The Technology Behind Pitchside AI

Pitchside is powered by a custom machine learning model trained specifically on small-sided football footage. The model was built this way because grassroots football does not look like professional 11-a-side football.

Small-sided games have tighter spaces, shorter passing patterns, faster transitions, different camera angles and more crowded visual cues. A model trained only on professional broadcast footage would miss too much context. Pitchside is designed around the footage real amateur teams can actually capture.

Pitchside started with 5-a-side because it is one of the hardest and most useful grassroots formats to understand. The game is fast, compact and full of repeated actions: goals, saves, passes, tackles, assists and quick turnovers.

The model has been trained using local small-sided game footage, with many hours spent annotating clips frame by frame. This helps Pitchside learn the visual patterns of real grassroots football instead of relying on assumptions from elite-level match footage.

The same approach also supports 6-a-side and 7-a-side football, where the pitch is still smaller than full 11-a-side and the game remains player-moment heavy.

Pitchside uses computer vision to read match footage and identify football actions from video. The goal is to understand what happened in the game, not just store a recording.

The system is being built to detect key football events, assign those events to teams and players, and turn long recordings into useful match output. That includes statistics, highlights and leaderboards for players who want proof of performance.

Pitchside is testing supported match output from suitable phone-recorded small-sided football footage.

Supported events currently being tested include goals, assists, saves, passes and tackles. Output can vary by footage quality, camera position, lighting, kit similarity and obstruction.

The goal is to turn a useful recording into stats, highlights and player records that can be reviewed after the match.

A stable, elevated landscape recording gives Pitchside a clearer view of the important playing area. Two phones can help cover opposing halves when one phone cannot keep the match in frame.

A one-phone setup can also work on smaller pitches if a short test recording shows both goals and the main playing area clearly. Avoid unnecessary panning, unstable mounts and positions where people can walk in front of the lens.

Footage quality, weather, lighting, camera height and venue layout all affect the final output. For practical setup steps, use the football recording setup guide.

Pitchside is still improving. Accuracy is not perfect yet, and detected events or player assignments may need review.

The current upload process can also take longer than ideal. At the moment, footage may take up to 45 minutes to upload and process because the system waits until the game has finished recording before uploading.

A future improvement is to stream the upload during the recording period, which should reduce waiting time after the match.

The beta is focused on improving event recognition, player assignment, upload handling and the quality checks that tell users when footage is suitable.

The long-term goal is to make football video analysis easier for grassroots players: record the game, upload the footage, receive stats, generate highlights and compete on player leaderboards without needing GPS vests or expensive camera hardware.

Most amateur players do not have analysts, camera operators or expensive football tracking systems. They have phones, teammates and matches worth remembering. Pitchside is being built for that reality.

The product focus is simple: help grassroots players turn suitable phone-recorded match footage into useful stats, highlights and player records.