What Movement Capture Records

A recording is not yet a risk claim

Movement capture turns a physical task into information that can be stored and compared. Cameras record changes in an image. Inertial measurement units record acceleration and angular velocity. Force platforms record forces and moments at the surface. These are different observations, not interchangeable versions of the same truth. A system may then estimate joint positions, contact events or other features from those observations.

The distinction matters because each processing step adds assumptions. A camera cannot see a joint centre directly. A wearable does not automatically know its orientation relative to a body segment. A force platform cannot explain why an athlete chose a particular landing strategy. When a report jumps straight to a coloured risk label, ask to see the measurement and the intermediate calculations underneath it.

Choose a method for the question

Marker-based optical systems use the positions of reflective markers to reconstruct movement. Their output depends on marker placement, the biomechanical model and movement of skin relative to underlying bones. Markerless systems estimate landmarks from video and can reduce setup demands, but clothing, lighting, camera angle and occlusion still affect tracking. Performance demonstrated in one task does not establish equivalent performance in every sport.

Wearables can capture repeated activity outside a laboratory, which is useful when real exposure matters. Placement, attachment, calibration, magnetic disturbance and drift can complicate interpretation. A measure called “load” may be a proprietary summary rather than a direct measure of tissue loading. Ask what its units mean and whether its validity has been studied for the intended activity.

The task is part of the measurement

A planned jump in a quiet room differs from an unexpected change of direction during a match. Instructions, approach speed, surface, footwear and familiarisation can all change the strategy an athlete uses. This does not make controlled testing useless. It means the test answers a narrower question than “How does this athlete move in every relevant situation?”

Write down a repeatable protocol before collecting data. Record warm-up, equipment settings, trial selection and the reason any attempt was excluded. Avoid choosing only the neatest-looking trial after seeing the result. If a system cannot analyse a movement, that missingness is information: failures may be more common in difficult tasks or particular groups, which can bias the remaining sample.

Reliability and validity answer different questions

Reliability concerns consistency when measurement conditions are sufficiently similar. Validity concerns whether the interpretation of the measurement is supported for its intended use. A device can produce highly consistent values that are systematically offset from a reference method. Conversely, agreement on average can conceal errors large enough to matter when judging one person.

Look beyond a single correlation coefficient. Correlation can remain high when two methods disagree in absolute terms. Reports of agreement, typical error, test–retest behaviour and uncertainty around individual measurements are often more useful. The International Society of Biomechanics offers recommendations on describing movement and joint coordinate systems; consistent definitions make comparisons easier, but do not validate a particular device automatically.

Keep the original question in view

Tools such as OpenSim can support musculoskeletal modelling, while OpenCap offers an approach to estimating movement from video. Both illustrate how recordings can feed more elaborate estimates. Their outputs still require careful attention to model assumptions and the population and movements used for evaluation. An estimated internal load is not a directly observed injury mechanism.

Before moving to a model, ask whether the chosen feature is sufficiently repeatable to detect the difference that matters. If expected change is similar to measurement noise, a categorical label may create false precision. Preserve the raw data where appropriate, document processing versions and report uncertainty. Only then is it sensible to ask whether the feature has a defensible relationship with a future outcome.