Movement analysis from video works when capture is consistent and a person checks the result. Pose and object tracking can count repetitions, measure angles and time actions, but camera placement, lighting and a clear definition of what counts matter as much as the model. These four steps set those up first and measure agreement with a human before anyone relies on the numbers.
Step 1 — Define what to measure
Goal: Agree exactly what counts as a repetition, a good form or an event.
- Pick one movement or activity to start with.
- Write a plain definition of a repetition, a fault or an event, with examples.
- Decide what the coach or supervisor needs to see to trust a result.
- Agree how consent and footage storage are handled for the people filmed.
You finish with: A written definition, review needs and consent and storage rules.
Step 2 — Standardize capture
Goal: Make every recording usable.
- Fix camera position, distance and angle for the chosen movement.
- Check lighting and background so the person or object stays visible.
- Record a set of sessions and hand-count them as a reference.
- Note conditions where tracking struggles, such as occlusion or low light.
You finish with: A capture setup and a hand-counted reference set.
Step 3 — Track, measure and review
Goal: Produce results a person can check quickly.
- Track pose or objects frame by frame.
- Compute counts, angles and timings from the tracking.
- Mark low-confidence frames instead of hiding them.
- Show clips and scores side by side for the coach or supervisor to confirm.
You finish with: Reviewed counts and scores with clips as evidence.
Step 4 — Measure agreement and expand
Goal: Earn trust before using results more widely.
- Compare automated results with the hand-counted reference.
- Track review time per session against your starting point.
- Fix capture or definitions where agreement is low.
- Add the next movement only once agreement is acceptable to the reviewers.
You finish with: Measured agreement and a capture routine ready for the next movement.
Common pitfalls
- Starting analysis before camera placement and lighting are consistent.
- Treating scores as decisions instead of advice for a coach or supervisor.
- Hiding low-confidence frames instead of marking them.
- Filming people without agreed consent and storage rules.