Parking Spot Detection Using AI
Juliann Weir-Jackson, Vedant Jani, Divyansh Gupta, Nicolas Ayala
Faculty Advisor: Dr. Debasis Mitra, Dept. of Computer Science, Florida Institute of Technology
Motivation
- Parking spaces at FIT can be very busy, hence scarce
- People driving around at low speeds looking for parking causes traffic congestion, it's time consuming, and adds to pollution
- This can be a major problem during peak hours
- There are many car detection Neural Network models for parking detection but lack user interface.
Features
- Users will be able see insights for each parking lot on the main page such as:
- % parking lot occupancy
- % parking lot availability
- specific section/s available
- User can select between two parking lots (one currently working, one currently in progress) featured on the main page of the web application
- Upon selection of desired parking lot users will be able to see a 2-D map of the chosen parking lot where available parking zone will be highlighted in a green box and unavailable parking zones will be highlighted in a redbox rendered on the 2-D map, along with the insights specified above for each parking lot
- Users will also be able to report whether data provided on the website is accurate
- If a parking section is marked as available but it is actually unavailable and vice versa
Goals
- To be able to provide the FIT community with a system to monitor a parking lot occupancy in real time using a user friendly web app.
- Display a Real-time mockup map of a parking lot through a web app (available and unavailable parking sections)
- To facilitate and make the search for a parking space in FIT’s parking lots more efficient.
Evaluation
- Majority of the time spent on this project was focused on getting the neural network classification to be as accurate as possible through training.
- Used transfer learning in order to increase efficiency of network , decrease the amount of time needed to train the neural network.
- Created a data simulation to tackle the problem of lack of data and train Neural Network effiectively.
Future Work
- Including a linear regression model that will give users hourly predictions of parking lot occupancy
- Increase in the number of parking lots.
- Increase the area of focus in each parking lot. Due to camera angle, data only had a partial view of the parking lot which will be improved using better camera.
- Incorporate the feedback that users give in the report to help train the Neural Network
- Host the system on a server ,system is currently being hosted locally .
Figure 1. System Architecture Diagram
Figure 4. User Interface of System
Figure 3. Final Training Loss and Accuracy Graph
Figure 2. Initial Training Loss and Accuracy Graph Loss vs Epoch#