IoB-Enhanced Deep Learning for Non-specific Thermal Respiratory Rate Monitoring

The growing trend of population aging has resulted in an increasing demand for long-term care and medical assistance. To address this, non-contact physiological measurements have made significant advancements by harnessing the potential of the IoB in emerging technologies. While thermal imaging techniques have been explored for respiratory rate measurement, limited research focuses on scenarios where facial obstructions, like glasses or masks, are present. In this study, we propose an IoB-enhanced method that utilizes low-resolution continuous thermal images and advanced image processing techniques. By accurately segmenting the foreground and transforming the data into frequency domain information after temporal domain denoising, we obtain valuable insights. A deep neural network, trained with the fusion of IoB insights and data from a piezoelectric respiratory belt, intelligently selects the most suitable region of interest to measure respiratory rate accurately. Our research prioritizes optimizing computational efficiency to ensure the method's feasibility within low-power embedded systems, achieving real-time and accurate measurements with a power consumption of only 10 watts. By integrating IoB principles into our method, we enhance connectivity, data exchange, and behavioral insights, resulting in a more effective respiratory rate measurement system. Experimental results demonstrate the superior accuracy of our IoB-enhanced method, even in scenarios where subjects wear glasses and masks while maintaining freedom of head movement. Although our model training requires a significant amount of data, our approach substantially reduces limitations imposed by facial obstructions in thermal imaging-based respiratory rate measurement. This breakthrough sets the stage for the future adoption and advancement of IoB-driven respiratory rate monitoring.