โ† Back to Projects
๐Ÿ”ฌ Research & InnovationLive Demo

Real-Time Hand Keypoint Detection

Live webcam keypoint detection and gesture recognition: 21 landmarks per hand, every frame, entirely on-device. Try it live.

MediaPipeComputer VisionWebAssemblyWebGLReact

Overview

Real-time computer vision running in the browser: 21 hand keypoints tracked per frame with gesture classification (thumbs up, victory, open palm and more), GPU-accelerated via WebAssembly. All inference happens on the visitor's device. No frames are ever uploaded. The same class of keypoint-extraction technique I've applied to utility-infrastructure analysis, where precise landmark detection drives compliance measurements.

Key Features

  • โœ“21 keypoints per hand at ~30 FPS
  • โœ“7-class gesture recognition with confidence scores
  • โœ“GPU-accelerated, fully on-device, privacy by design
  • โœ“Lazy-loaded: model only downloads when the camera starts