Exploring Human Activities Using eSense Earable Device

Abstract

Detecting head- and mouth-related human activities of elderly people are very important for nurse care centers. They need to track different types of activities of elderly people like swallowing, eating, etc., to measure the health status of elderly people. In this regard, earable devices open up interesting possibilities for monitoring personal-scale behavioral activities. Here, we introduce activity recognition based on an earable device called ‘eSense’. It has multiple sensors that can be used for human activity recognition. ‘eSense’ has a 6-axis inertial measurement unit with a microphone and Bluetooth. In this paper, we propose an activity recognition framework using eSense device. We collect accelerometer and gyroscope sensor data from eSense device to detect head- and mouth-related activities along with other normal human activities. We evaluated the classification performance of the classifier using both accelerometer and gyroscope data. For this work, we develop a smartphone application for data collection from the eSense. Several statistical features are exploited to recognize head- and mouth-related activities (e.g., head nodding, head shaking, eating, and speaking), and regular activities (e.g., stay, walk, and speaking while walking). We explored different types of machine learning approaches like Convolutional Neural Network (CNN), Random Forest (RnF), K-Nearest Neighbor (KNN), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), etc., for classifying activities. We have achieved satisfactory results. Our results show that using both accelerometer and gyroscope sensors can improve performance. We achieve accuracy of 80.45% by LDA, 93.34% by SVM, 91.92% by RnF, 91.64% by KNN, and 93.76% by CNN while we exploit both accelerometer and gyroscope sensor data together. The results demonstrate the prospect of eSense device for detecting human activities in various healthcare monitoring system.