Why Pose Estimation Works Here
17-point skeletal tracking
Tracks the standard COCO joint set — head, shoulders, elbows, wrists, hips, knees, ankles — as connected skeleton lines updated every frame.
Multi-person tracking, stable IDs
Turn on Multi-Person to track everyone in frame at once, each with their own stable color ID — most free browser pose tools only handle one person.
Live joint-angle measurement
Reads elbow and knee bend in real degrees as you move, not just a skeleton overlay — useful for form-checking, biomechanics, or sports/dance analysis.
Gesture detection, no training step
Recognizes held poses like "Arms Up" or a T-pose and labels them live. Geometric, not a trained classifier — works instantly, nothing to train.
Runs at up to 60fps
Pose detection is GPU/WASM-accelerated in-browser, keeping pace with live camera input rather than processing frame-by-frame after the fact.
Skeleton visuals, trails & smoothing
Joint connections render as a proper skeleton overlay with adjustable line style, width, and color, plus motion trails and a jitter-smoothing toggle.
Export every joint — and every angle — as CSV
Per-person joint coordinates and computed joint angles export frame-by-frame as CSV telemetry — useful for motion analysis or a separate animation pipeline.
Stream live to TouchDesigner
Pose data can broadcast over the same OSC/WebSocket telemetry stream as blob tracking, for real-time performance visuals driven by a dancer's movement.
Frequently Asked Questions
How many body points does the pose tracker detect?
17 skeletal joints per person — the standard COCO keypoint set covering head, shoulders, elbows, wrists, hips, knees, and ankles.
Can it track more than one person at once?
Yes — enable the Multi-Person toggle to track every skeleton in frame at once, each with its own stable color ID. It defaults to a single tracked subject for the clearest single-skeleton view, since that's what most dance/movement use cases want.
Does pose tracking work from a recorded video, or only live?
Both. Load a video file for frame-by-frame pose tracking, or use a live webcam feed — the same detection pipeline runs either way.
Can I export the skeleton overlay as a video?
Yes — Record captures the composited output (source video plus skeleton overlay) as WebM or MP4.
Can it measure joint angles, not just show a skeleton?
Yes — enable Joint Angles to read elbow and knee bend in real degrees as the subject moves, and export those angles alongside the joint coordinates in the CSV telemetry.
Does it recognize specific poses or gestures?
Yes, for simple held poses — Gesture Labels detects things like "Arms Up" or a T-pose geometrically (not a trained ML classifier), so it labels them instantly with no training step.
Is this free motion capture?
For 2D movement, yes: it records 17 joints per person from ordinary video and exports them per frame as CSV. It doesn't produce depth or a 3D rig, so it suits movement analysis and 2D animation reference rather than a full mocap pipeline. See the browser motion capture guide.
