基于递推近似最小一乘的多传感器系统偏差稳健估计算法

对于多传感器多目标跟踪问题,系统偏差对航迹融合精度有较大影响,因此在信息融合系统中,首先要对各传感器的系统偏差进行估计,而在含错误关联和观测野值的复杂环境下,传统系统偏差估计方法的性能会严重退化.对此,提出一种具有递推形式的近似最小一乘稳健估计算法,以减少异常噪声对偏差估计的不利影响.使用平方根平滑逼近函数替代最小一乘法的目标函数,基于牛顿方向及其秩1修正推导出该方法的递推求解框架.基于条件数分析,证明所提出算法的数值稳定性好于Huber方法.通过两个仿真算例,将所提出算法与已有其他算法进行对比验证.仿真结果表明,在含错误关联和观测野值的条件下,所提出算法可以改善偏差估计精度,并且明显好于已有的其他算法.

Multi-sensor bias robust estimation based on recursive approximate least absolute deviation

For the problem of multisensor-multitarget tracking, the sensor bias has great influence on the accuracy of track fusion. Thus, the sensor bias should be estimated at first in the system of information fusion. However the performance of the traditional methods of bias estimation will degrade dramatically in the complex environment that exists the misassociations and observed outliers. Therefore, an algorithm of robust estimation based on the recursive approximate least absolute deviation(RALAD) is proposed, aiming to decrease the adverse impacts of the misassociations and observed outliers. A smooth approximate function in the square-root form is used to replace the least absolute cost function, and the recursive framework is derived based on the Newton method and its rank-one modification. It is verified by the condition number analysis that the proposed algorithm has better numerical stability than the Huber-based method. Performance comparisons between the proposed and existing algorithms are carried out through two simulation examples. The results show that the proposed algorithm is obviously superior to the existing algorithms, and it can improve the estimation accuracy significantly in the case of the misassociations and observed outliers.