UCLA

In recent years, sensing and perception techniques have evolved to be heavily reliant on learning-based pipelines. There is a specific need to explore computational imaging (joint design of hardware and software) in the era of AI. This work bridges this gap by understanding what we term as “neural sensing” through three pillars: the sensor, the data, and the learning algorithm. In the context of contactless heart rate monitoring of humans using visual sensors and beyond, we show that each of these three pillars pose specific, critical problems with the current state of the art: equity across demographic groups, lack of scalable, diverse data, and low signal to noise ratio in sensor measurements inhibiting accurate vital sign monitoring. We explore each pillar with the aim of addressing these limitations and demonstrate how a fundamental understanding and treatment of each of this pillars is critical towards building an operational perception systems. Through this thesis, we make contributions towards understanding the various pillars of neural sensing for and beyond contactless heart rate sensing, while also advancing the state of the art in remote plethysmography.