Estimating detection statistics within a Bayes-closed multi-object filter
Javier Correa, Martin David Adams · International Conference on Information Fusion · 2016
In multi-target tracking, correct models of detection statistics, namely the probability of detection and clutter rate, are required for effective multi-target state estimation. Within a multi-target filter, the detection statistics are usually assumed as known and static. Estimating the detection statistics' parameters before the execution of the filtering algorithms is not always feasible and in some scenarios, these parameters could be time varying, which would invalidate offline estimation. To overcome these issues, this paper presents a Random Finite Set (RFS) based algorithm which is capable of estimating both the probability of detection and the clutter rate, while jointly estimating the multi-target state of the system. The proposed algorithm is based on previous work, the Kronecker Delta Mixture and Poisson (KDMP) filter, which is a Chapman-Kolmogorov and Bayes closed solution to the RFS-based filtering problem. Importantly, the resulting robust filter remains closed under the filtering procedure. Results show that the algorithm converges to the correct detection statistics in simulated environments and, as opposed to other methods, it can even continue to estimate the probability of detection when no targets are present in the environment.