【摘要】 传统的基于高斯过程的扩展目标跟踪(Gaussian Process Extended Target Tracking, GP-ETT)算法通常将高斯过程的输入视为精确值，输入不确定性不可忽略时，引起跟踪性能的降低。针对这一问题，定量推导了GP-ETT算法中GP输入噪声的前二阶矩，提出基于泰勒级数与数值近似的3种带输入噪声的GP-ETT算法，并求得理想环境下带输入噪声的GP-ETT最优理论性能界。仿真结果表明，改进算法的马氏距离更小，得到的性能界更合理。更多还原

【Abstract】 In the traditional Extended target tracking(ETT) based on Gaussian Process, the GP input is assumed to be known accurately, that will degrade the tracking performance when the input uncertainty occurs. In order to address this problem, the first two moments of the GP input in the GP-ETT is derived in this paper. Then, based on the Taylor series expansion and the numerical approximation, three modified GP-ETT methods are proposed. Finally, the optimal theorical performance bound of GP-ETT with input noise is derived under the ideal conditions. Simulation results show that the proposed GP-ETT method achieves a smaller Mahalanobis distance and the derived performance bound is more reasonable.更多还原