In this paper we focus on the problem of human activity recognition without identification of the individuals in a scene. We consider using Wi-Fi signals to detect certain human mobility behaviors such as stationary, walking, or running. The main objective is to successfully detect these behaviors for the individuals and based on that enable detection of the crowds overall mobility behavior. We propose a method which infers mobility behaviors in two stages: from Wi-Fi signals to trajectories and from trajectories to the mobility behaviors. We evaluate the applicability of the proposed approach using the StudentLife dataset which contains Wi-Fi, GPS, and accelerometer measurements collected from smartphones of 49 students within a three-month period. The experimental results indicate that there is high correlation between stability of Wi-Fi signals and mobility activity. This unique characteristic provides sufficient evidences to extend the proposed idea to mobility analytics of groups of people in the future.
Download