Human Activity Recognition Using Photoplethysmographic Data

Abstract

Accurately recognizing human activities is a challenging task with numerous potential applications, including fitness tracking and healthcare monitoring. In this study, we used Photoplethysmographic (PPG) sensor data to classify seven different activities performed during fitness training or in a gym setting. The dataset was obtained from two publicly available sources and a combined dataset was generated which consisted of PPG data only. We trained three models on the dataset and achieved an overall performance of 91% in activity classification. The Inception-v3 model slightly outperformed the other two models, which were based on the Inception-ResNet-v2 and ResNet-101 models. Previous work in this area has typically focused on classifying a limited number of activities using PPG data, making our results, which were obtained using two different datasets and real-life settings, particularly encouraging.