Usable and Ubiquitous Privacy-Aware Sensing Devices

Hybrid Event: 3725 BBB / Zoom

Abstract: The proliferation of smart devices is bringing us closer to a future where everyday objects can monitor users, anticipate their needs, and track health metrics. However, privacy concerns have significantly hindered the adoption of information-rich sensors, such as microphones and cameras, particularly in sensitive home environments like bedrooms and bathrooms—locations where critical self-care behaviors and health events, such as falls, are most likely to occur. To address these concerns, ubiquitous sensing technologies must be designed with principles from the usable privacy community, ensuring privacy guarantees that promote adoption and maximize real-world impact. This dissertation presents a Privacy by Design approach to sensor-level privacy across three key domains. First, I introduce privacy-preserving microphones that do not capture speech frequencies but instead leverage inaudible ultrasound to outperform traditional microphones in acoustic event recognition. These microphones can also transform captured signals into locality sensitive hashes, ensuring that privacy-invasive raw audio cannot be reconstructed. In real-world deployments, these privacy-aware microphones demonstrate on-device detection of urinary voiding—an important kidney health metric—using only inaudible frequencies. Second, I explore privacy-preserving cameras that utilize thermal imaging to sanitize personally identifiable information on-device. This approach enables critical machine learning and computer vision applications, such as fall detection, without compromising performance. Finally, I present novel privacy-aware sensing techniques that inherently prevent the capture of sensitive information, such as airborne sound, by leveraging alternative signal sources like surface-acoustic waves. These methods enable activity recognition in the home without invasive data collection. Together, these contributions demonstrate how privacy-aware sensing can bridge the gap between user privacy and real-world sensing applications, paving the way for safer, more adoptable smart home technologies.