We're really frustrated with looking for parking, and our team member Khalid has experience with using drones. We thought, why not have our own personal homing pigeon that would alert us when a spot was available. And thus, Spotty was born.
What it does
Spotty launches a drone from the entrance of a parking lot location and searches for open spots for the user. After using identifying available parking spaces, it then presents these results to the user in the form of a web app visible on the user's phone. The drone can then lands near the car or on its roof.
How we built it
For our model we built a SSD (Single Shot Detector) using a mobileNet backbone for our architecture. We trained the object detector on a combination of data we captured and labeled as well as part of the CARPARK dataset. We also used the dji SDK to communicate with the drone.
Our front end was built using ReactJS and express to make http requests.
Challenges we ran into
Some issues we ran into were the available GPUS being taken up on Google Cloud and some minor git/github problems. Another problem we faced was getting accurate gps coordinates for available parking spots.
Accomplishments that we're proud of
We initially thought that this idea would sputter, but after going out testing live data, we're happy with our current result.
What we learned
Some of the many things we learned throughout this hackathon include how to use express for http requests, train an object detector from scratch, collect and label data, integrate a machine learning model into an actual production app.
What's next for Spotty
Next up would be including on-street parking, as well as tighter integration with the DJI SDK and Google Maps. Also convert it to a mobile app. Maybe pitch Elon Musk on our idea.