Fall Detection System for Elderly (FDS-E) using Low-cost Camera Based on LSTM and OpenPose

Recently, many countries have been dealing with an increasing amount of elderly who are living alone. Unfortunately, fall incidents in the elderly tend to happen, and without proper dependable care would lead to fatal injuries. The elderly must be closely monitored under constant observation. Accide...

Full description

Saved in:
Bibliographic Details
Published in:IEEE International Instrumentation and Measurement Technology Conference (Online) pp. 1 - 6
Main Authors: Miawarni, Herti, Sardjono, Tri Arief, Setijadi, Eko, Wijayanti, Purnomo, Mauridhi Hery
Format: Conference Proceeding
Language:English
Published: IEEE 22.05.2023
Subjects:
Bayes methods
Cameras
Data models
Elderly
Fall detection
Fall Detection System
Low-cost Camera
LSTM
Older adults
Openpose
Sensitivity
Task analysis
ISSN:2642-2077
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, many countries have been dealing with an increasing amount of elderly who are living alone. Unfortunately, fall incidents in the elderly tend to happen, and without proper dependable care would lead to fatal injuries. The elderly must be closely monitored under constant observation. Accidental falls are a leading cause of injury, mortality, and mobility limitations in the elderly. The expenses of national healthcare systems are also significantly impacted by accidental falls. These findings highlight the importance of investing heavily in the study and advancement of fall detection and intervention technologies. It is essential to recognize them early and provide fast assistance by developing technologies that can simultaneously improve the standard and the safety of the elderly's living environment. In this paper, we present a computer vision algorithm for falling incident detection using a low-cost camera and deep learning model. Ultimately, this algorithm can be used for the cases of elderly, in which falling risk is high and fatal. Furthermore, one self-developed dataset was used to validate the suggested approach experimentally. We used an OpenPose-based feature using a low-cost camera input and then classified each event or activity as either fall or non-fall. Classification tasks are carried out by Long Short-Term Memory (LSTM) with multiple validation metrics. The LSTM in this work is also optimized using Bayesian method. Finally, our approach to recognize falls has accuracy of 99.5%, which is distinct from other past works.
ISSN:2642-2077
DOI:10.1109/I2MTC53148.2023.10175919