smart-city

Smart City 🏙️

Live: https://smart-city-cyan.vercel.app/

Hackathon platform where kids train computer-vision models to detect urban hazards (potholes, faded crosswalks, litter, broken signs), tour real Israeli streets virtually (default city: Sderot), and watch their model’s detections appear as pins on a real map.

Built on the proven thinkCV Hakaton pipeline: browser frame extraction, YOLO dataset export, free Colab GPU training, TF.js in-browser inference.

Modules

• 🗺️ City map — MapLibre + CARTO dark, hazard pins, heatmap, coverage layer
• 🚶 Street tour — Google Street View embed + geotagged frame viewer + 🔴 live tour: the trained model runs on every frame; detections above threshold auto-save as map pins
• 🎓 Training studio — collect → curate → tag → export YOLO ZIP → Colab → load model back; shared pool merges all teams into a “city model”
• 🏭 Data factory (admin) — drive video + GPX → sharp geotagged frames → city dataset
• 📋 Hazard board — moderation (approve/reject), team leaderboard, CSV report for the municipality

Stack

Static ES modules (no build step) · MapLibre GL · TensorFlow.js · JSZip · Supabase (auth, sc_* tables with RLS, public storage) · YOLOv8n trained on free Colab GPU.

Hebrew RTL throughout. Storage keys are ASCII-only (Hebrew stays in DB rows).