FPDS Dataset

Introduction

Here we release the new version of Fallen People Data Set (FPDS), a novel benchmark for detecting fallen people lying on the floor. It consists of 6982 images, with a total of 5023 falls and 2275 non falls corresponding to people in conventional situations (standing up, sitting, lying on the sofa or bed, walking, etc). Almost all the images have been captured in indoor environments with very different situations: variation of poses and sizes, occlusions, lighting changes, etc.

FPDS images with annotations

Download Version 2

Download Version 1

Best practice: Recommendations on using the dataset

FPDS dataset is divided into three dataset: training, validation and test. Any approach reporting results for the FPDS bechmark must be trained using any data except the provided test images. Furthermore, the test data must be used strictly for reporting of results alone - it must not be used in any way to train or tune systems, for example by running multiple parameter choices and reporting the best results obtained. For that purpose we recommend to use the training and validation sets which are provided on the same package.

YOLO weight files

YOLOv3 optimized trained weight files for the E-FPDS dataset (1 class-7000 epochs, 2 classes-4000 epochs and 20 classes-2000 epochs).

Other datasets

Ground-truth annotations of other public datasets.

Demo video

Demo video of the assistive robot with the ability of detecting fallen people while patrolling.

Acknowledgements

This work is supported by project PREPEATE, with reference number TEC2016-80326-R.