基于仿真的滤波技术及其在方位角跟踪中的应用

用贝叶斯方法求解目标跟踪问题，实质就是求状态的后验概率密度。基于蒙特卡洛仿真技术实现的滤波算法PF可通过带权重的采样点来近似后验概率密度，实现对状态变量的期望和方差的求取。当采样点足够多时，PF算法可产生对待估量的近似最优而非次优的估计。将基于无迹变换技术实现的滤波算法嵌入到PF算法中，可避免在观测精度非常高时PF算法的失效，同时还能提高估计精度。引入了方位角跟踪的实例，在对该例的仿真中比较了PF类算法和传统卡尔曼类滤波算法在解算精度和滤波时间耗费等方面的性能差异。

Simulation-based Filter and Its Application to Bearings Tracking

The Bayesian solution to the problems of target tracking is essentially to calculate the exact posterior density. The PF algorithm based on the sequential Monte Carlo method can approximate the posterior density function with a series of weighted sample particles, thus get an estimate of the state, and a measure of the accuracy of the estimate as well. When the number of the particles is sufficiently large, the PF can get an almost optimal estimation of the state. Failures when using generic PF under some circumstances can be avoided if an Unscented-Transformation-based algorithm is embedded in PF. This improved PF algorithm also performs better in getting a higher estimation accuracy. A bearings tracking simulation is discussed here to compare the PF-kind algorithms with those Kalman filter based algorithms.