Gesture Recognition & Virtual Hand-Sign Checker

These are this tool's own real classification rules, applied directly to the 21 hand landmark points tracked each frame.

1. Click "Start" and allow camera access when prompted.
2. Hold up your hand clearly in frame.
3. Try each of the five recognized gestures to see how reliably each is detected under your current lighting.
4. Adjust lighting or hand position if detection feels inconsistent — clear, even light on your hand helps most.

This tool runs MediaPipe Hands, a hand-tracking model Google trained specifically for real-time, on-device use, entirely inside your browser via WebAssembly and WebGL. Nothing is sent to a server — the model downloads once and every frame after that is processed locally.

Each frame, the model returns 21 landmark points per detected hand: the wrist, and three joints plus a tip for each of the five fingers. Gesture classification itself is simple geometry on top of those points, not a separately trained gesture classifier — a finger counts as extended when its tip sits meaningfully further from the wrist than its own middle knuckle. Five fingers extended reads as an open palm; only the thumb reads as thumbs up; and so on.

Zoom's Reactions menu includes a physical Raise Hand gesture that recent desktop clients detect automatically when your camera is on — holding an open palm up near your face triggers the same reaction as clicking the button. Teams has an equivalent Raise Hand reaction, though gesture-based auto-detection is rolling out more gradually across client versions than in Zoom.

Discord has no native gesture detection, but hand gestures still matter if you use OBS or an Elgato Stream Deck plugin that triggers on-screen overlays or effects from a webcam gesture — those plugins depend on exactly the kind of clean, well-lit landmark tracking this tool checks for.

Keep your hand fully inside the frame and reasonably close to the camera — landmark tracking degrades quickly once a hand shrinks below roughly a tenth of the frame width. Backlighting is the most common failure mode: a bright window behind you silhouettes your hand, and the model can no longer resolve individual finger joints from a dark outline.

Frame rate matters too. A webcam struggling to hold 30fps, whether from USB bandwidth limits or driver issues, gives the tracking model fewer frames to lock onto a gesture, which shows up as flickering or missed detections rather than a hard failure.