Online decision analytics with deep learning : non-invasive fever screening

Abstract

Research in artificial intelligence (AI) has focused on performance advancement while sacrificing consideration for real-time applications. Consequently, online decision-analytic systems utilizing multiple AI computer vision algorithms to replace human decisions receive less attention. This dissertation reports the design, implementation, and study of a live-streaming AI fever screening system (LAFSS) to replace human decisions. The LAFSS addresses four major challenges of designing an online-decision analytic system on fever screening. The first challenge is the difficulty in long-distance non-invasive temperature screening. In current practice, temperature screening with Infrared Thermography (IRT) is limited to a narrow distance range for febrile detection to bypass the inaccuracies due to the distance. Results show that our novel proposed model can compensate for the loss due to the effects of distance and extends the temperature screening distance in a controlled thermal environment. The second challenge is the influence of ambient temperature on long-distance temperature screening. Data shows that our system can compensate for the effects of distance and ambient temperature for semi-outdoor temperature screening. This system is also the first of its kind. Moreover, we study the possibility of noise suppression in non-invasive temperature measurement with human tracking. Our study shows that the temporal information by human tracking suppresses the noise effectively. Last but not least, the fourth challenge is the design and implementation of a large-scale real-time fever screening system with multiple AI. Our system can detect febrile people in a moving crowd and track them across multiple cameras in real-time. LAFSS has been designed, implemented, and deployed in multiple cross-border checkpoints, libraries, schools, and elderly centers. Finally, lessons learned are discussed to facilitate more real-time implementation of AI algorithms, especially on non-invasive temperature screening applications.

Date of Award	2021
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Hau Yue Richard SO (Supervisor)