Amazon
Home Presence Detection and Localization using WiFi CSI
Indoor localization is the foundation block to enable many customer delight applications, such as turning a TV screen on- and off, or turning lights on- and -off, when there is someone in the home or no one is in the home. WiFi Channel State Information (CSI) technology has shown a lot of promise in presence detection and localization and this solution adds almost zero costs to current devices; however, one of the problems with the CSI presence or localization is that it can’t differentiate if the motion has happened near the Access Point (AP) or near the device. Additionally, there is the big challenge of generalization to different RF environments and cases of false positives. This student team will work to detect presence in a home and localize the motion, whether it has happened near the AP or near a device. This student team may work to build ML models with CSI amplitude and phase or use WiFi CSI with another sensor signal such as audio from microphone or Ultra-sonic to calibrate the WiFi CSI model for a particular environment. This student team will also work to collect their own indoor CSI dataset to train and test algorithm on it to check performance and generability of the chosen method. Ultimately, the outcomes this student team will work to achieve are to ensure the true positive rate (TPR) for presence is >90 % and localization error is in +-10 cm. The solution should be generalizable to different indoor home environments.
Faculty Adviser
David Laning,
Affiliate Professor,
Electrical & Computer Engineering
Students
Anusha Prasad
Cheng-hung Hsieh
Gaurav Sharma
Shreyas R K
Yabin Xu