FallVision: A benchmark video dataset for fall detection

Abstract

This article presents a comprehensive video dataset curated specifically for fall detection research, comprising categorized fall and no-fall videos. The dataset encompasses three primary categories of falls: falls from a bed, chair, and standing position. Initially collected as raw footage, these videos were subsequently processed to produce landmark videos, both with and without a background.

Recorded using handheld devices such as mobile phones and digital cameras, the dataset was sourced from voluntary participants, ensuring ethical compliance and informed consent. The dataset holds significant value for advancing fall detection algorithms, offering a robust platform for algorithm development and testing.

Fall detection systems are of paramount importance, particularly in scenarios where individuals are alone and unable to regain their footing post-fall or in cases where elderly individuals experience medical emergencies resulting in falls requiring immediate assistance. Leveraging this dataset, researchers can explore a plethora of techniques, including computer vision and deep learning, to devise and refine fall detection systems. Given its accessibility to researchers, this video dataset can be used in the advancement of fall detection technology to enhance safety measures for vulnerable populations.