Localization from Pairwise Distance Relationships using Kernel PCA

In this paper, we present a method for estimating the relative localization for a set of points from information indicative of pairwise distance. While other methods for localization have been proposed, we propose an alternative that is simple to implement and easily extendable for different datasets of varied dimensionality. This method utilizes the Kernel PCA framework [6] (or equivantly the Multidimensional Scaling framework [8]) for producing localization coordinates. To localize, Kernel PCA is performed on a matrix of pairwise similarity values, assuming similarity values are reflective of the data generation distance metric. We test this localization method on 3D points comprising a human upper body and signal strength data from Mote processor modules.