Implementing a Robust Method for Detecting Human Actions in Health Monitoring by Employing Sensors on Mobile Internet of Things Devices and Utilizing a One-Dimensional Convolutional Neural Network

This research is focused on developing a precise human action recognition system by utilizing accelerometer and gyroscope sensor data. The aim is to analyze various human actions like jogging, sitting, standing, walking, going upstairs, going downstairs, and falling, using multiple sensors found in modern smartphones. Previous studies have explored human action recognition using different sensors, cameras, and machine learning and deep learning techniques. However, some existing approaches, particularly those based on channel state information (CSI), have shown less satisfactory outcomes. In contrast, this paper employs deep learning techniques, specifically one-dimensional convolutional neural networks (1D-CNN), for more accurate human action analysis. One significant advantage of the proposed system is its potential as a wearable device. The paper will introduce a one-dimensional convolutional neural network tailored for human action recognition and an optimized approach for determining the magnitude of accelerometer's X, Y, and Z coordinates, as well as gyroscope sensor values. By utilizing these coordinates, the One-Dimensional Convolutional Neural Networks can effectively classify various human actions like jogging, sitting, standing, walking, going upstairs, going downstairs, and falling. To evaluate the performance of the 1D-CNN approach, a comparison is made with other methods, including SOM, SVM, and FCM. The results demonstrate that the proposed method (1D-CNN) outperforms other classification techniques with an impressive accuracy of 92.42 per cent.