Automatic Defensive Block Detection From Tracking Data: Two Dimensions That Classify Every Sequence

The first dimension is height — where the team's defensive structure sits on the pitch. A low block defends close to the own goal, a high block presses near the opposition box, a mid block sits between. The measurement is the collective spatial distribution of the team's outfield players during a defensive sequence: take the centroid or median y-coordinate of all ten outfield players, normalise against pitch length, and classify into bands.

The second dimension is length — the vertical distance between the team's top line (typically the forward line) and the team's base line (the defensive line). A compact block has a small vertical distance; a stretched block has a large one. The measurement separates coordinated defensive units from teams whose first line presses without support from the lines behind.

Combining the two dimensions produces four structural types, each corresponding to a distinct tactical situation. High compact block: a coordinated high press where the forwards trigger and the lines behind step up in sync. Mid compact block: stable defensive shape, opponent allowed into the middle third but no further. Mid long block: the forwards press but the lines behind don't step up — a stretched team, structurally vulnerable to balls played through the gap. Low compact block: organised deep defence, the classic 4-5-1 or 5-4-1 sitting in front of their own box.

A real match has many sequences from each type. The analytical value is in measuring time-spent and outcome-by-block-type — which block does this team concede most chances in, which block produces the most progressive opposition passes, which block successfully prevents shots. Those answers are usable as pre-match preparation against a specific opponent and as in-season trend data for a coach's own team.

A common analyst trap is averaging block characteristics across an entire match. A team that spends 30 minutes in a low compact block and 30 minutes in a high compact block has an average block height that's neither — and that average is misleading for tactical purposes. The team is operating in two distinct modes, and the right model separates them rather than blending them.

Sequence-level classification handles this directly. Each defensive sequence — the period from when the team loses possession to when they win it back or the opponent shoots — gets its own block classification. The team's match profile becomes a distribution of sequences across block types, not a single average. That distribution is what predicts performance against specific opponents, and it's what coaches actually find actionable.

Once the classification is automatic, the analytical questions become specific. In which block type does the team concede the most dangerous attacks? When does the team become too long — what triggers the front line to press without support? Which defensive structure is most effective at preventing progression into the final third? Which block type yields the most turnovers in the opposition half?

For a coach preparing for a specific opponent, the questions are inverted. What block does this opponent struggle to break down? Where does their build-up break against a compact press? Which side of the pitch yields more turnovers? The classification scales to every team in the league — a coach can compare their own block-type distribution against a league average, against a specific upcoming opponent, and against their own historical trends to spot drift.

• Height dimension — median y-coordinate of outfield players, normalised against pitch length.
• Length dimension — vertical distance between top and base of the defensive unit.
• Four types — high compact, mid compact, mid long, low compact.
• Sequence-level — classify each defensive phase, not the match average.
• Use cases — own-team drift detection, opponent-specific preparation, league-wide benchmarking.