Recognizing And Counting Repetitive Motions With Commodity Wi-Fi

Human activity recognition is the core technology that enables a wide variety of applications such as health care,business and military.Traditional approaches use specialized hardware.These devices-based approaches have the fundamental limitations of their inconvenient and human privacy potentially.Human activity recognition with commodity Wi-Fi has many advantages,such as low-cost,non-invasive,ubiquitous and device-free.In this paper,we propose a CSI based multi-person repetitive motions recognition and counting method.The extensive experiments under different indoor environment demonstrates that proposed method performs superiorly in both single person and multi-person with repetitive motions recognition and counting.In summary,the main contributions are as follows:1.Repetitive motions recognition is implemented by multi-model vote strategy.In this paper,factor analysis is used to further compress CSI data and its strengths are elaborated.The dynamic threshold algorithm is leveraged to detect start of motions and many features of repetitive motions are extracted from the time domain and frequency domain.According to accuracies of different machine learning algorithms,the relationship between CSI and the periodic motion types is established by Random Forest,furthermore,we theoretically and experimentally verify the vote strategy can improve prediction accuracy.2.Based on periodic motions recognition,repetitive motions counting is implemented.In this paper,Hankel matrix is used to hold the structural features of CSI correlation.Based on the relationship between the rank of Hankel matrix and the number of person,independent signals are effectively separated from the multi-person signals by CP decomposition technique.The uniqueness of CP decomposition is proved theoretically.The stable signal matching algorithm is applied to find the decomposed signal pairs for each person,and then the signals are fused by averaging.Finally,the peak detection method is adopted to estimate repetitive motion for each person.