UltraMic:Privacy-preserving Indoor Activity Tracking and Recognition with Microphone Array
Our work leverages high-frequency sound beyond human hearing (8kHz-48kHz) for privacy-preserving activity recognition and localization using a 49-channel microphone array and deep learning. This system avoids privacy concerns of traditional microphones while enabling accurate identification of activities like cooking or boiling water. By integrating transfer learning, few-shot learning, and user-specific association, we aim to enhance activity recognition, address multi-user scenarios, and expand the scope of detectable activities, paving the way for innovative, privacy-centric applications in health, wellness, and home automation.
The work is still under development, hope to see it soon, but let’s check out the quick demo here
Big thanks to Cameron, Yasha (IN VIDEO) !
05-12-2024