Measurement and Metric Glossary
Which metric comes from which test, how it is measured, and what we do not measure.
The five categories below produce results from real training footage today. Each has one primary KPI — the single number a coach turns into a decision — plus supporting data that puts it in context. The measurements are in field use in the football and tennis pilots.
Speed
Test: Repeated sprint test (RST)
- Continuous speed-time curve
- Maximum speed (km/h)
- Trend against previous tests
How it is measuredPlayer detection and a pose model locate the foot contact point, and a homography to the pitch plane scales distance. Because the run is straight and one-directional, this is the category with the lowest error from camera angle.
Agility
Test: Change-of-direction test (COD, 5-10-5 pro-agility)
- Total test time
- Speed-drop curve at the turn
- Right versus left turn comparison
How it is measuredThe same pipeline as sprint, plus turn-point detection. CODD is preferred because total time also contains the athlete’s straight-line speed: two fast athletes can post the same total time with very different change-of-direction ability.
Endurance
Test: Maximal aerobic speed (MAS) — 30-15 IFT
- Total distance covered
- Repetitions completed
- Point of fatigue
- Season-long trend
How it is measuredThe same tracking pipeline over a fixed area and shuttle geometry. It was chosen because it is the most widely used intermittent aerobic test in team sport, and its output converts directly into training-intensity prescription.
Strength
Test: Resistance training (RT) — velocity-based training (VBT)
- Load-velocity profile
- Velocity loss within a set (%) — a fatigue indicator
- Estimated 1RM
How it is measuredBarbell tracking — object tracking rather than athlete pose, so a separate component of the pipeline. THE OUTPUT IS VELOCITY, NOT FORCE: this is the only camera-compatible approach in the category.
Power and explosiveness
Test: Plyometric jump
- Flight time (ms)
- Variation between repetitions
- Left-right asymmetry (single-leg variant)
How it is measuredDetection of the take-off and landing frames; height is computed from flight time. It is the only category that needs no homography, which makes its capture setup the simplest.
FIFA 11+ (separate track)
Test: Warm-up and injury-risk-reduction programme
- Adherence to duration and repetitions
- Compliance trend per player and per squad
How it is measuredNOT YET IN THE FIELD — unlike the five categories above, this track is awaiting development. It sits outside the main measurement model because the programme’s real outcome — a fall in injury incidence — belongs to 10-12 weeks of cumulative practice, not to the biomechanical precision of a single exercise. What is measured here is therefore PROTOCOL COMPLIANCE, not biomechanical accuracy: was the exercise performed, and did duration and repetitions meet the target.
What we do not measure, and why
You can judge the honesty of a measurement list by whether it states what it leaves out. The following are deliberately out of scope because a single camera cannot measure them reliably.
Force and power in newtons
These cannot be measured directly from a single camera; measuring them requires a force plate. Test types whose output is force — Nordic hamstring, isometric tests, static single-leg variants — were therefore left out of scope. This is a deliberate boundary rather than a gap: it is entirely possible to PRODUCE a number in newtons from a biomechanical model and the athlete's body mass, but the number produced is not a measurement. Listing "peak force" from a phone camera is technically feasible; it does harm the moment a club trusts that number to set training load. What we measure in strength training is velocity: barbell velocity, velocity loss within a set as a percentage, and the load-velocity profile. Published work reports agreement of r = 0.940 between camera-based barbell velocity and reference systems. The 1RM estimate is derived from that profile and requires the load lifted as an input — so wherever it is an estimate, it is labelled as one.
Heart-rate-based metrics
A camera cannot measure these reliably. Heart rate is an internal physiological variable; what a camera sees is movement itself, so protocols whose primary output is heart rate, such as HIIT, are not in scope. The same reasoning applies to recovery indices and heart-rate variability. We read endurance from movement instead: aerobic speed reached in the 30-15 IFT test (VIFT), total distance covered, repetitions completed and the point of fatigue. These do not substitute for heart rate, but they convert directly into training-intensity prescription, which is what team sport practice actually asks for. Where a club already collects heart-rate data, the two sources complement each other: we measure the external load, the chest strap measures the internal response.
Crowded and contact scenes
Small-sided games involve many players and heavy occlusion, while Olympic lifts and complex or PAP sets involve fast movement very close to the body. Pose estimation loses reliability in both: when one player's joint sits behind another's, the model keeps producing an estimate, but the error in that estimate will not carry a measurement. Cone and ladder drills, reactive drills and tests with multiple turns are outside the MVP for the same reason — every additional turn accumulates error, and the literature reports that accuracy in change of direction is lowest at the moment of the turn. That is why the MVP is limited to the single-turn 5-10-5: low risk, and one of the two most frequently used agility tests in the literature.
Literature on the method
The following are independent published studies on markerless measurement — they are not TechForSport’s own validation results. They are here to show where the method is strong and where it is weak.
- Needham et al. 2021 (Sensors) — Markerless speed measurement with a Kalman filter shows an error of 0.041 ± 0.257 m/s against a marker-based system.
- Van Hooren et al. 2023 (Scand. J. Med. Sci. Sports) — Markerless accuracy decreases as speed increases.
- Helwig et al. 2025 (Scientific Reports) — In change of direction, accuracy is lowest at the moment of the turn.
- Balsalobre-Fernández et al. 2020 (J. Sports Sciences) — Barbell velocity measurement for velocity-based training shows r = 0.940 agreement.
Want to try it on footage from your own field?