Inspiration

As a commuter student, it is really hard to find parking, especially in busy places like the co-rec during peak hours. due to this, I am always leaving for class an hour before it begins. Our team wanted to make a application that can help students see how many cars are in the parking lot before they leave.

What it does

Our website allows for us to upload an image and and see how many spots are taken. This is done through an object detection model that we trained and then connected to our website.

How we built it

We took parking lot image datasets from Kaggle and trained a model on Roboflow to classify parking spaces as "occupied" and "unoccupied." We trained around 76 images with this classification technique until we deemed the model to be accurate enough.

Challenges we ran into

The model would consider random objects like shadows and road patterns as objects, making the space seem occupied when it really wasn't. Furthermore, our model could not differentiate an unoccupied space from a wide space like a road or driveway. We fixed this by adding extra classes like a car class and a road class, to help the model have more parameters to train on.

Accomplishments that we're proud of

We managed to prevent a road and unoccupied space from being the same thing by tuning the model overlap threshold (my first ever experience with feature selection and tuning). our model can also very accurately classify a space as occupied or having a car.

What we learned

We learned how to use Kaggle datasets, Roboflow model training, and model tuning and feature selection.

What's next for Purdue Parking Pro

if we can connect it to a camera for a live parking feed then it will be more applicable for daily life. The model currently uses static images.

Share this project:

Updates