A Machine Learning Approach to Passive Human Motion Detection Using WiFi Measurements From Commodity IoT Devices

Human motion is a primary indicator of indoor occupancy and activity. Motion sensing has paramount importance in the energy management of modern smart buildings and is used for automated controls of lighting and heating, ventilation, and air conditioning (HVAC) equipment. The all-pervasive WiFi infrastructure in urban buildings offers an opportunistic method of human motion detection through passive sensing of WiFi received signal strength indicator (RSSI) and channel state information (CSI). This technique unfolds a plethora of building IoT-related services, in addition to sustainable energy utilization and reduced emission of greenhouse gases. In this article, a device-free human motion detection method through WiFi RSSI and CSI collected from commercial-off-the-shelf (COTS) IoT devices is proposed. Using a laptop, a smartphone, and an ESP32 as receivers, WiFi RSSI and CSI samples were collected from two residential buildings, to constitute six datasets. A 4-D feature vector that exploits the data spread in the time domain is extracted from the collected samples and used to train a two-stage ensemble machine learning model. A comparison of various RSSI-based datasets indicates a mean cross validation accuracy of up to 99.7% and 97.8% in line-of-sight (LoS) and through-the-wall scenarios, respectively. The detection accuracy in non-LoS environments can be enhanced using CSI-based features, enabling motion detection in different rooms using a single WiFi router.