未知系统偏差下的预检测变分PHD滤波器OA

In multi-target tracking,zero-mean-shifted Gaussian noise is frequently observed.Measurement offsets relative to true target states,namely system biases,are thereby generated,and inaccurate target state estimates are consequently obtained.To al-leviate the adverse effects exerted by system biases on filter performance,a variational probability hypothesis density filter under unknown system biases is proposed based on variational inference in this paper.To facilitate accurate modeling of system biases,Bernoulli variables are introduced for pre-detection of the presence of system biases at each time step,and bias offsets are modeled as Gaussian distributions.Joint analytical solutions for target states,Bernoulli variables and bias offsets are intractable to obtain for target state estimation.Accordingly,variational Bayesian approximation is adopted to iteratively optimize the parameters of all vari-ables,and a likelihood function is derived to update stage of the probability hypothesis density filter.Simulations results show that superior tracking performance is attained by the proposed algorithm under unknown system bias scenarios in comparison with the Gaussian mixture probability hypothesis density filter and the extended sequential Monte Carlo probability hypothesis density filter.