An Optimized Space Partitioning Technique to Support Two-Layer WiFi Fingerprinting

WiFi fingerprinting has been a common solution to indoor localization, however its accuracy still cannot meet the requirements of high accurate indoor location based services. This paper proposes an optimized space partitioning technique followed by a two-layer WiFi fingerprinting based on SVM classification and regression. The proposed space partitioning technique first applies K-means clustering algorithm to partition the localization area to subregions according to signal similarity of the reference locations. An outlier removal algorithm is then applied to remove the outliers according to physical proximity. To obtain the optimal partitioning, an optimization algorithm is proposed and applied onto the space partitioning technique. Evaluations show that the optimized space partitioning technique outperforms space partitioning with floor plan and SVM-C. Comparison with SVR, NN and KNN shows that the two-layer WiFi fingerprinting with space partitioning outperforms one-layer WiFi fingerprinting and is able to improve the localization accuracy effectively.