Parking Spot Detection and Classification

This project utilizes computer vision techniques and deep learning to detect and classify free parking spots in a given video.

Overview

The system processes a video feed of a parking lot, identifying whether each parking spot is occupied or vacant. It employs a Convolutional Neural Network (CNN) trained on labeled images of parking spots to make predictions in real-time.

Features

• Video Input: Supports any video file where parking spots need to be analyzed.
• Real-time Classification: Each frame of the video is analyzed to determine the status of each parking spot.
• Output Video Generation: An annotated video is generated showing identified occupied and vacant spots.
• Evaluation Metrics: Provides metrics such as accuracy, precision, recall, and F1 score to evaluate the model's performance.

Installation

Prerequisites

• Python 3.x
• OpenCV (pip install opencv-python)
• TensorFlow (pip install tensorflow)
• Scikit-image (pip install scikit-image)

Setup

1. Clone the repository:

   git clone https://github.com/adrianna-d/Parking_detection.git
   cd parking_project

2. Install dependencies:

   pip install -r requirements.txt

3. Download or configure the video and mask file paths in your code (video_path and mask_path variables).

Usage

1. Ensure the video and mask are correctly set up.

2. Run the main script:

   python parking_spot_detection.py

3. The program will process the video, classify parking spots, and generate an annotated output video.

Model Training

• The CNN model is trained using images from two directories:

  • empty_folder_path: Images of empty parking spots.
  • occupied_folder_path: Images of occupied parking spots.

• The model architecture includes convolutional layers with max-pooling, fully connected layers, and dropout to prevent overfitting.

Files

• parking_spot_detection.py: Main script for detecting and classifying parking spots.
• requirements.txt: List of Python dependencies.
• parking_1920_1080.mp4: Example input video
• mask_1920_1080 - Mask for the project
• project_presentation - .pdf of what was done for the project

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• This project was developed as part of IronHack bootcamp project on computer vision.
• @sofiaggoncalves is a co-author of the project.