A Particle Bernoulli Filter Based on Gaussian Process Learning for Maneuvering Target Tracking

In this work, the Bayes-optimal Bernoulli filter (BF) is studied for the target tracking where the target is randomly present or absent in the view field of the sensor while the sensor may provide imperfect measurement which contains miss detection and false alarm. To solve the issue that the dynamic model of the target is switching in an unknown mode, we employ the Gaussian process (GP) regression tool, which is a data-driven approach for learning the motion model online, to approximate the transitional density in the formulation of the BF. To deal with the nonlinear measurement model, the proposed GP-based BF is implemented using particles. In the simulation experiment, the proposed approach is performed on a maneuvering target tracking scenario and compared with the Bernoulli particle filters utilizing the full or partial model changing information.