A Multiperson Behavior Feature Generation Model Based on Noise Reduction Using WiFi

Abstract

Benefiting from modeling human behavior and the multipath propagation of wireless signals in the environment, passive behavior-sensing systems can effectively identify individual behaviors by model-based or pattern-based methods. The existing methods use an individual sample to model or profile behavior feature, which leads to their failure in multiperson scenarios due to time-varying environments. In this paper, an online method on modeling the position-behavior feature of multiperson scene is presented. First, a position-based model of individual behavioral feature transformation is proposed. The behavior features in other positions are generated by the behavior of the individual in a particular position. Second, a multiperson behavior feature generation method based on noise reduction is proposed to generate the same behavior feature in an online pattern. Finally, taking advantage of the models, a multiperson fitness coaching system is designed and implemented, named multiuser fitness coach. The system can identify the irregular behavior of individuals in the multiperson environment. The performance of the system is evaluated in different scenarios, and the results show that the precision of feature generation can be effectively applied to the decision of irregular behavior in multiperson scenarios.