About SignBridge
SignBridge is a real-time American Sign Language recognition web app built as a college mini project. It demonstrates how modern in-browser machine learning can make accessibility tools fast, private, and easy to deploy.
Project overview
SignBridge captures video from your webcam, tracks 21 hand landmarks per frame using MediaPipe, and applies a geometric rule-based recognizer to identify ASL letters A through F. Recognized letters are aggregated into words and sentences that you can copy or export.
Technologies used
- React 19 + TanStack Start (SSR-ready routing)
- MediaPipe Tasks Vision — HandLandmarker (WASM, runs in-browser)
- TypeScript, Tailwind CSS v4, shadcn/ui components
- Rule-based landmark classifier (no training required)
- Lucide icons, Sonner toasts
How it works
- You grant camera access; a live video stream begins.
- Each frame is passed to MediaPipe HandLandmarker, producing 21 (x, y, z) landmarks.
- Landmarks are drawn as an overlay so you can see the model's view.
- A geometric recognizer inspects finger extension and thumb position to classify the gesture.
- Predictions above your confidence threshold, held stable across frames, are appended to the transcript.
Developer
Built by a student developer as a mini project demonstration. Replace this placeholder with your name, roll number, and department before submission.