Multi-target joint detection, tracking and classification using generalized labeled Multi-Bernoulli filter with Bayes risk

The objective of multi-target joint detection, tracking and classification (JDTC) is to simultaneously estimate the number of targets, their kinematic states, and decide on the class labels with varying target number and observation uncertainly. Good solutions require solving these problems jointly. Recently, the conditional joint decision and estimation (CJDE) algorithm based on the generalized Bayes risk was proposed to solve problems involving inter-dependent decision and estimation conditioned on data. However, there are difficulties in applying CJDE to multi-target JDTC directly because association and track information is unavailable. In this paper, the CJDE approach is introduced into δ-Generalized Labeled Multi-Bernoulli (δ-GLMB) filter to solve multi-target JDTC problem. The optimal solution is derived based on the new CJDE risk involving classification error and estimation error of multi-target cardinality and states. As an analytic solution to the multi-target Bayes filter, δ-GLMB filter produces track-valued estimates based on association maps, so the costs can be computed by using the estimated state parameters of tracks. In addition, the conditional δ-GLMB prediction and update equations are proposed to estimate the multi-target states given hypotheses and decisions. Our algorithm takes advantage of both δ-GLMB filter and CJDE. The proposed risk can be calculated reasonably and the coupling between decision and estimation is utilized. As illustrated by the simulation results, our method outperforms the traditional JDTC algorithms.