Engineering guidance, not a report of client results. Code fragments illustrate architecture and require testing before use.
Correction: unsupported pricing, physical thresholds and research attributions were removed. No client performance is claimed. Source review: .
What single-camera reconstruction aims to recover
Game state reconstruction connects athlete identities and locations to a top-view representation of the pitch. The SoccerNet-GSR paper introduces a dataset, an evaluation metric and an end-to-end baseline for this problem. It describes a challenging research task, not a universal solution for every camera. Read SoccerNet-GSR.
Start with the coaching question and representative footage. Off-screen players, occlusion and uncertain identity must remain explicit gaps rather than invented trajectories.
Calibration is part of the measurement
Image coordinates are not automatically physical distances. Estimate the camera-to-pitch mapping using known venue geometry and visible references. Check the mapping when a broadcast view changes. Use the actual venue dimensions rather than assuming one pitch size.
A bounding box's bottom centre can be an approximate ground-contact anchor, but occlusion, jumping and detector error can invalidate it. Reject unstable projections and expose uncertainty instead of presenting precise-looking coordinates.
Evaluate identity and location separately
Review detection misses, identity switches and pitch-localization errors on labelled footage. The SoccerNet-GSR work introduces GS-HOTA for evaluating reconstruction. That research metric does not establish whether a particular coaching report is correct.
Keep an analyst review path back to source frames. Record camera setup, dataset permissions, annotation rules, software versions and difficult conditions with each assessment. No installation price, minimum camera height or deployment accuracy is asserted here.
Keep tactical interpretation reviewable
Only derive tactical measures from observations that support them. Missing coverage may make a requested conclusion unavailable. Separate reconstructed positions from a coach's interpretation, and document how each measure is calculated.
For an illustrative pilot, deliver reviewable player tracks, a calibration report and a list of excluded measures. Compare against analyst labels before expanding. This is a proposed scope, not measured client evidence. Explore sports video analytics or read the TSX project context.
References
What the sources support
- SoccerNet-GSR introduces game-state reconstruction data, a baseline and GS-HOTA. CVPR Workshops 2024; arXiv 2404.11335. Research task; not a validated result for Alector Lab or any client.
- SoccerNet Game State Reconstruction: End-to-End Athlete Tracking and Identification on a Minimap — Somers et al., CVPR Workshops 2024
