With the development of information technology and Internet-of-Things (IoT), various smart home applications have been proposed to improve the safety and comfortableness of people’s lives. Occupant monitoring is essential to enable these smart home applications, and multiple systems have been explored, including vision-, audio-, radio- and wearable-based methods. However, these methods face certain limitations and requirements in real-world environment such as line-of-sight constraints for vision based methods, the need for silent environment for audio-based methods, sensitive to multi-user movement for radio-based methods, and considerations of comfort and battery life for wearable-based methods. There is a pressing need for a robust and scalable solution for occupant monitoring in complex and diverse real-world deployments.

In this dissertation, we introduce a cyber-physical augmented structural vibration sensing system for occupant monitoring. Compared to other methods, the advantages of structural vibration sensing are non-intrusive (device free), enabling sparse deployment and decreasing privacy concerns. However, these advantages are accompanied by limitations that restrict the system’s robustness and scalability in real-world deployment.

To improve the robustness and scalability, we approach this problem from twoaspects: cyber and physical augmentation to enhance robust data acquisition and scalable information inference. First, we leverage the individual information from wearable sensing to cyber-augment the scalability of information inference. We introduce CMA, a cross-modal signal segment association scheme that associates theidentity information from wearable sensing to enable individual occupant monitoring for structural vibration sensing without label data. Second, we utilize activity information from wearable sensing to cyber-augment the robust (high-quality) data acquisition. We introduce AutoQual, an autonomous sensing quality assessment framework to quantify the impacts of deployment environment on the sensing task performance. The wearable sensing enables the system to automatically select the occupant induced signal for quality assessment without requiring additional human effort. Third, we combine cyber and physical augmentation to enhance scalable information inference and robust data acquisition. We present CPA, a cyber-physical augmentation scheme to enhance the vibration sensing signal via a physical arc structure and achieve high-accuracy, labeling-free event detection. Our experiments verified the efficiency of CMA, AutoQual, and CPA in real-world experiments. Fourth, we explore cyber and physical augmentation to enhance the capability of single-point sensing for occupant tracking. We introduce LEVO, which utilizes the metamaterial filter made of LEGO® bricks to embed the direction information into the waveform. The preliminary experiment shows the feasibility of LEVO for occupant tracking by signal sensor.