MAIS: Multiple Activity Identification System Using Channel State Information of WiFi Signals

An extensive set of papers have employed channel state information of WiFi signals to perform human activity identification. Given the satisfactory performance, WiFi signals provide a device free, low cost, and non-intrusive alternative to traditional approaches including sensor based and camera based monitoring systems. Unfortunately, most existing papers have focused on the scenario where only a single subject presents. In this paper, we propose a novel human activity identification scheme termed Multiple Activity Identification System (MAIS), targeting at identifying multiple activities of different subjects in the same environment. In designing MAIS, we identify several challenges in identifying activities of multiple subjects and present corresponding solutions, including noise filtering, two-step detection of start/end point of activities, and kNN (k-Nearest Neighbors) algorithm to predict the number of people and the exact activities they are performing. Our experiments show that MAIS achieves an accuracy of 98.04% for anomaly detection, 97.21% for predicting the number of people, and 93.12% for predicting the activities they perform. To the best of our knowledge, this is the first system that achieves high accuracy identifying multiple activities performed by multiple people.