Computer vision / 2026 / Live in the browser
SAT0RU
Hand-tracked cursed-technique visualiser: MediaPipe landmarks drive a Three.js particle engine, with an ML classifier picking the technique in real time.
01
The problem
Gesture projects usually hard-code one pose and break the moment lighting changes.
02
How I approached it
A trained classifier reads 21-point hand landmarks per frame and votes across a rolling window; a deterministic geometric fallback takes over whenever confidence drops, so the visual never freezes.
03
The outcome
Techniques fire reliably at webcam frame rate on a laptop GPU, with graceful degradation instead of a dead screen.
Related work
Build voxel structures in the air over your webcam feed: pinch to place, pinch and drag to paint, make a fist to turn the build. A second synced page drives presets and shows live status.
A 3D globe you steer with your hands: pinch to zoom, move your finger in mid-air to spin it. No camera? Drag, scroll, pinch with two fingers or use the arrow keys.
ASL fingerspelling to text from a webcam. It reads hand geometry instead of pixels, so all 26 letters work with no training, and two seconds of calibration per letter fits it to your own hand.