PoseTracker gives an application body tracking without building the computer vision for it. The API handles pose estimation in real time, counts exercise repetitions, measures joint angles for flexibility work, and compares a live pose against a reference position to give feedback. It reports a 40ms average response and about ten minutes to integrate.
Buying this rather than building it is almost always correct. Pose estimation is a solved research problem and a hard engineering one, and a fitness app whose value lies in coaching and programming has no reason to spend its runway reimplementing it.
Movelytics builds it on TensorFlow’s MoveNet, with an iOS app for testing before integration. Free gives 500 API calls a month for non-commercial use only, and commercial use starts at EUR 50 a month for 5,000 calls, which is a low ceiling for a consumer app. Accuracy depends on camera angle and lighting like every vision system, and a pose API is not a physiotherapist.
