Introduction

KinesiaMotion is the SDK the Kinesia app uses to measure a patient’s movement with the iPhone camera. It takes body points frame by frame and returns what a rehab app needs: the current angle, the reps done, the ones done with poor form, and whether the patient is well placed in front of the camera.

Two modules

Module	What it contains	Depends on
`KinesiaMotion`	The engine: poses, measures, exercises, rep counting, session. Pure Swift, no camera, no UI.	Foundation, CoreGraphics, Observation
`KinesiaMotionVision`	`VisionPoseProvider`: live poses from the camera with Vision’s body detector.	`KinesiaMotion`, AVFoundation, Vision

The split is deliberate: the engine can be tested on macOS with recorded poses, and the pose source can be swapped (another detector, a file, a server) without touching anything else.

What it does

Measures in degrees with four kinds of measure: inner angle, flexion, angle from vertical and tilt.
Counts reps with hysteresis thresholds, and times holds (keeping a position for N seconds).
Checks form: each exercise has form rules; it cues live and rejects reps that break them.
Checks placement: detects whether the patient should turn, move closer or step back.
Measures quality: detector confidence per rep, peak velocity, rise and return times.
Baseline measurement: the median of the first N clean reps.

What it doesn’t do

It doesn’t draw UI. It gives you an observable MotionSnapshot and an event stream; the screen, voice and haptics are yours.
It doesn’t store data. You decide where to persist the reps (session.reps).
It doesn’t translate. Corrections (FormFeedback, SetupIssue) are enum cases; your app turns them into text and speech in the patient’s language.

Requirements

iOS 17 or later (macOS 14 for the engine and tests).
Swift 6 with strict concurrency: every public type is Sendable.
KinesiaMotionVision only works on iOS (it needs a camera).