Analysis of Frequency Spectral Features of Coughing Activity from Tri-Axial Accelerometer sensor at Multiple Body Points

Abstract

Human activity recognition using sensors has wider applications such as daily activity and health monitoring, robotics, security purpose, monitoring human beings in the workplace, and others. Activities such as sitting, standing, walking, walking upstairs, and walking downstairs are commonly classified. Cough event detection and counting have always been the most important research topic in the medical field. We aim to study the non-cough and cough activity in human beings at five body positions of varying heights subjects. Previous studies have shown that cough during walking can be accurately detected with 92, 73, 62, and 82% accuracy at the chest, stomach, shirt pocket, and upper hand respectively from raw acceleration signals in the time domain. We analyzed the frequency domain characteristics, the Spectral Maximum (SM), and Spectral Summation (SS) at four frequency bands in the 0–20 Hz range for the accelerometer axis: x, y, and z. Our study reveals a 24 – 142 % increase along the Y-axis and a 14 – 146 % increase along the Z-axis in SS of cough signal compared to a non-cough signal at the five-position considered in our study. Evaluation of the 3D plot of spectral features shows the clear difference of a cough signal from a non-cough.