Intelligent PPG-based Heart Rate Signal Analysis for Car Drivers Monitoring

• Autor(a) principal: Baiense, João Pedro
• Data de Publicação: 2024
• Outros Autores: Eerdekens, Anniek, Schampheleer, Jorn, Deruyck, Margot, Pires, Ivan Miguel, Velez, Fernando José
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.6/14469
• topic: Artificial Intelligence; Convolutional Neural Networks; Data cleaning; Data processing; Driver monitoring; Photoplethysmogram; Heart rate signal processing; Wearable devices

This research aims to contribute to enhancing road safety through the development and exploration of an intelligent wristbandbased health monitoring solution for car drivers. It focuses on using various sensors, such as the photoplethysmogram (PPG) and an accelerometer, to accurately estimate the drivers’ heart rate. The primary goal was to create a robust and accurate model capable of real-time heart rate estimation from PPG signals, with the potential to improve the effectiveness of Internet of Medical Things (IoMT) applications in the healthcare sector. The study delves into the multiple processing steps involved in improving the quality of data to make it suitable for efficient processing by the deep learning model, encompassing data analysis, signal interpretation, and applying diverse techniques such as filters, data shifting, and data manipulation. The research integrated the leave-one-session-out (LOSO) cross-validation technique for model training and evaluation alongside fine-tuning hyperparameters to optimize model performance and efficiency. The achieved Mean Absolute Error (MAE) of 3.450 ± 1.324 bpm and Mean Squared Error (MSE) of 69.50 ± 93.57 bpm2 represent notable outcomes, resulting in a 54.9% improvement in MAE from the original study. Additionally, the research integrated the model into a user-friendly mobile application, visually presenting the results and enabling users to examine their health status in real-time. These findings highlight the significance of eticulous data analysis and processing in wearable device applications and the high accuracy of the proposed model.