Fall detection dataset UG (Underground GATE)

Fall Detection Dataset UG (Underground GATE)

This dataset contains 40 short videos (~5 seconds each) that contain a person either falling or some other activity (walking, sitting, standing, jumping, etc.). Its main purpose is performance evaluation of fall detection methods.

Contents

– videos/: 40 `.avi` files (fall and non-fall events)
– labels.csv: A CSV file mapping each video to its class label (`fall` or `not_fall`)
– LICENSE.txt: Dataset license (CC BY 4.0)
– citation.bib: BibTeX citation entry

Motivation

Fall detection from video data is a very active topic in the field of Computer Vision. This dataset is made specifically for testing an already trained ML method. The videos are short and contain only one type of movement. This removes the need to implement sliding (and overlaping) windows which extract portions of the video and allows the user to focus only on the peroformance of the fall detection method. The following points characterize the set:

– Indoor scenario
– Constant brightness
– Only one moving person
– Static background
– High contrast between the moving person and the background

Recording Details

– Resolution: 1920×1080 
– Frame Rate: 30 FPS
– Video Duration: ~5 seconds each
– Format: AVI
– Camera: Lenovo ThinkPad E14 Gen5 Webcam
– Lighting: Constant indoor LED lighting
– Scene: Indoor, static camera, single moving person/object

Labels

The `labels.csv` file contains:

filename,label
video_01.avi,not_fall
…
video_21.avi,fall
…