基于变分贝叶斯势均衡多目标多伯努利滤波的 多扩展目标跟踪算法

Multiple extended-target tracking based on variational Bayesian cardinality-balanced multi-target multi-Bernoulli

中文关键词  多扩展目标  变分贝叶斯  势均衡多目标多伯努利滤波  随机有限集  目标跟踪

英文关键词  multiple extended target  variational Bayesian  cardinality-balanced multi-target multi-Bernoulli filter  random finite set  target tracking

中文摘要

      由于传统多扩展目标跟踪算法在量测噪声协方差未知情况下跟踪性能急剧下降, 本文提出一种基于变分贝叶斯的势均衡多目标多伯努利滤波(VB--CBMeMBer)跟踪算法, 并给出了高斯混合实现. 该算法在未知量测噪声协方差的情况下, 将量测看成随机分布在扩展目标上的量测产生点所产生, 利用变分贝叶斯方法近似地求出各量测产生点状态和量测噪声协方差的联合概率密度, 并给出其递归形式以估计量测产生点, 继而将得到的量测产生点状态进行聚类得到扩展目标的状态. 仿真实验表明, 所提算法可以自适应地跟踪未知数目、未知量测噪声协方差的多扩展目标. 其跟踪精度与传统的CBMeMBer跟踪算法相比, 有明显提高.

英文摘要

      Because the performance of the conventional extended target-tracking declines greatly under the circumstance of unknown measurement noise covariance, we propose a new multiple extended target-tracking algorithm based on the variational Bayesian cardinality-balanced multi-target multi-Bernoulli (VB--CBMeMBer), and give its Gaussian mixture implementation. With unknown measurement noise covariance, the measurements of this algorithm are assumed to be produced by the measurement producers randomly distributing on the extended target. Then, the variational Bayesian (VB) approximation technique is applied to approximate joint probability density of the states of measurement producers, and the unknown measurement noise covariance. Their recursion forms are derived and are used to track measurement producers. Next, clustering algorithms are applied to the states of the tracked measurement producers to determine the states of the extended target. Simulation results show that the proposed algorithm can adaptively track unknown numbers of multiple extended targets with unknown measurement noise covariance. In addition, it has an improved precision when compared with conventional CBMeMBer algorithms.