Multiple Classifiers Global Dynamic Fusion Location System based on WiFi and Geomagnetism

Abstract

The existing WiFi and geomagnetism based positioning methods using single classifier show low accuracy because they are sensitive to changing environments. In this paper, we propose a global dynamic fusion location algorithm for multiple classifiers based on WiFi and geomagnetic fingerprints. In the offline phase, we first divide a positioning environment into some grid points and construct RSS and geomagnetic fingerprints for each grid point. Then, we train multiple classifiers by using the constructed fingerprints. Second, we derive a global dynamic fusion weight training method for each grid point through the global supervised optimization learning. In the online phase, given an RSS testing sample, we select the matching weights for fusion by using K-nearest neighbor (KNN). Our proposed multiple classifiers global dynamic fusion algorithm can make full use of the intrinsic complementarity of multiple classifiers, thus effectively improving the positioning accuracy of RSS and geomagnetic fingerprints. Experimental results show that the proposed algorithm outperforms some existing methods in complex indoor environments.