What Is a Personal Movement Baseline?

A population reference is what most people picture when they hear "normal range": a distribution of many people's values, with a band across the middle that most of them fall inside. Compare one reading against it and you learn where the person sits among strangers. That is a real question, and for some purposes the right one.

It is not the question most movement monitoring is asking. A care team, a physiotherapist, a coach or the person themselves usually wants to know whether something has changed — and a population band cannot say. Two people can sit at the same point in it for opposite reasons: one has always been there, one has drifted there from somewhere else. The reading is identical; the story is not.

A personal baseline is not a value. It is a set of observations of the same measure, from the same person, made under conditions alike enough to compare — and it has two properties, both of which matter.

• A centre. Where this person's readings tend to fall. Often nowhere near the population centre, and that is not a problem; it is the point.
• A spread. How much the readings vary from one ordinary day to the next. Nobody walks identically twice, and the spread is what a change has to exceed before it counts as one.

The hero figure draws both. Bring the observations in one at a time and the first few form a cluster; the band drawn around them is the spread. The latest reading is then judged against that band — not against a number, and not against a population.

There is no universal number, and any story that offers one should be read carefully. What a baseline needs is enough observations to know its own spread: to have seen what an ordinary good day and an ordinary bad day look like for this person, so that a reading outside that range means something.

Three observations show a tendency. More show the variation around it. The figure draws its band only once three are in and keeps widening it as ordinary variation appears — because a band drawn from two very consistent days would flag the third, perfectly ordinary, day as a change.

A baseline describes a person at a period of their life, and people change. Someone recovering from an injury has a baseline that is supposed to move; freezing it at the first week would flag every improvement as an anomaly. Someone whose walk changes with the season, with new footwear or with a new routine has a baseline that has genuinely shifted, not a change worth a closer look.

Telling these apart is not something the measure can do alone. It takes the history — how fast the centre moved and whether it stayed — and it takes a record of the conditions each observation was made under.

The Engineering GaitAI story on camera angle made the point that two recordings from different mounts are two instruments. For a baseline this is decisive. If the observations that built the band came from a side-on camera and today's reading came from an oblique one, the reading is outside the band because the geometry changed — not because the person did.

• Every observation carries its capture. Angle, distance, lighting, occlusion, which sensor. A baseline is built only from observations whose capture is comparable, or corrected to be.
• A degraded capture does not enter the baseline. The fourth Foundation's figure showed an assessment with a degraded capture sitting in a sequence; it is shown, and it is not counted.
• A change in setup resets the comparison, not the person. When the instrument changes, the honest reading is "not comparable yet", and the baseline starts again alongside the old one.

A reading outside a personal baseline says one thing: this person's movement, measured this way, under these conditions, is not where it usually is. That is a great deal more than a population threshold can say — and it is still not a finding.

• It can say that something changed, in which measure, and roughly when.
• It can say whether the change persists across observations or was one unusual day.
• It cannot say why. Recovery, fatigue, footwear, a slope, a moved camera and a pose error all move a reading.
• It cannot say what to do. It is offered to a person who can look at the history, the capture and the person — as decision support.

That is the whole design. The population reference answers a question about everyone. The personal baseline answers a question about someone — and it can only do that if it is built carefully, allowed to move for the right reasons, and kept together with the conditions that made it.