A box particle filter for stochastic and set-theoretic measurements with association uncertainty

This work develops a novel estimation approach for nonlinear dynamic stochastic systems by combining the sequential Monte Carlo method with interval analysis. Unlike the common pointwise measurements, the proposed solution is for problems with interval measurements with association uncertainty. The optimal theoretical solution can be formulated in the framework of random set theory as the Bernoulli filter for interval measurements. The straightforward particle filter implementation of the Bernoulli filter typically requires a huge number of particles since the posterior probability density function occupies a significant portion of the state space. In order to reduce the number of particles, without necessarily sacrificing estimation accuracy, the paper investigates an implementation based on box particles. A box particle occupies a small and controllable rectangular region of non-zero volume in the target state space. The numerical results demonstrate that the filter performs remarkably well: both target state and target presence are estimated reliably using a very small number of box particles.