🚧 OpenPotholeMap

🫳 What it does

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.


👊 How we built it

  • 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.

💪 Challenges we ran into

  • 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.

🙌 Accomplishments we’re proud of

  • 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.

🫶 What we learned

  • 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.

🤞 What’s next

  • 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.

Built With

+ 1 more
Share this project:

Updates