Platform

HOW
IT WORKS

From raw match footage to predictive game intelligence. Three phases, one unified system.

AWS EC2 G5.16xlarge
YOLOv9 Detection
ByteTrack Persistence
Jersey OCR
Homography Calibration
No Proprietary Hardware
01Live

Detection & Tracking

Every player. Every second. Every centimetre.

YOLOv9ByteTrackJersey OCRRe-IDHomographyAWS G5.16xlarge
01
Multi-camera ingestion

Four angles processed in parallel: broadcast wide, dedicated wide lens, behind the goals left, behind the goals right.

02
Player detection

YOLOv9 identifies every player on-field across all camera feeds with high-precision bounding boxes.

03
Persistent tracking

ByteTrack maintains unique player IDs through occlusions, player crossings, and close groupings — IDs never drift.

04
Player identification

Jersey-number OCR and re-identification logic confirms each player's identity and links them to the squad database.

05
Camera calibration

Homography maps each camera's pixel space to true field dimensions in metres, accounting for lens distortion and camera angle.

06
2D field coordinates

Every player and the ball is placed on a unified top-down field map with X/Y coordinates updated per frame.

02In Build

Action Recognition & Pattern Analysis

What players do, and what it means.

Action MLEvent SyncPattern MiningARIMASequence Models
01
Event feed alignment

Structured JSON event feeds (kicks, handballs, tackles, marks) are aligned frame-by-frame with tracking data using timestamps and player IDs.

02
Action recognition

Computer vision models trained to classify player actions directly from video — reducing dependency on manual coding feeds over time.

03
Contextual enrichment

Every action is tagged with context: field position, game state, pressure rating, nearest opponents, available options.

04
Pattern mining

Unsupervised learning surfaces recurring patterns in player movement, team structure, and opposition tendencies.

05
Predictive modelling

ARIMA and sequence models begin forecasting likely next events based on current game state and historical patterns.

03Roadmap

Predictive Intelligence Layer

The intelligence coaches actually use.

NLP InterfacePredictive AIVideo OverlaysDashboardsScenario Modelling
01
Natural language interface

Coaches and analysts query the system in plain English: "How does Geelong structure their forward press in wet conditions?"

02
Pre-game briefings

AI-generated opposition scouting reports drawing on the full historical match library — ready the night before.

03
Video overlays

Player tracking paths, spatial zones, and tactical annotations burned directly into match footage.

04
Interactive dashboards

Role-specific views for head coaches, analysts, list managers, and fitness staff.

05
Scenario modelling

Simulate how different player combinations, game plans, or opponent setups affect predicted outcomes.

Seen enough?

Book a demo and we'll walk through what iCoach sees in your own footage.