基于单声矢量传声器虚拟扩展的多机动声目标跟踪算法

为解决单声矢量传声器(Acoustic vector sensor, AVS)可跟踪声目标数目少、跟踪性能差的问题, 提出了基于AVS虚拟扩展的多机动声目标跟踪算法. 首先, 引入高阶累积量预处理过程并建立高阶似然函数, 不仅能够抑制高斯噪声、提高估计精度, 还可通过AVS的虚拟扩展增加可跟踪目标数目. 然后, 在边缘化$\delta$广义标签多伯努利(Marginalized $\delta$-generalized label multi-bernoulli, M$\delta$-GLMB)滤波框架下, 提出了基于累积量的增广运动模型状态的M$\delta$-GLMB (Cumulants-based augumented motion model state M$\delta$-GLMB, Cum-AMMS-GLMB)算法. 算法引入多种运动模型, 并将表征不同模型的索引标号作为目标状态的增广参数, 通过各模型间的加权混合获取优于单一运动模型的跟踪性能. 除此之外, 算法的序贯蒙特卡洛(Sequential Monte Carlo, SMC)实现过程中, 依据高阶预处理获得的归一化空间谱拟合检测概率函数, 抑制了杂波向可用粒子扩展, 进一步增强了高似然区域的粒子. 最后, 推导了AVS目标跟踪的后验克拉美罗下界(Posterior cram$\acute{e}$r-rao lower bound, PCRLB), 并通过仿真实验验证了算法的量测噪声抑制能力和声目标跟踪性能.

Multi-maneuvering Acoustic Targets Tracking Algorithm Based on Virtual Extension of Single Acoustic Vector Sensor

To solve the problem of poor performance and fewer trackable targets in the multi-targets tracking of acoustic vector sensor (AVS) in multi-target tracking, a multi-maneuvering acoustic targets tracking algorithm based on virtual extension of single AVS is proposed. First, the higher-order cumulants processing method is introduced to establish a higher-order likelihood function, which can not only improve the estimation accuracy by suppressing the Gaussian noise, but also increase the number of estimable targets by virtually extending the AVS. Then, under the marginalized $\delta$-generalized label multi-bernoulli (M$\delta$-GLMB) framework, a cumulants-based augmented motion model state M$\delta$-GLMB (Cum-AMMS-GLMB) algorithm is proposed. The algorithm introduces multiple models, and uses the model index labels that distinguish different motion models as an augmented parameter for the target state, and obtains a better tracking performance than a single motion model through weighted mixing of the updated states of each model. In addition, in the sequential Monte Carlo (SMC) implementation of the algorithm, the detection probability function is fitted based on the normalized spatial spectrum obtained by higher-order cumulants preprocessing can suppress the diffusion of clutter to the available particles, and further enhance the particles in the high-likelihood region. Finally, the posterior cram$\acute{e}$r-rao lower bound (PCRLB) for targets tracking of single AVS is derived, and the performance of measurement noise suppression and acoustic targets tracking is verified by simulation experiments.