WiFi-based Indoor Localization Using Clustering and Fusion Fingerprint

Due to the free deployment of the additional network infrastructure, WiFi-based indoor localization has drawn researchers' attention in recent years. However, the accuracy and robustness of WiFi-based indoor localization systems are severely undermined by the fluctuation of received signal strength (RSS). To mitigate the problem, in this paper, we propose a WiFi localization framework via fingerprint clustering and adaptive k-nearest-neighbors (KNN) based on fusion fingerprint. First, in the offline phase, we cluster the offline fingerprints via Gaussian mixture model (GMM) to divide the localization area into several subareas. Then a random forest-based subarea classifier is trained by the offline data and corresponding subarea labels. In the online phase, the subarea of the query fingerprint is firstly predicted by the trained RF-based classifier. Finally, a fusion fingerprint-based adaptive KNN algorithm is utilized to estimate the location in the predicted subarea. In the experiment conducted, the localization performance of the proposed method is evaluated and compared with other representative methods. The results obtained demonstrate that the proposed localization framework significantly reduces the localization error without any hardware calibration.