Abstract

Remote medical diagnostic solutions have recently gained more importance due to global demographic shifts and play a key role in evaluation of health status during epidemic. Contactless estimation of vital signs with image processing techniques is especially important since it allows for obtaining health status without the use of additional sensors. Thermography enables us to reveal additional details, imperceptible in images acquired with standard visible light cameras, yet, low resolution is its significant limitation. In the presented doctoral dissertation, original artificial intelligence solutions were proposed based on performed analysis of innovative thermal image processing methods using Deep Learning techniques for the needs of remote medical diagnostics. Possibility of modifying architecture of deep neural network designed for classification of visible light images in such a way that distribution of extracted features will be recreated enabling detection of facial areas from low resolution thermal data was verified in conducted experiments. Effectiveness of the proposed deep neural network architecture was demonstrated in practical applications, increasing resolution of thermal images and leading to better image quality metrics in comparison to stateof-the-art convolutional models, as well as increasing accuracy of facial areas detection, contactless estimation of respiratory rate and person recognition.

Author (1)