基于视觉的机动目标定位跟踪的滤波算法

在视觉定位系统中，由于各种噪声的影响，运动目标的三维位置和姿态的计算精度受到一定限制。为了提高运动物体的定位跟踪精度，提出了一种有效的滤波算法。和已有的方法相比，这种算法具有以下两个特点：第一，不再局限于平缓运动的物体，它对于未知运动规律的机动目标同样有效，第二，由于避免了扩展卡尔曼滤波器的使用，滤波复杂度有所下降。通过分析噪声对位姿计算误差的影响，建立了一组描述位姿测量值和真实值关系的线性测量方程。然后，分别给出了两种滤波算法：基于有限记忆的检测自适应滤波和基于数值微分模型的卡尔曼滤波。在检测自适应滤波算法中，给出了分别适用于快机动和慢机动的最优机动检测函数。一旦检测出机动发生，系统采用有限记忆滤波进行矫正。在第二种滤波算法中，系统采用数值微分技术构造出了描述机动目标运动行为的鲁棒估计模型。并且，引入了衰减因子，以防止滤波器的发散现象。该衰减因子可以根据位姿计算值自适应估计。最后，通过伪贝叶斯估计算法，将两种滤波器进行数据融合，有效的降低了机动时刻位姿估值的误差抖动，进一步提高了定位跟踪精度。仿真结果验证了本算法的有效性。

Filter Algorithm for Visual Positioning and Tracking of Maneuvering Target

It presents a filter algorithm for accurately visual tracking of a maneuvering target.Emphasis is given to find a solution for the 3D pose(position and orientation) estimation degradation caused by noise in the images.Superior to previous approaches that were limited to the assumption that the target motion is slow and smooth,this algorithm is applicable for a maneuvering target that acts in an unknown manner.Moreover,the computation load is reduced by avoiding the need of extended Kalman filtering.First,by analyzing the effect of noise in 2D images on 3D pose errors,linear measure equations relating the nominal pose parameters(obtained by 3D reconstruction) and the true ones are derived.Then,two filter schemes are introduced respectively.The first filter uses maneuver detection technique,in which the criterion of an optimal detector is deduced,and two sub-optimal detectors suitable for fast and slow maneuvers are given respectively.When a maneuver is declared,limited memory filtering is adopted for update.In the second filter,a robust dynamic model is constructed by the numerical differentiation technique.To restrain the divergence caused by truncation errors of estimate model of numerical differentiation,a fading factor is imported,which can be controlled adaptively based on the measurements.Finally,generalized pseudo Bayes algorithm is employed to combine the two filters for more accurate tracking.Experiment results illustrate the capacity of this algorithm.