Dataset Card for PKLot

PKLot is a robust dataset for parking lot classification containing 12,416 images captured from three different parking lots (PUCPR, UFPR04, UFPR05) under various weather conditions (sunny, cloudy, rainy). Each image includes detailed annotations for individual parking spaces with occupancy status, resulting in approximately 695,900 segmented parking space instances.

This is a FiftyOne dataset with 12,416 samples.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = load_from_hub("Voxel51/PKLot")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

The PKLot dataset is a comprehensive parking lot classification dataset designed for computer vision research in parking space detection and occupancy classification. The dataset contains:

• 12,416 high-resolution images (1280×720 pixels)
• 3 different parking lots: PUCPR (Pontifícia Universidade Católica do Paraná), UFPR04, and UFPR05 (Federal University of Paraná)
• 3 weather conditions: Sunny, Cloudy, and Rainy
• Time-series data: Images captured at 5-minute intervals throughout different days
• ~695,900 parking space instances: Each image contains 45-100 annotated parking spaces
• Rich annotations: Each parking space includes polygon boundaries and occupancy status

The dataset is particularly valuable for:

• Parking space detection algorithms

• Occupancy classification models

• Temporal analysis of parking patterns

• Weather-robust computer vision systems

• Smart city and intelligent transportation system research

• Curated by: Paulo R. L. de Almeida, Luiz S. Oliveira, Alceu S. Britto Jr., Eunelson J. Silva Jr., Alessandro L. Koerich

• Funded by [optional]: Federal University of Paraná (UFPR) and Pontifícia Universidade Católica do Paraná (PUCPR)

• Shared by [optional]: Vision, Robotics and Imaging Laboratory (VRI) - UFPR

• Language(s) (NLP): Not applicable (computer vision dataset)

• License: Creative Commons Attribution 4.0 International License

Dataset Sources

• Repository: PKLot Official Page
• Paper: Almeida et al., "PKLot – A robust dataset for parking lot classification", Expert Systems with Applications, 2015
• Download: PKLot.tar.gz (4.6GB)

Uses

Direct Use

The PKLot dataset is intended for:

1. Parking Space Detection: Training and evaluating algorithms to detect individual parking spaces in aerial/surveillance imagery
2. Occupancy Classification: Developing models to classify parking spaces as occupied or vacant
3. Temporal Analysis: Studying parking patterns over time and predicting future occupancy
4. Weather Robustness: Testing computer vision models under different weather conditions
5. Smart Parking Systems: Developing real-time parking availability systems
6. Benchmark Dataset: Comparing performance of different parking detection algorithms

Out-of-Scope Use

This dataset should not be used for:

• Identifying individuals or vehicles (images are not high-resolution enough for identification)
• Real-time commercial applications without proper validation
• Training models for different parking lot layouts without additional data
• Applications requiring night-time or low-light conditions (dataset only contains daylight images)

Dataset Structure

FiftyOne Dataset Fields

Each sample in the FiftyOne dataset contains the following fields:

Field	Type	Description
filepath	string	Path to the image file
source	string	Parking lot identifier (pucpr, ufpr04, ufpr05)
weather	Classification	Weather condition label (sunny, cloudy, rainy)
date	date	Date of image capture (YYYY-MM-DD)
parking_timestamp	datetime	Full timestamp of capture (YYYY-MM-DD HH:MM:SS)
parking_spaces	Polylines	Collection of parking space polygons

Parking Space Annotations (Polylines)

Each parking space polyline contains:

Attribute	Type	Description
label	string	Always "parking_space"
points	list	Normalized polygon vertices [[x,y], ...] in [0,1] range
index	int	Unique parking space ID (1-100)
closed	bool	True (parking spaces are closed polygons)
filled	bool	True (for visualization as filled polygons)
occupancy_status	string	"occupied", "not occupied", or "unknown"
space_id	int	Parking space identifier

Dataset Statistics

• Total Samples: 12,416 images
• Parking Lots Distribution:
  • PUCPR: ~4,474 images
  • UFPR04: ~3,791 images
  • UFPR05: ~4,152 images
• Weather Distribution:
  • Sunny: ~50% of images
  • Cloudy: ~35% of images
  • Rainy: ~15% of images
• Temporal Coverage: September 2012 - April 2013
• Capture Frequency: 5-minute intervals

Dataset Creation

Curation Rationale

The PKLot dataset was created to address the lack of robust, publicly available datasets for parking lot classification research. Key motivations included:

1. Standardized Benchmark: Providing a common dataset for comparing parking detection algorithms
2. Real-World Conditions: Capturing diverse weather conditions and lighting variations
3. Temporal Dynamics: Understanding parking patterns over time
4. Scale: Offering sufficient data for training deep learning models
5. Reproducible Research: Enabling researchers to compare results on the same dataset

Source Data

Data Collection and Processing

The data collection process involved:

1. Camera Setup: Fixed surveillance cameras installed at three parking lots
2. Capture Protocol: Automatic image capture every 5 minutes during daylight hours
3. Weather Diversity: Deliberate collection across different weather conditions
4. Time Period: Data collected from September 2012 to April 2013
5. Image Resolution: All images captured at 1280×720 pixels
6. Quality Control: Manual verification of image quality and weather labels

Who are the source data producers?

The data was produced by researchers at:

• Federal University of Paraná (UFPR), Brazil
• Pontifícia Universidade Católica do Paraná (PUCPR), Brazil
• Vision, Robotics and Imaging Laboratory (VRI)

Annotations

Annotation process

The annotation process consisted of:

1. Parking Space Delineation: Manual marking of parking space boundaries using rotated rectangles and polygons
2. Occupancy Labeling: Binary classification (0=vacant, 1=occupied) for each parking space
3. XML Format: Annotations stored in XML files with both rotated rectangle and contour representations
4. Consistency: Same parking space IDs maintained across all images from the same parking lot
5. Validation: Cross-checking of annotations for accuracy

Who are the annotators?

Annotations were created by the research team at the Vision, Robotics and Imaging Laboratory (VRI) at UFPR, with quality control and validation performed by multiple team members.

Personal and Sensitive Information

The dataset contains surveillance imagery of parking lots but:

• Images are taken from elevated positions at resolution insufficient for personal identification
• No license plates or individual features are distinguishable
• Focus is on parking space occupancy, not vehicle or person identification
• The dataset complies with privacy regulations for public space surveillance

Bias, Risks, and Limitations

Known Limitations

1. Geographic Bias: All data from two universities in Curitiba, Brazil
2. Temporal Bias: Limited to daylight hours (approximately 6 AM to 7 PM)
3. Seasonal Bias: Data from September 2012 to April 2013 only
4. Weather Distribution: Unbalanced weather conditions (more sunny than rainy days)
5. Parking Lot Types: Only university parking lots, may not generalize to other environments
6. Camera Angles: Fixed camera positions, limited viewpoint diversity

Technical Limitations

• No night-time or low-light conditions
• No snow or extreme weather conditions
• Fixed parking space layouts (no dynamic spaces)
• Resolution limitations for fine-grained vehicle classification

Recommendations

Users should be aware that:

1. Generalization: Models trained on this dataset may need adaptation for different geographic locations or parking lot types
2. Lighting Conditions: Additional data may be needed for 24-hour operation systems
3. Real-time Deployment: Validation on target deployment environment is essential
4. Privacy Considerations: Ensure compliance with local regulations when deploying models
5. Weather Robustness: Test model performance across all weather conditions in the dataset

Citation

BibTeX:

@article{almeida2015pklot,
  title={PKLot--A robust dataset for parking lot classification},
  author={Almeida, Paulo and Oliveira, Luiz S and Silva Jr, Eunelson and Britto Jr, Alceu and Koerich, Alessandro},
  journal={Expert Systems with Applications},
  volume={42},
  number={11},
  pages={4937--4949},
  year={2015},
  publisher={Elsevier}
}

APA:

Almeida, P. R., Oliveira, L. S., Britto Jr, A. S., Silva Jr, E. J., & Koerich, A. L. (2015). PKLot--A robust dataset for parking lot classification. Expert Systems with Applications, 42(11), 4937-4949.

Glossary

• Parking Space: Individual parking slot/bay in a parking lot
• Occupancy Status: Binary classification of whether a parking space contains a vehicle
• Polyline: Closed polygon defining the boundary of a parking space
• Rotated Rectangle: Bounding box with rotation angle for non-axis-aligned parking spaces
• Normalized Coordinates: Coordinates scaled to [0,1] range relative to image dimensions

More Information

For more information about the dataset, visit the official PKLot page or read the original paper.

Dataset Card Authors

• Harpreet Sahota (FiftyOne integration and dataset card)
• Original dataset by Paulo R. L. de Almeida et al.

Dataset Card Contact

For questions about the original PKLot dataset, please contact the Vision, Robotics and Imaging Laboratory at UFPR.