RSSI Based Indoor Localization with Reduced Feature Dimension

Authors : Mustafa Eren YILDIRIM

Pages : 106-109

Doi:10.17694/bajece.956866

View : 18 | Download : 11

Publication Date : 2022-01-30

Article Type : Research Paper

Abstract :Wifi based indoor localization gains the interest of researchers for several purposes. Among various techniques, fingerprinting based on Wifi received signal strength indicator insert ignore into journalissuearticles values(RSSI); is a widely used feature in indoor localization because of its simplicity in implementation and minimal hardware requirement conditions. However, the amount of access points insert ignore into journalissuearticles values(AP); at which the RSSI is measured from in the network increases the computational load. This paper presents an alternative approach for dimension reduction in RSSI based indoor localization. We focus on recognizing the building and floor of the test user which is a multi-class problem for both cases. In a multiple class problem, inter-class differences are obtained by Manhattan distance in pair-wise manner. From each pair calculation, top-25 and top-50 features with the largest variances are chosen and merged to generate the final feature set. The proposed algorithm is implemented and evaluated on UJIIndoorLoc dataset. According to the outcomes, our method provides 99.1% accuracy for building and 82.8% accuracy for floor estimation
Keywords : Received signal strength indicator, dimension reduction, indoor localization, UJIIndoorLoc dataset