. The process involves a few key steps:

1. Calculate "Time to Intercept" (Ti,j) 

This is the heart of the model. For each defending player (i) and each attacking player or the ball (j), the model calculates the estimated time it would take for defender i to reach target j.

This calculation isn't just distance divided by speed. It considers:

• Current positions of both players (ři(t) and řj(t)).
• Current velocities (speed and direction) of both players (vi(t) and vj(t)).
• A reaction time (Tr): A small delay before the defender reacts and accelerates.
• Maximum running speed (vmax): How fast the defender can sprint.
• A penalty (Tß): An extra time penalty added if the defender is initially facing or moving away from the target. This makes intuitive sense – it takes longer to intercept if you first have to turn around.

2. Convert Time to Probability (Pi,j) 

Knowing it takes, say, 1.2 seconds for defender i to intercept attacker j isn't super intuitive. The paper converts this time into a probability using the logistic function.

This function smoothly maps time onto a probability scale (0 to 1).

• The function is set up so that very short interception times result in a high probability (e.g., close to 1), while longer times result in lower probabilities.
• The paper uses specific parameters (T = 1.5 seconds, σ = 0.45). This means it calculates the probability that defender i can reach target j within 1.5 seconds.
  
  If Ti,j is much less than 1.5s, the probability is high. If Ti,j is much more than 1.5s, the probability is low.

3. Calculate Total Pressure on an Object (Pj)

An attacker might feel pressure from multiple defenders at once. To get the total pressure on attacker j, the model combines the individual probabilities (Pi,j) from all defenders (i).

It calculates this as: 1 - (the chance that *none* of the defenders can intercept j).

• Mathematically: Pj = 1 - product(1 - Pi,j) for all defenders i.
• This means if even one defender has a high probability (Pi,j is close to 1), the total pressure Pj will be high. If all defenders have low probabilities, the total pressure will be low.
• (Note: This calculation assumes the probabilities from different defenders are independent, which is a simplification but works reasonably well).

4. Refinement

Active Pressing Filter: Raw calculations might show pressure even from a defender who is just jogging slowly nearby.

To focus on intentional pressure, the paper introduces an "Active Pressing" speed threshold (e.g., 2 meters per second).

If a defender's current speed is below this threshold, their calculated probability of intercepting is set to zero.

This filters out "passive pressure" and focuses on defenders actively moving to close down opponents.

The output can then be visualized, often as a grid showing the pressure each defender exerts on each attacker at a specific moment

Applying Pressing Intensity In Other Sports (e.g., Basketball)

The core ideas behind this Pressing Intensity metric aren't limited to soccer. They could be adapted for other team sports where applying pressure is key, like basketball.

Imagine trying to measure defensive pressure in basketball using tracking data:

• Targets: The "attacking objects" would be the player with the ball (ball handler) and potential receivers standing open or cutting to the basket. Passing lanes themselves could also be targets.
• Defenders: You will calculate the Time to Intercept for the on-ball defender to reach the ball handler, and for off-ball defenders to reach potential receivers or intercept passing lanes.
• Time to Intercept Calculation: You will need basketball-specific parameters for reaction time, maximum sprint speed on court, and maybe acceleration. The penalty for facing away would still apply.
• Probability: You will convert these times to probabilities, perhaps using a shorter time window (T) than in soccer, reflecting the faster pace and smaller court (maybe T=0.75 or 1.0 seconds?).
• Total Pressure: You will calculate the total pressure on the ball handler, the pressure on each potential receiver, and the pressure on key passing lanes.
• Use Cases: This could help quantify:
  • How effectively a defender is "locking up" the ball handler.
  • How well help defenders are positioned to deny passes or contest shots.
  • The effectiveness of defensive schemes like traps or zones in generating pressure.
  • Which offensive players are good at relieving pressure or making plays despite it.

Adapting the parameters and potentially the exact definition of "interception" (reaching the player vs. reaching the pass lane) would allow this framework to provide valuable insights into defensive intensity in basketball and potentially other sports like hockey or American football (measuring QB pressure).

Where the Model Can Improve?

No model is perfect, and the authors acknowledge areas for improvement:

Pitch Boundaries

The current model calculates Time to Intercept as if the field is infinitely large. In reality, sidelines limit movement.

A player being forced towards the sideline might feel intense pressure, but if the calculated time to intercept on the pitch is greater than 1.5 seconds (because the defender has to run a long way around), the model might register zero pressure.

The paper suggests a potential fix: adding a virtual "pressure" exerted by the sidelines themselves.

Independence Assumption

Calculating total pressure assumes the probability of each defender intercepting is independent of the others.

In reality, defenders coordinate. If two defenders pressure one attacker, their actions might influence each other.

This assumption simplifies the math but isn't perfectly realistic.

Defining "Targets"

The paper mainly focuses on pressure on opposing players.

Redefining the target as the closest point on a passing lane allows measuring how well defenders close down passing options, which is a different but related aspect of pressing. This requires more complex calculations.

Measuring Efficiency (Smart Pressing)

The paper proposes combining Pressing Intensity with Metabolic Power data (which estimates energy expenditure). This could help measure "smart pressing" – achieving high pressure without wasting too much energy.

Future work could focus on addressing the pitch boundary issue, refining the total pressure calculation, extending the model robustly to pass lanes, and deeply integrating energy expenditure to evaluate pressing efficiency.

Wait.. But How Do You Get The Tracking Data?

To measure pressing intensity in football, we need accurate tracking data for every player and the ball throughout the game.

The system calculates intensity using physics-based models like time-to-intercept. But before any of that works, we need the raw tracking data, which comes from computer vision (CV) systems analyzing match footage.

CV models learn by example. If we want a system to track players and the ball during a fast-paced match, we need to first label it what those things look like.

That means manually labeling thousands of video frames marking each player and the ball.

Without this step, the model wouldn’t know how to find players(Object Detection or Segmentation), follow them(Object Tracking), or understand the game(Event Tagging).

Types of Annotation Used

To track players and calculate pressing intensity, we need a few key types of annotation:

• Object Detection
  Annotators draw boxes around each player and the ball. Every box gets labeled (like "Team A Player" or "Ball"). This helps the model learn what to look for in new footage.
• Player Tracking / Re-Identification
  Once players are detected, annotators link the same player across different frames using a unique ID. This allows the model to follow individual player movements, which is crucial for measuring speed, direction, and predicting interceptions.
• (Optional) Semantic Segmentation
  In some cases, annotators label every pixel that belongs to players, the ball, or the pitch. This gives the model a deeper understanding of the scene, although it's not always required for pressing intensity.

Doing this all manually takes an enormous amount of time and effort. That’s where Labellerr comes in.

Our platform helps automate video annotation with smart tools and AI-powered workflows.

You can label thousands of frames faster, track players more accurately, and build better models, without burning out your team.

Book a demo now!

Conclusion

The "Pressing Intensity" metric captures the fast-paced, high-pressure moments in games like the World Cup final.

Instead of just counting events or measuring distance, it uses tracking data, like player positions, speed, and direction, to show pressure in a more detailed, frame-by-frame way.

Coaches and analysts can now see who is pressing, where it’s happening, and how effective it is. It also helps break down pressing tactics and even measure how physically demanding different actions are.

While it still needs improvement, like better handling of pitch edges, it’s a big step forward. It turns complex data into clear insights and helps us see the invisible tactics that shape the game.

FAQs

What is pressing intensity in sports?

Pressing intensity measures how actively a team pressures opponents to regain control of the ball, especially in football or basketball.

How does computer vision help measure pressing intensity?

Computer vision tracks players’ positions and movements in real-time, helping teams analyze how often and how fast players press the opponent.

Why is tracking pressing intensity useful?

It helps coaches understand team effort, improve defensive tactics, and make data-driven decisions about training and performance.