【作者】 杨秀华；

【导师】 陈涛；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 机械制造及其自动化， 2004， 博士

【摘要】 预测目标的运动参数实现复合控制是提高快速捕获和精密跟踪精度的重要手段。复合控制可以提高系统的无静差度,消除速度和加速度滞后误差,使跟踪精度提高几倍至十几倍。同时可以降低由于其它扰动引起的误差。 实现复合控制必须能够获得目标运动的角速度、角加速度信号,通过预测滤波技术可以得到空中目标运动的位置及其导数,因此,预测目标的运动参数对于提高精密跟踪精度具有十分重要的意义。 通常,目标的运动状态方程在直角坐标下描述,而传感器的测量值是在极坐标下取得的,因此,光电目标运动参数的滤波属于非线性滤波。本文研究高精度光电目标运动参数的非线性滤波。 首先,根据噪声的统计特性,建立了分别基于白噪声和有色噪声的直角坐标下的光电跟踪目标状态模型和分别基于极坐标系和直角坐标系的测量模型。 针对推广Kalman滤波算法精度低,易发散的缺点,认为线性化误差是影响跟踪精度的因素之一,为此,本文提出了实际测量噪声协方差的在线估计算法,该算法简单实用,计算量小。从而很好地解决了在实际应用中,测量噪声的时变协方差矩阵无法验前已知这一难题。同时在研究中发现目标的距离与线性化误差有关,而目标的角位置的非线性与线性化误差关系较小。 提出了改进的推广Kalman滤波算法:①采用混合坐标系,在直角坐标系用相对简单的状态方程描述目标的运动特性,目标轨迹外推,协方差矩阵的传播,增益的计算均在直角坐标下进行,而目标新息(残差)计算在极坐标下进行;②增加了线性化误差补偿。将距离的非线性通过附加噪声加以补偿,距离测量误差的方差每一步被重新估计。该算法有效的减小了极坐标测量值非线性的影响,跟踪性能优于推广Kalman滤波器。 为减少改进的推广Kalman滤波器的计算量,提出了贯序滤波方法(SMEKF)。采用贯序滤波算法,可以避免直接计算矩阵的逆,因此可以大大减少计算量。同时在研究中发现如果按照俯仰、方位、距离的顺序进行测量更新,可以提高估计精度。仿真研究表明,采用贯序滤波的改进的推广Kalman滤波算法不仅可以减小计算量,而且跟踪精度也大大提高。 提出了一种针对光电机动目标的修正机动加速度方差自适应算法。状态模型更多还原

【Abstract】 Combined control is an important method for acquiring fast moving targets and improving tracking precision by prediction filtering parameters of target. The no difference grade of control system is increased by it. The lag errors of velocity and acceleration can also be removed. Combined control provides greatly superior performance than conventional control system. Furthermore, it can reduce some disturbances.Combined control must obtain the signals of target’s velocity and acceleration by prediction filtering technology. So target’s parameter estimation is important for tracking system.Usually a Cartesian coordinate frame is well suited to describe the target dynamics, in this description, the dynamics equations are often linear and uncoupled .Because measurements of the target location are expressed as nonlinear equations in Cartesian coordinates, the tracking problem is connected with nonlinear estimation. In this dissertation, applications of nonlinear filtering algorithms to target tracking are studied.At first, linear dynamic models are proposed based on white noise and colour noise. Observation models are built in Cartesian coordinates and Polar coordinates respectively.Extended kalman filter (EKF) is simple and practical. But the EKF has the disadvantage of model linearization error. This filer is always divergence. The author analyzed the relationship between tracking accuracy and linearization error and presents an algorithm that can estimate the statistic properties of measurements noise, it also compensates for the errors of model linearization to overcome this disadvantage. It was founded nonlinearities of the range measurements influence on the tracking accuracy. But nonlinearities of two angle measurements do not appear significant.A new filtering algorithm for improving tracking performance is developed in mixed coordinate system which includes correct evaluation of the measurement error covariance. The target dynamics in inertial rectangular coordinate is modeled. Theestimation of target’s trajectory and calculation of gain are performed in it. While residual is calculated in polar coordinates. The compensation of linearization error is included in MEKF. The range measurement nonlinearity is treated as additional measurement error. The measurement error variance for the range is calculated adaptively at each time. The algorithm reduces nonlinearity effect. So the performance of this filter is better than EKF.In order to reduce computational complexity, it is preferable to sequentially process scalar component of the measurement vector instead of processing it as a single data. The advantage of sequential processing is that instead of requiring the matrix inversion, which leads to considerable computational savings. Furthermore, in the case of nonlinear measurements, sequential processing of the measurement vector in particular order may produce additional benefit of improving estimation accuracy. .Simulation results show that the proposed method offers superior performance.An improved adaptive filtering algorithm based on current statistical model is presented by using the relationship between maneuvering acceleration and its covariance. The acceleration of a maneuvering target is considered as a time-correlation random process with non-zero mean values. The prediction of acceleration is as mean value of maneuvering acceleration and its covariance varies adaptively. The simulation results show that adaptive algorithm can estimate the position, velocity and acceleration of target and require less computation, no matter the target is maneuvering at any way.In the end, the algorithms discussed in this dissertation are tested in experiments for different targets and input noises. The results verify the proposed algorithms.更多还原