Parking Spot Detection Dataset

Project Overview

This project aims to create a high-quality dataset for parking spot detection using outdoor parking lot videos. The dataset includes annotated images for training machine learning models, specifically YOLO, to detect parked cars and empty parking spots. This project also includes preprocessing scripts, data augmentation, and model evaluation tools to assist in training object detection models.

Dataset Structure

The dataset is organized into training, validation, and test sets in YOLO format, along with the relevant metadata.

Installation

1. Clone the Repository:

git clone https://github.com/your-username/parking-spot-dataset.git
cd parking-spot-dataset

2. Install Dependencies:

Set up a Python environment and install the required packages listed in requirements.txt (if provided):

pip install -r requirements.txt

3. Extract Frames from Videos:

Run the script to extract frames from raw videos:

python src/extract_frames.py --input data/raw_videos/ --output data/extracted_frames/

4. Data Augmentation (Optional):

You can apply data augmentation to increase the variability of the dataset:

python src/data_augmentation.py --input data/extracted_frames/ --output data/data_augmentation/

5. Train YOLO Model (Optional):

Use the YOLOv5 configuration to train the model on the dataset:

python src/train_yolo.py --data output-dataset/parking-spot-dataset-v1.0/data.yaml --weights yolov5s.pt

Usage

1. Dataset: The final dataset (version 1.0) is organized into train, validation, and test sets, ready for model training.
2. Scripts: Use the provided scripts for frame extraction, annotation verification, and model evaluation.
3. Notebooks: Explore the dataset, analyze metrics, and visualize results using the Jupyter notebooks in the notebooks/ directory.

Contributing

We welcome contributions! If you'd like to contribute to this project, please fork the repository, create a new branch, and submit a pull request with detailed information.

License

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

References