Source-Observation Weighted Fingerprinting for machine learning based localization

High Resolution Position Information has become increasingly vital to the development of Location Based Services and the expansion of the Internet of Things (IOT). Due to the attenuation of Global Positioning System (GPS) signals in Indoor applications, alternative methods have been proposed to refine location estimates. In search of practical methods, researchers have considered the use of currently deployed 802.11 networks as the basis of positioning, adopting Received Signal Strength Indicator (RSSI) as the standard distance measure. However, attempts at accurate localization have failed due to reliance on heavily distorted power measurements acquired on saturated 2.4 and increasingly crowded 5 GHz channels. In this paper, A Dual Source-Observation Weighted Localization method is proposed as a solution to the Wi-Fi positioning problem, estimating user position through Tikhonov Regularization Cost Functional Minimization. This novel solution combines a) Weighted Kernel Ridge Regression (WKRR), and b) Weighted Radial Basis Function (RBF) Kernels to develop an algorithm which increases estimation accuracy by up to 1/4 meter compared to the current leading localization technology, Weighted K-Nearest Neighbors (WKNN), and substantially reduces error variance, due to the dual Empirical Loss, Complexity objective.