Extracting Context Information from Wi-Fi Captures

Abstract

Inferring a user's current situation is the basis of context-aware services. However, users rarely provide access to their sensor data, and hence extracting context information remains challenging in real-world scenarios. In this paper, we present an overall concept for inferring mobility, location, and role information from users based on passively recorded Wi-Fi signals. Several methods are investigated and an extended Viterbi-based approach is presented to determine dwelling and motion periods. This information is used to enhance the mobility model for probabilistic indoor localization. In addition, we compute various features to classify users according to their role. The presented concept is evaluated on simulated data and discussed on real Wi-Fi captures. Our results show, that the proposed Viterbi-based approach performs best for inferring mobility states and can improve the localization accuracy in most instances. Furthermore, it helps to increase the classification performance and indicates strong cluster tendencies in our real-world dataset.