Measurement quality

An angle is only as good as the image it comes from. KinesiaMotion gives you three tools to know when to trust a measurement.

1. Placement: `SetupIssue`

On every frame, SetupCheck looks at how the patient stands in front of the camera, and the session publishes the problems in snapshot.setupIssues (and in .setupChanged when they change, after ~0.5 s of stability).

Case	What happens	What to tell them
`turnSideways`	Side-view exercise, but they face the camera: the angle is foreshortened	“Turn sideways”
`faceCamera`	Front-view exercise, but they’re turned sideways	“Face the camera”
`moveCloser`	The measured joints look tiny: points jitter more	“Come closer”
`stepBack`	A joint is at the edge of the image and will leave it when moving	“Step back a little”

Anything that can’t be judged (a joint out of view, no imageSize) is left out rather than guessed.

if session.snapshot.isVisible, let issue = session.snapshot.setupIssues.first {
    Banner(text(for: issue))
}

Tuning the thresholds

var check = SetupCheck()
check.maxSideRatio = 0.45   // shoulder width ÷ torso above which they "face the camera"
check.minFrontRatio = 0.35  // below which they're "sideways"
check.minSpan = 0.15        // minimum size of the joints ÷ the image's shorter side
check.edgeMargin = 0.03     // minimum margin to the edges
session.setupCheck = check

session.framesToConfirmSetup = 12       // debounce (~0.5 s at 24 FPS)
session.framesToConfirmVisibility = 6   // debounce (~0.25 s)

2. Detector confidence

snapshot.confidence is the lowest confidence among the primary measure’s joints on the last frame, 0…1. Each Rep stores the average over the rep.

if let c = session.snapshot.lastRep?.confidence, c < 0.5 {
    Text("Unreliable measurement: improve the light or framing")
}

It’s nil when the pose source gives no confidence.

3. Per-rep metrics

Beyond the peak, each Rep carries data to assess movement control:

for rep in session.reps {
    print(rep.peak, rep.riseDuration, rep.returnDuration, rep.peakVelocity)
}
session.reps.peakSpread   // standard deviation of the peaks: low = consistent

A return much faster than the rise (returnDuration ≪ riseDuration) usually means the patient “drops” the limb instead of controlling it.