Live Pose Detection

Built on MediaPipe PoseLandmarker and OpenCV, it tracks 33 full-body keypoints from a webcam feed.

• Exercise reps: squat and push-up counting with configurable angle thresholds
• Form scoring: green/yellow/red skeleton overlay from joint-angle quality
• Posture analysis: forward head, uneven shoulders, rounded back, neck tilt
• Live angle timeline: a chart of joint angles over the last 300 frames
• Reference skeleton: semi-transparent guide overlay to compare poses
• Video input: run recorded .mp4 / .avi / .mov files instead of the camera
• Session export: annotated MP4 plus CSV/JSON keypoint data per session
• Multi-person: up to 3 people tracked with proximity-assigned IDs

It's a PySide6 app with a live feed and tabbed panels, built on OpenCV + NumPy/SciPy + Matplotlib, packaged with uv, and tested with pytest (15 tests).