Checkpoint

Inspiration

The idea came from the need to improve security during school pickups, where unauthorized individuals could potentially pick up students. We wanted a seamless, tech-driven solution to ensure student safety while making the pickup process more efficient and stress-free for parents and schools.

What it does

Our system uses object detection and license plate recognition to verify vehicles in school pickup lines. It scans cars, checks them against a secure database, and notifies teachers when the right car arrives. Parents get real-time updates when their child checks in or out, and guest pickups are also securely verified through a driver-side app.

How we built it

We started by using a Raspberry Pi 5 to manage the real-time camera feed for object detection and license plate recognition. The key component of our detection system is YOLOv8, a state-of-the-art object detection model, which we optimized to specifically detect vehicles and zoom in on their license plates. For text extraction, we integrated EasyOCR, an optical character recognition library, to read the license plate characters from the detected images.

The detected license plate text is cross-referenced with a Firebase Realtime Database, where the school’s authorized vehicle information is stored. If a match is found, the system sends the student’s name and verification status to a frontend UI designed for teachers, allowing them to quickly call the student for pickup.

We also built a driver-side app prototype on Figma where parents can register their vehicle by scanning their government-issued ID. This ID is verified by the school’s administration for approval. Once verified, parents can add their vehicles under their child’s profile in the app. This app also allows for secure guest pickups, with verification of the guest’s ID through the same process.

The entire system is designed to work in real-time, ensuring smooth, secure communication between the Raspberry Pi, the Firebase backend, and the frontend display, providing a seamless experience for both the teachers and the parents.

Challenges we ran into

Integrating the object detection model and OCR posed several challenges, especially given the limited time during the hackathon. Optimizing YOLOv8 for license plate detection was tricky. Initially, the model struggled with accurately detecting and focusing on license plates in varying lighting conditions or from odd angles. We had to fine-tune the model and retrain it on a dataset of vehicle images to improve its accuracy for this specific use case.

Handling real-time updates between the Raspberry Pi, Firebase, and the frontend was another major hurdle. Since our system is designed to instantly verify vehicles and update the teacher-side display, it required low-latency communication across all components. Balancing this with the need for accurate detections from YOLOv8 and OCR without creating delays was technically demanding.

Additionally, OCR integration with EasyOCR had its own challenges, particularly when dealing with license plates that were dirty, had reflections, or were partially obstructed. We had to experiment with different image preprocessing techniques to ensure the text recognition was as accurate as possible.

Finally, developing the driver-side app prototype to handle secure ID verification and vehicle registration added another layer of complexity. We wanted the process to be secure but also user-friendly, which meant designing the app’s flow to make ID scanning and guest pickup management simple for parents while ensuring robust security for the school.

Accomplishments that we’re proud of

We successfully built a working model that scans cars, reads license plates as well as a prototype demo to simulate our product that verifies ‘authorized’ pickups via a hack-uta badge and face, and updates a real-time frontend.

What we learned

We learned how to optimize object detection models for specific use cases like license plate recognition and efficiently handle real-time data using Firebase. We also gained experience in app development and cloud integration for both driver and teacher interfaces.

What’s next for Checkpoint

Next, we plan to enhance the system’s scalability, improve the user interface, and add more security features for even tighter security. We also aim to integrate a comprehensive dashboard for school administrators to manage pickup records, notifications, and guest approvals.