Stochastic Sampling-Based Data Association

Abstract

This paper considers how to determine the origin of a single measurement originating from one of a group of objects moving in close proximity. During the time in which measurements are being received, the dynamics of the various objects are the same except for initial conditions. We present a method that uses techniques from filtering theory to represent a distribution using a finite number of parameters. This method, which we call stochastic sampling based data association (SSBDA), is similar to a particle filter but differs in that we use a modified probabilistic data association filter (PDAF) in the propagation of the distribution associated with the object's location. Using the PDAF it is possible to see the effect that the addition of each measurement has on the covariance of the posterior distribution. We discuss how the covariance of the posterior can be used for making decisions on whether or not a particular measurement originated from a predetermined object of interest.