Pushing the Boundaries of Wearable Sensing: A Tale of Two Modalities

Description

Wearable sensing—where human activities and behaviors can be perceived by wearable devices—plays a crucial role in human-centric applications. Although significant progress has been made over the past decades, the field of wearable sensing is still in its infancy and far from maturity. This project aims to address the key challenges in wearable sensing by undertaking the following two research tasks. Conventional wearable sensing modalities require an external power supply to function, making continuous monitoring a power-intensive process for wearable devices. To solve this issue, the first task aims to propose a novel wearable context sensing system utilizing a recent breakthrough in photovoltaic technology, namely the transparent solar panel. The proposed system promises energy efficiency through the dual use of photovoltaic energy harvesters for both context sensing and energy harvesting. First, we will propose a theoretical model and simulator to facilitate the performance analysis and design of the proposed sensing system. Then, we will develop a novel algorithm that can achieve accurate and fine-grained context sensing in various scenarios in a data-efficient manner. Earable sensing has emerged as a new sensing paradigm to shape the future of wearable computing. However, earables bring not only tremendous opportunities but also new challenges, such as dynamic head movements, wearing variability, and new activity categories. Therefore, the second task aims to improve the accuracy of earable sensing systems by addressing these key challenges. First, we will propose a source separation method to mitigate the impact of dynamic head movements. Then, we will propose a novel orientation-independent feature extraction method to address wearing variability. Finally, we will design a novel transfer learning scheme that utilizes publicly available dataset to boost the performance of earable sensing. While Task 1 focuses on energy issue and Task 2 aims at accuracy issue, they are interconnected. The technology proposed in Task 1 can be used in future earable devices, which will extend the battery life, and hence facilitate the study in Task 2. In turn, the technical outcomes of Task 2 can inspire the design of the context sensing system proposed in Task 1. Jointly, the two novel sensing modalities push the scientific frontier of wearable sensing forward. This project will not only lay theoretical foundations, discover key principles, and generate new knowledge in the scientific community, but will also enable a wide range of practical sensing applications that can benefit the wider community