A Faster implementation of Multi-sensor Generalized Labeled Multi-Bernoulli Filter

Abstract

The recent multi-sensor Generalized Labeled Multi-Bernoulli (GLMB) is an efficient analytic implementation to the multi-sensor multi-object state estimation problem. The multi-sensor multi-object posterior is recursively propagated using the multi-sensor multi-object filtering density, by updating it with multi-sensor measurements at each time step. The measurement update step requires solving a series of NP-hard multidimensional assignment problems. In this paper, we introduce a faster implementation of this algorithm by an intuitive approximation, and combine that with the Gibbs sampler based truncation approach to produce an efficient multi-sensor multi-object estimation solution suitable for practical applications.