Photoplethysmography and accelerometer sensors signals for recognizing physical activity

Abstract

The utilization of wearable devices to monitor human physiological parameters has been popularized, and due to their low cost, the most common method of monitoring human information in such devices is the use of photoplethysmography (PPG) signals. However, accurate estimation of the PPG signal recorded from the subject's wrist during various physical exercises is often a challenging problem, as the original PPG signal is heavily corrupted by motion artefacts. The article starts with an introduction to how PPG and Accelerometer (ACC) work, and then moves on to the programming, which is then used to provide data processing support for subsequent deep learning by importing data and calculating operations. Long short time memory (LSTM) is built for the paper to recognize activities. The experimental results showed that over 95 % accuracy was achieved in the classification of the test data.