Principal Component Analysis and Clustering Based Indoor Localizaion

This paper proposes an improved method which applies principal components analysis (PCA) algorithm to an existing fingerprinting localization method based on iterative K-means, grid scoring (KS) and AP scoring (AS). In the off-line phase, the suggested method evaluates the localization capability of every access point (AP) for the first step, and then generates only a few new principal components from APs. To obtain balanced components, component rotation is needed. Finally, the balanced principal components (BPC) can be used as AP for following KS algorithm in on-line phase. Compared with the former one, the suggested method has an outstanding performance in large monitored area with large amount of APs, for it greatly reduces the computational quantity by reducing the dimensions of radio map by a large margin.