OpenPotholeMap is a crowdsourced pothole mapping system. Using a phone camera and AI, it detects potholes in real time, logs them to a central database, and visualizes them on a live interactive map. Drivers get instant alerts (“⚠ Pothole 50m ahead”), and admins can verify reports or export data to authorities.
- Backend: Node.js + Express, MongoDB (Mongoose), Socket.io for real-time events, Roboflow Universe for detection, Firebase Auth + JWT for login.
- Frontend: React + TypeScript + Vite, Google Maps integration, camera streaming with WebSockets, REST + Socket.io clients.
- Infra: GitHub Actions for CI/CD, hosting via Vercel, cloud inference via Roboflow.
- Keeping video streaming stable while sending frames at a usable rate.
- Tuning the AI model confidence threshold (p > 0.6) to reduce false positives.
- Matching detections with GPS coordinates reliably.
- Designing a scalable geospatial data model that avoids duplicate entries.
- A working end-to-end pipeline: camera → AI → DB → live map → alerts.
- Clean Google Maps–like UI that feels intuitive.
- Admin dashboard and verification tools for data integrity.
- Bridging AI and geospatial mapping into a real safety tool.
- Real-time distributed system design is harder than it looks.
- Best practices in WebSocket communication, rate limiting, and recovery.
- Handling geospatial queries and data visualization at scale.
- Trade-offs between cloud inference and edge AI deployment.
- Detect more hazards (speed bumps, flooding, debris).
- Add gamification for user engagement.
- Direct integration with FDOT/authority APIs for repair reporting.
- Native mobile app for better performance.
- On-device inference to cut latency and cloud costs.