未知量测噪声分布下的多扩展目标CBMeMBer滤波算法

Multi-extended target CBMeMBer filtering algorithm under unknown measured noise distribution

在实际应用场景中,量测噪声协方差准确模型很难被建立,传统的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迅速下降。为了解决量测噪声未知对多扩展目标跟踪结果造成的影响,将变分贝叶斯方法引入到CBMeMBer滤波算法中。VB-GM-CBMeMBer算法能在量测噪声未知情况下通过估计噪声协方差进行滤波计算,但该算法存在目标数目估计不准确的问题。针对此问题,提出一种改进的VB-GM-CBMeMBer算法,该算法在滤波算法预测步骤后引入椭球门限,使用保留在门限内的量测来进行下一步计算,以减少杂波量测,降低杂波量测对扩展目标量测的影响,提高对扩展目标状态聚类的精度。实验结果表明,该算法适用于多扩展目标数目未知、量测噪声协方差未知的情况,且其跟踪精度比GM-CBMeMBer和VB-GM-CBMeMBer滤波算法有一定提高。

In the practical application scenarios,it is difficult to establish an accurate model of measured noise covariance,and the tracking performance of the traditional multi-extended target tracking algorithm decreases sharply when the measured noise covariance is unknown. In order to get rid of the influence of unknown measured noise on the results of multi-extended target tracking,the variable Bayesian method is introduced into CBMeMBer filtering algorithm. The VB-GM-CBMeMBer algorithm can perform filtering calculation by estimating the noise covariance in the case that the measured noise is unknown,but the estimation of target number is inaccurate. Therefore,an improved VB-GM-CBMeMBer algorithm is proposed,which introduces an ellipsoid threshold after the prediction of filtering algorithm,and uses the measurements kept within the threshold to carry out the next calculation,so as to reduce the clutter measurement,lower the impact of clutter measurement on the extended target measurement,and improve the accuracy of clustering in the extended target state. The experimental results show that the proposed algorithm is suitable for the situations that both the number of multi-extended targets and the measured noise covariance are unknown,and its tracking accuracy is improved to a certain extent in comparison with the filtering algorithms of GM-CBMeMBer and VB-GM-CBMeMBer.