Inspiration
2,554 vehicles were stolen per day, one of which was our teammate Raghav's. Wondering why there wasn't an easy system to track stolen vehicles, especially with the amount of recording there is off the roads, from traffic stops to police body cams we built License To Steal.
What it does
License to Steal utilizes computer vision and a machine learning model to read license plates and cross-reference them with a database of stolen car license plates to detect whether the car is stolen. This can be done with live video or a recorded video that can later be uploaded, which can be integrated with the police body cam system. Furthermore, the application logs the latitude and longitude of the stolen license plate and creates a heat map so that the police can easily determine where there is a trend of car theft.
How we built it
We started off by training our own model, using YOLOv8, to detect license plates, and also used the base model for YOLOv8 to detect cars and associate a license plate to each car. Next, we used OpenCV to take only the license plate image and convert it to greyscale and then we used EasyOCR to read the text and determine the license plate number. Then we cross-referenced that license plate with a sample database of stolen vehicles (we made our own database as the police one wasn't available to us), and we made this database via MongoDB. We then recorded the longitude and latitude of each video based on its IP address, which then creates a heat map to show the localization of stolen vehicles.
Challenges we ran into
We have never worked with computer vision to this extent previously and had to learn how to implement computer vision through OpenCV and EasyOCR, systems that none of us have used before. Furthermore, training the model and running the inference of the model so that it wouldn't hinder the efficacy of our application, was a massive struggle.
Accomplishments that we're proud of
Going from a novice in Python to being able to implement a computer vision system. Making a high-accuracy reliable, reader for license plates.
What we learned
How to implement Computer Vision, data visualization via a heat map, utilizing databases, and how to integrate various libraries for machine learning.
What's next for License to Steal
We want to potentially work together directly with police officers and potentially with toll booths and traffic cameras, as an increase in data would increase the finding of stolen vehicles and make it a lot easier to track down stolen cars. Furthermore, we hope to make it an application so police officers can just keep it on their phones instead of needing to have their computers.
Log in or sign up for Devpost to join the conversation.