Human Motion Detection and Classification Using Ambient WiFi Signals

We investigate the detection and classification of human motions using ambient 802.11 wifi signals. Extensive experiments are conducted and show that different human motions introduce unique patterns in the received signal waveforms. As a starting point of the research, the Received Signal Strength (RSS) is explored as a reliable measurement due to its easy implementation and robustness against frequency nonsynchronization. A two stage algorithm was proposed by first detecting human motion and then classifying different motion types using the random forest algorithm. We studied statistical features not only for the raw RSS values but also for its dominant frequency component and time differencing variant. These features allows us to accurately characterize different human motions and provide sufficient yet concise information for the detection and classification. Experiment results shows that our proposed algorithm gives a high detection rate and a satisfactory classification accuracy in various environments.