Falling Estimation Based on PoseNet Using Camera with Difference Absolute Standard Deviation Value and Average Amplitude Change on Key-Joint

Abstract

Elderly people need special attention and some of them need to be monitored continuously and in real-time. Fall detection is one of the systems used to monitor the daily life of the elderly. Various devices and methods were developed to monitor the condition of the elderly on daily activity. The system that has been proposed in previous studies uses a number of sensors that are placed on the body. However, this system tends to be high cost, complex installation, and inconvenient to use. Therefore, an alternative system is needed to overcome this problem. The purpose of this study is to developed a fall detection method using PoseNet with pose calculations based on key joins. Testing on larger data sets is needed to verify the proposed method's performance further. The use of cameras can be a solution to monitoring the activities of the elderly. With the image processing method, it is possible to estimate the activities of the elderly. The purpose of this study is to developed a fall detection method using PoseNet with pose calculations based on key joins. This study developed a fall detection method using PoseNet with pose calculations based on key joins. The key-join used is Left and Right Shoulder and is only measured at the y-coordinate. We calculated the difference absolute standard deviation value (DASDV) and average amplitude change (AAC) on the Y-coordinate. From 10 falling and non-falling conditions trials, we obtained 85% and 80% accuracy for AAC and DASDV. The result of this research can be used as a source or comparison for future research.

Empirical Mode Decomposition and Hilbert Spectrum for Abnormality Detection in Normal and Abnormal Walking Transitions

“…PoseNet uses deep learning to generate 3D spatial coordinates of joint human skeleton poses. Examples of vision-based HAR applications include the detection of falls in the elderly [22] or the detection of the subject's pose [23].…”