Autonomous Vehicle Tracking System

Here is the updated project description, reflecting that only the OCR logging-in and logging-out functionality was completed.

Inspiration

We were inspired by the need for a simple, automated way to manage vehicle access on a campus. Manually logging every car, bike, and visitor is slow and inefficient. We wanted to build the foundational first step: an automated "Digital Sentry" that could handle the check-in and check-out process, even with basic CCTV cameras.

What it does

Our project is the core "Digital Sentry" module for vehicle access control. Its primary function is to:

• Log Vehicles In: When a vehicle approaches the entry gate, it uses an ALPR (Automatic License Plate Recognition) model to read the license plate via OCR. It then saves this plate with an entry timestamp to a database.
• Log Vehicles Out: When a vehicle leaves from the exit gate, it captures the license plate again and adds an exit timestamp to the vehicle's log.

This creates a simple, searchable digital record of which vehicles entered and exited the campus and at what times.

How we built it

We focused on the entry and exit points. The system was built using:

1. ALPR (OCR) Model: We implemented a license plate recognition model to capture and transcribe the license plate text from a camera feed.
2. Database: We set up a simple database (like SQLite or MySQL) to store the logs. Each entry contains the license plate, an entry timestamp, and an exit timestamp (which is an empty field until the vehicle leaves).
3. Python Script: A main script connects the camera feed to the OCR model and handles the database operations (writing the new log on entry, updating the log on exit).

We did not build the dashboard, internal tracking ("Digital Twin"), or the automated "AI Officer" violation detection.

Challenges we ran into

Our single biggest challenge was the reliability of the OCR model. We ran into constant problems due to:

• Bad Image Quality: The simple, low-resolution CCTVs provided blurry or pixelated images.
• Poor Lighting: At night, in heavy rain, or with sun glare, the camera struggled to capture a clear image of the plate.
• Angled or Dirty Plates: Vehicles approaching at an angle, or having dirty/damaged plates, made the OCR fail completely.

A significant portion of our time was spent just trying to get a semi-reliable reading, and we found it was a much harder problem than we first anticipated.

Accomplishments that we're proud of

We are proud of successfully building the "book-ends" of a vehicle management system. Despite the challenges, we created a functional pipeline that captures an image, processes it through an OCR model, and correctly logs the data in a database. We built the essential data-collection layer that all other advanced features (like a dashboard or tracking) would need to be built upon.

What we learned

We learned that real-world computer vision is incredibly difficult. A model that works perfectly on clean data (like the "Digital Sentry" concept) can fail constantly when faced with low-quality cameras, bad weather, and unpredictable angles. We learned the hard way that the core "logging" feature, which we thought was simple, is actually a major hurdle that requires a lot of refinement to get right.

What's next for Autonomous Vehicle Tracking System

The next steps would be to build all the other modules we originally planned.

1. Solve the OCR problem: First, we need to dramatically improve the logging accuracy, possibly by adding image enhancement filters or using better cameras.
2. Build the Dashboard: Create a web interface to view the live logs, see which vehicles are currently on campus, and search for a vehicle's history.
3. Build the "Digital Twin": Develop the internal tracking logic (using color/type) to track vehicles between cameras, since we now know we can't rely on OCR alone.
4. Create the "AI Officer": Add the other AI models to detect speeding, parking, and safety violations.

Built With

• opencv
• python
• ultralytics
• yolo