Smart Parking Lot Classifier

This is the classifier script used to deploy a Deep Learning model on Raspberry Pi 4 to detect whether a parking space is empty or occupied.

This is a part of the Smart Park - Capstone Senior Project.

Model Description

• The model is trained using Transfer Learning method with TensorFlow Lite to be used on Raspberry Pi 4.
• Dataset: PKLot - A robust dataset for parking lot classification consisting of 695,899 images of occupied and empty parking spaces captured from two parking lots with different camera views and under different weather conditions.
• This model only works with pre-labeled spaces.

Model Benchmark

• When being tested at a parking lot with 222 spaces being observed:

  • The model achieved an accuracy of 85%.
  • Precision is 73%
  • Recall is 88% which is the proportion of actual vacant slots identified correctly.

  

Built With:

• Python
• OpenCV
• TensorFlow Lite