Localized local fisher discriminant analysis for indoor positioning in wireless local area network

Subspace learning methods have been used to improve indoor positioning accuracy in wireless local area network (WLAN). However, these methods all suffer from the multimodal signal distributions. Furthermore, the variability of RSS over physical locations presents challenge to learning methods. This paper proposes local fisher discriminant analysis (LFDA) for improved WLAN positioning. LFDA adapts multimodality of signal distributions effectively and extracts more separate location features than previous methods. This is because LFDA further considers preserving the within-class local structure of signal space, thereby more freedom is left for maximizing the between-class separability of physical locations. Moreover, we do not perform monolithic LFDA model over the whole region. Instead, clustering analysis is incorporated to take advantages of spatially localized LFDA and reduce complexity. The proposed method is carried and compared with previous methods in a realistic WLAN indoor environment. Experiments show that the proposed method achieves significant accuracy improvement while reducing computation cost.