MOCAP: Device Free Indoor Localization via RFID Tomography

Device-free localization methods allow users to benefit from location-aware services and smart environments without the need to wear a transponder or carry a mobile device continuously. However, conventional radio tomographic imaging approaches that place active wireless sensor nodes around the perimeter of a living space for localization require wired power or continual battery maintenance, limiting usability and deployability. We present a real-time multi-user UHF RFID tomographic localization system that employs a novel signal processing pipeline that uses communication channel parameters such as RSSI, RF Phase, and Read Rate to create tomograms which are processed by our custom-designed convolutional neural network to predict user's locations. Results show that our system is highly accurate, with an average mean error of 17.0 cm for a stationary user and 20.2 cm when walking and moving. We also demonstrate multi-user tracking with an average mean error of 39.4 cm. Overall, the method empowers a minimally intrusive, scalable, and deployable system for locating un-instrumented users in indoor environments.