Parking Spot Availability Detector

Project Overview:

This project leverages computer vision and machine learning to accurately detect and classify parking spot occupancy using a single camera feed. The system was trained locally and combines a machine learning classifier with efficient real-time processing techniques to identify available parking spots.

Technologies Used:

• Libraries: OpenCV, Scikit-learn, Pandas, Pillow, Scikit-image, Matplotlib.

• Development Environment: The system was developed and trained locally for optimal performance.

Methodology:

• Classifier Training:

  • Using a labeled dataset of 3,046 images of empty parking spaces and 3,046 images of occupied spaces, a Support Vector Classifier (SVC) was trained with hyperparameter tuning via grid search, chosen for its effective handling of non-linear data boundaries.

  • The classifier achieved a 100% accuracy rate on the test data, making it highly reliable for distinguishing between empty and occupied spots. However, it’s important to note that while the test and training data are from different frames, they were all taken from the same video, with consistent lighting, weather conditions, etc. For a real-life project, more extensive and diverse training data would be needed to ensure robust performance across various conditions.

• Application on Footage:

  • The trained classifier was then applied to real-time footage to detect the status of parking spots.

  • To optimize performance, the system checks for potential changes in occupancy by comparing the average pixel differences between consecutive frames. Only spots with a pixel difference exceeding a threshold of 0.4 are passed to the classifier for further analysis. This approach minimizes computational load by avoiding unnecessary classification on unchanged spots.