Partially Occluded Pedestrian Dataset used in "Replacing the Human Driver: An Objective Benchmark for Occluded Pedestrian Detection" Gilroy et al 2023.
This dataset contains 820 person instances in 724 images for the characterization of detection model performance for partially occluded pedestrians. Pedestrian instances are occluded by a diverse mix of objects across a range of occlusion levels from 0 to 99% occluded.
Images are annotated using the objective occlusion level annotation method described in An objective method for pedestrian occlusion level classification Gilroy et al 2022.
Annotation files are in COCO format.
Replacing the human driver: An objective benchmark for occluded pedestrian detection
@article{gilroy2023replacing,
title={Replacing the human driver: An objective benchmark for occluded pedestrian detection},
author={Gilroy, Shane and Mullins, Darragh and Parsi, Ashkan and Jones, Edward and Glavin, Martin},
journal={Biomimetic Intelligence and Robotics},
pages={100115},
year={2023},
publisher={Elsevier}
}
An objective method for pedestrian occlusion level classification
@article{gilroy2022objective,
title={An objective method for pedestrian occlusion level classification},
author={Gilroy, Shane and Glavin, Martin and Jones, Edward and Mullins, Darragh},
journal={Pattern Recognition Letters},
volume={164},
pages={96--103},
year={2022},
publisher={Elsevier}
}