MM-CBMeMBer滤波器跟踪多机动目标

摘要

多模型(Multiple Model, MM)概率假设密度(Probability Hypothesis Density, PHD)滤波器能同时估计机动目标个数及状态,但其序贯蒙特卡罗(Sequential Monte Carlo, SMC)实现运用粒子聚类算法提取目标状态,不仅引入额外计算量,且可能导致目标丢失.针对这一问题,该文提出一种基于多模型的势平衡无偏多目标多伯努利(Multiple Model Cardinality Balanced Multiple target Multi-Bernoulli, MM-CBMeMBer)滤波器,在每次扫描杂波数低于20,检测概率大于0.9的环境中,该方法利用一组伯努利参数近似机动目标状态的后验概率,并通过对伯努利参数的简单运算估计出目标状态,有效地避免了常规聚类算法.仿真结果表明,该方法与多模型概率假设密度滤波器相比,表征估计误差的最优子模型分配距离明显降低.%　　The existing multiple model hypothesis density filter can estimate the number and state of maneuvering targets at the same time. Yet its Sequential Monte Carlo (SMC) implementation involves clustering algorithm, which is unstable and time consuming, and may result in tracking target loss. To solve the problem, this paper proposes a Multiple Model (MM) Cardinality Balanced Multiple target Multi-Bernoulli (CBMeMBer) filter. When the clutter number of per-scan is less than 20 and detection probability is higher than 0.9, this algorithm transmits the posterior density of maneuvering targets through a set of time-varying Bernoulli parameters, according to which, the targets state can be computed by simple operations, thus effectively avoids the clustering algorithm. Simulation results shows that compared with multiple model hypothesis density filter, the algorithm proposed decreased the OSPA distance which chooses to estimate tracking errors.