基于区间箱粒子多伯努利滤波器的传感器控制策略

多目标跟踪中的传感器控制本质上是一个最优非线性控制问题, 其在理论分析和计算上极具挑战性. 本文基于区间不确定性推理, 利用箱粒子多伯努利滤波器提出了一种基于信息测度的传感器控制策略. 首先, 本文利用箱粒子实现多伯努利滤波器, 并通过一组带有权值的箱粒子来表征多目标后验概率密度函数. 其次, 利用箱粒子的高斯分布假设, 将多伯努利密度近似为高斯混合. 随后, 选择柯西施瓦兹(Cauchy-Schwarz, CS) 散度作为评价函数, 并详细推导了两个高斯混合之间的CS散度的求解公式, 以此为基础提出相应的传感器控制策略. 此外, 作为一种对比方案, 利用蒙特卡罗方法, 本文还给出了通过对箱粒子进行混合均匀采样, 进而通过点粒子求解CS散度的递推公式, 并提出了相应的控制策略. 最后, 仿真实验验证了所提算法的有效性.

Sensor Control Based on Interval Box-particle Multi-Bernoulli Filter

In multi-target tracking, sensor control is essentially an optimal nonlinear control problem. And it is also challenging in theoretical analysis and calculation. On the basis of interval uncertainty reasoning, this paper proposes an information measure based sensor control via box-particle multi-Bernoulli fllter. First, the box-particle multi-Bernoulli fllter is given and the posterior multi-Bernoulli density is approximated by a set of box particles with weights. Then, by constructing a box particle as a Gaussian distribution, the multi-Bernoulli density is approximated by mixed Gaussian components. Subsequently, this paper chooses the Cauchy-Schwarz (CS) divergence as the evaluation function, and deduces the CS divergence in detail between two Gaussian mixed multi-Bernoulli densities. The corresponding sensor control strategy is also proposed. Furthermore, as a compared scheme, this paper also gives a recursive formula for solving CS divergence by sampling particles from a box in the mixed uniform way using Monte Carlo method and presents the corresponding control strategy. Finally, simulation results verify the efiectiveness of the proposed algorithm.