Efficient Performance Prediction of the Bernoulli Filter for Optimal Detection and Tracking

In this paper, we propose a tracker-aware method for radar detection-threshold optimization in single-target tracking based on the Bernoulli filter and the generalized optimal sub-pattern assignment (GOSPA) metric. We use the predicted GOSPA metric as the optimization objective and propose two efficient methods for predicting it. The first method employs measurement normalization and gating techniques to achieve an efficient integral computation. The second method precomputes integrals offline and leverages them online to approximate the predicted GOSPA metric efficiently. The optimal detection and tracking method using the Bernoulli filter is then proposed, where the detection threshold is optimized in a tracker-aware manner to minimize the predicted GOSPA metric. Simulation results show that the proposed methods achieve significant improvements in computational efficiency while accurately predicting the actual GOSPA metric. Furthermore, the Bernoulli filter with optimal detection and tracking demonstrates superior performance compared with the fixed-threshold approach.