On-Device Pose Detection
The counting pipeline processes camera frames with Apple Vision on the phone. This supports the app’s no-account workflow without sending workout video to a remote inference service.
iOS · ON-DEVICE VISION
Workout Counter iOS
Count bicep-curl repetitions without taking attention away from the exercise.
The Problem
Keeping track of repetitions can distract from the exercise, and checking a counter repeatedly adds another interruption. Automating the count with a camera introduces a second problem: joint detections fluctuate, wrists can leave the frame, and a partial movement should not count as a complete repetition.
The Approach
The app uses AVFoundation to capture camera frames and Apple Vision to estimate body joints on the device. A bicep-curl counter follows arm extension and flexion, updates the repetition total, and signals progress toward a chosen goal. A live skeletal overlay shows what the camera is tracking.
Count Complete Movements
The counter combines joint angles, smoothing, confidence checks, and extension/flexion states. Motion history and missing-frame tolerance help handle brief tracking gaps instead of treating every change in angle as a repetition.
Goals and Feedback
Repetition goals and audio and haptic cues signal progress without requiring users to watch the display. A live skeletal overlay helps users see whether their pose is being tracked.
Current Scope
Bicep-curl counting is available now. An ExerciseCounter interface separates movement logic from camera capture and the UI, providing a foundation for additional exercise counters. Accurate counting depends on camera placement and visible joints. The app tracks repetitions; it does not assess exercise form.