An efficient approximate implementation for labeled random finite set filtering

Abstract In this paper, we propose an approximate implementation for labeled random finite set filtering with low computational cost. In contrast to the earlier implementations of the generalized labeled multi-Bernoulli (GLMB) filter and the labeled multi-Bernoulli (LMB) filter, the proposed approach adopts a sampler based on marginaling the joint association probability that can generate the components of the GLMB density from marginal probabilities with high parallel level. Additionally, this marginal probability approximation results in an efficient implementation of the LMB filter, which calculates the parameter set of the filtering LMB density from the marginal probabilities. Furthermore, we exploit a simplified belief propagation algorithm to obtain approximate marginal probabilities with linear complexity in the number of measurements and objects. Simulations show that the proposed implementation can handle both linear and non-linear scenarios and is more efficient against the existing implementations.