Lone Striker Studio

The problem Most football is played and coached far from professional stadiums: on amateur pitches with faded or missing lines, dirt instead of grass, long shadows, and one coach with a phone. Today's tools for measuring players need fixed multi-camera installations, GPS vests or broadcast footage of well-marked pitches. Grassroots coaches get none of it, so they can't tell how far each player runs, where the team stands or how possession really went. That holds in matches, and even more in training. What it does The coach records a match or a training session with one phone and marks the pitch once, tapping a few points they recognize: the halfway line, the center circle, a corner of the box, the posts. From then on every player's position is in meters. - SAM 3.1 tracks every player and the ball from a short noun phrase built from the jersey color ("player in red jersey"). There is no training, no labeling and no dataset, so a new league, kit or pitch just works. - The coach's taps become a homography, which the worker follows frame by frame as the phone pans, tilts or zooms. SAM's white line masks re-anchor it wherever lines are still visible. - Muse Spark turns the measurements into short tactical findings (possession, distance and meters per minute, top speed, team width and length), each linked to the exact moment in the video. It never sees player names. - The app is honest about what it did not measure. Camera cuts, close-ups, frames without a calibrated pitch and long stoppages such as hydration breaks are flagged and left out instead of guessed. Results on real footage - On our own amateur pitches, including a dirt pitch where the lines are almost gone, 7–8 taps on one frame calibrate the whole pitch. - On a full 49-minute match with a moving camera, SAM 3.1 kept 187 player tracks with a median length of 99 s, where a local detector baseline produced 6,102 tracks of 4.3 s. The pitch was in meters in 86 % of the frames. - Possession matched hand-labelled ground truth in 95 % of the frames. - Cost: about US$0.36 per minute of video. How it's built - Android app (Kotlin, Jetpack Compose): records or imports the match, prepares a light 10 fps copy in chunks with a per-frame pitch mask, uploads in resumable parts and shows the analysis. - API: PHP and MySQL on shared hosting. - Python worker: calls SAM 3.1 and Muse Spark through the Meta Model API. - Privacy by design: video-use consent before the first upload, names that never leave the phone (important with youth players), and automatic deletion of the video after the analysis. Why it matters It brings measurement in meters, which used to need expensive gear, to any coach with a phone, on the pitches where most of the game is actually played.