A Pose Estimation-Based Fall Detection Methodology Using Artificial Intelligence Edge Computing

Wan-Jung Chang,Chia-Hao Hsu,Liang-Bi Chen

doi:10.1109/access.2021.3113824

Wan-Jung Chang, Chia-Hao Hsu + Show 1 more

Open Access

https://doi.org/10.1109/access.2021.3113824

Copy DOI

Abstract

As the population worldwide continues to age and the percentage of elderly people continues to increase, falls have been become the second leading cause of death from accidental or unintentional injuries. Although many imaging sensing devices have been used to detect falls for elderly people, most involve using the Internet to transfer images taken by a camera to a large back-end server, which performs the necessary calculations; however, algorithm limitations and computational complexity may cause bottlenecks in image outflow, and the image transfer can result in privacy problems. To address these problems, in this paper, an artificial intelligence (AI) fall detection method is proposed that operates using an edge computing architecture, called the pose estimation-based fall detection methodology (PEFDM), which is based on a human body posture recognition technique. The proposed PEFDM can effectively reduce the computational load, runs smoothly on mainstream edge computing systems and possesses artificial intelligence computing capabilities. By using edge computing, the privacy and upload bandwidth issues caused by image outflow can be eliminated. Experiments with real humans show that the PEFDM can detect falls for elderly people with a recognition accuracy of up to 98.1%.

Highlights

Due to modern advances in medical treatments and public health, the average human life span has increased substantially
To accurately detect and report falls in real time and address the privacy problems caused by image transmission, this paper proposes a fall detection method based on artificial intelligence edge computing; it combines image recognition with edge computing to avoid the limitations of local computing resources
This article addressed three major issues that occur in automated fall detection: security and privacy issues caused by transferring fall detection images over a network, network bandwidth issues caused by image transmission, and cost issues in system construction

Summary

INTRODUCTION

Due to modern advances in medical treatments and public health, the average human life span has increased substantially. According to statistics from the World Health Organization (WHO), an American individual’s average life expectancy is 79.3 years and increases every year [1]. In this regard, the United States has created an aging society. Many devices have been created to detect and report falls in elderly individuals These devices can be roughly divided into wearable systems and image recognition systems. To accurately detect and report falls in real time and address the privacy problems caused by image transmission, this paper proposes a fall detection method based on artificial intelligence edge computing; it combines image recognition with edge computing to avoid the limitations of local computing resources.

RELATED WORKS

EXPERIMENTAL DESIGN AND RESULTS

COMPUTATIONAL EFFICIENCY EXPERIMENTS

FALL DETECTION ACCURACY EXPERIMENTS

CONCLUSION