IDAct: Towards Unobtrusive Recognition of User Presence and Daily Activities

Hanchuan Li, Chieh-yih Wan, Rahul C. Shah, Alanson P. Sample, Shwetak Patel
IDAct recognizes human activities by correlating object state derived from RFID channel parameters with user interactions.


The Internet of Things (IoT) promises to revolutionize the way people interact with their surrounding environment and the objects within it by creating a ubiquitous network of physical devices. However, recent advancements have been focused on creating battery-powered electronics. There remains a huge gap between the collection of smart devices and the massive number of everyday physical objects. In this work, we bridge this gap by enhancing the sensing capabilities of everyday objects using commercial long-range RFID. We apply signal processing and machine learning techniques towards its communication channel parameters to detect the presence of users and to understand their daily activities. Different from prior work, our system can adapt to different environments and objects types. In a naturalistic user study deployed in a home environment, IDAct detected user presence with an F1 score of 96.7% and recognizes 24 different daily activities with an F1 score of 82.8%.