GobblerPark

ML-powered real-time parking lot monitoring

Comment

The Problem

Campus parking is unpredictable, inefficient, and stressful:

  • Wasted Time: Students are often late to classes, exams, and meetings because parking isn’t predictable.
  • Congestion: Cars circling for spots clog campus roads and worsen traffic flow.
  • Accessibility Gap: Students with disabilities or tight schedules face the greatest challenges.
  • Environmental Impact: Extra circling wastes fuel and increases emissions.
  • No Analytics: Administrators lack real data on usage, turnover, and demand.

Our Solution: GobblerLot

GobblerLot is an ML-powered parking monitoring system that makes campus parking smarter and more efficient:

  • Real-Time Detection: Custom CNN detects free vs. occupied spots using existing camera systems.
  • Student App: Displays available spaces before students leave home.
  • Admin Dashboard: Provides analytics like turnover, duration, and congestion patterns.

The Technology Behind It

  • Custom CNN Model:
    • Trained on 38,000 images from the CNR Parking Lot dataset.
    • Achieved 99.6% accuracy across diverse conditions (lighting, weather, angles).
    • Just 12k parameters vs. YOLOv8’s ~3.2M → <100ms latency, deployable on edge devices.
    • Outperformed fine-tuned YOLO models on parking occupancy detection.
  • Lower Cost: Smaller model = reduced computational load on VT servers = cheaper to run.
  • Admin Setup: Upload a lot photo once, draw boundaries manually (saved permanently), and scale across multiple lots.
  • Student Experience: Live updates, instant free/occupied views, and rich analytics on lot usage.

Obstacles & How We Overcame Them

  • Designing a CNN small enough to run in real time while maintaining high accuracy.
  • Generalizing across variable lighting, weather, and camera angles.
  • Building a one-time setup process that remains scalable for multiple lots.

Impact Achieved

  • Higher accuracy and lower latency than YOLO-based models.
  • Edge deployment for real-time updates without heavy server usage.
  • A single platform that benefits both students (convenience) and administrators (data-driven insights).

Key Takeaways

  • Efficiency matters as much as accuracy when deploying ML in the real world.
  • Small usability choices (like manual boundary setup) unlock scalable adoption.
  • Real-time systems require careful optimization for both speed and reliability.

Next Steps & Vision

  • Expand to all campus lots with centralized coverage.
  • Add predictive analytics to forecast spot availability.
  • Integrate with navigation apps and extend to city garages and event venues.

Built With

Share this project:

Updates