约束无迹粒子滤波及其在车辆导航中的应用

Constraints unscented particle filter and its application in vehicle navigation

针对城市高楼密集区,卫星导航信号易被干扰或遮挡、航位推算长期定位误差积累导致车辆组合导航定位精度差的问题,提出一种约束无迹粒子滤波算法。首先,该算法利用无迹卡尔曼滤波对实时状态的均值和方差进行估计,生成的高斯分布作为粒子采样的重要性函数,克服了粒子滤波重要性函数难以选取的问题。其次,采用从观测方程中提取约束条件、构建约束方程的方法,解决约束条件难以构造和新增约束方程导致算法计算量激增的缺陷。再次,通过构造拉格朗日函数,得到无迹粒子滤波的状态估值投影到约束平面的最小值。然后,设计车辆组合导航系统的车辆运动约束方程和道路约束方程,对状态估计值进行约束,修正误差大的估计值,提高了状态量的估计精度。最后,将提出的约束无迹粒子滤波应用到全球定位系统/航位推算车辆组合导航系统中进行仿真验证,并与无迹粒子滤波和自适应无迹粒子滤波进行比较。结果表明:提出的算法估计得到位置误差均值控制在1.5 m左右,而2种比较算法估计得到的位置误差均值控制在3 m左右;提出算法的位置误差估计精度明显优于2种比较算法,车辆组合导航定位性能得到了改善。该方法为驾驶人提供了可靠的反馈信息,避免了交通事故的发生,从而减小了人员伤亡和经济损失。

A constraints unscented particle filter algorithm was proposed to deal with the low accuracy of vehicle integrated navigation, which was caused by interference and blockage of satellite navigation signal or long-term position error accumulation of dead reckoning in urban built-up area. Firstly, unscented Kalman filter was used to estimate the mean and variance of state in real-time. As the importance density function of particle sampling, Gaussian distribution overcomes the hard selection problem of importance function. Secondly, the method of constraint condition and constraint equation, which were constructed from observation equation and adopted to solve the defects of constraint conditions, which was hard to constructed and new constraint equation and leads to the computation surge. Then, Lagrange function was constructed to obtain the minimum value, which was projected to constraint plane by the state estimation. The vehicle kinetics constraint function and road constraint function were designed to constrain the state estimation and modify the big error estimation in order to improve the estimation accuracy of the state. Finally, the proposed algorithm was applied to simulation testing of GPS/DR integrated navigation system. The results show that compared with unscented particle filter and adaptive unscented particle filter, the precision of position error of proposed algorithm is controlled at about 1.5 m, while the precision of position error of other two algorithms are controlled at about 3 m. The estimated accuracy of position error of proposed algorithm performs better than other two algorithms and the positioning performance of integrated navigation is improved. Because of the improvement of the positioning performance of integrated navigation, reliable feedback information is provided and the occurrence of traffic accidents is avoided which in turn reduces casualties and economic losses.