Particle Filtering for Positioning and Tracking Applications

Abstract: A Bayesian approach to positioning and tracking applicationsnaturally leads to a recursive estimation formulation.The recently invented particle filter provides a numericalsolution to the non-tractable recursive Bayesian estimation problem.As an alternative, traditional methods such as the extended Kalmanfilter, which is based on a linearized model and an assumption onGaussian noise, yield approximate solutions.In many practical applications, signal quantization and algorithmiccomplexity are fundamental issues. For measurement quantization,estimation performance is analyzed in detail. The algorithmiccomplexity is addressed for the marginalized particle filter, wherethe Kalman filter solves a linear subsystem subject to Gaussian noise efficiently.The particle filter isadopted to several positioning and tracking applications and comparedto traditional approaches.Particularly, the use of external database information to enhanceestimation performance is discussed.In parallel, fundamental limits are derivedanalytically or numerically using the Cram'{e}r-Rao lower bound, andthe result from estimation studies is compared to the correspondinglower bound. A frameworkfor map-aided positioning at sea is developed, featuring an underwaterpositioning system using depth information and readings from a sonarsensor and a novel surface navigation system using radar measurementsand sea chart information. Bayesian estimation techniques are alsoused to improve position accuracy for an industrial robot. Thebearings-only tracking problem is addressed using Bayesian techniquesand map information is used to improve the estimation performance. Formultiple-target tracking problems data association is an importantissue. A method to incorporate classical association methods when theestimation is based on the particle filter is presented. A real-timeimplementation of the particle filter as well as hypothesis testing isintroduced for a collision avoidance application.