遮挡情况下刚体位姿估计的自适应无迹卡尔曼分布式融合

Adaptive unscented Kalman filter distributed fusion for rigid body pose estimation under occlusion

针对视觉目标位姿估计系统中常出现的因为特征点遮挡而造成系统估计结果不准确的问题,本文提出了一种利用自适应无迹卡尔曼滤波(AUKF)作为局部滤波器的分布式融合估计方法.通过引入改进的Sage-Husa噪声估计器自适应过程噪声.根据特征点识别量将遮挡情况分为部分遮挡和严重遮挡,对部分遮挡子系统根据先验信息修复缺失观测点后进行局部滤波估计,严重遮挡子系统不参与融合,利用当前时刻整体估计结果对其进行初始化.通过仿真获取了区分遮挡情况的阈值,实验结果表明所提方法能够提升系统在遮挡情况下的估计精度与鲁棒性.

Aiming at the problem of inaccurate estimation results caused by occlusion of feature points in visual target pose estimation system,a distributed fusion estimation method using adaptive unscented Kalman filter(AUKF)as local filter is proposed in this paper.The process noise is adapted by introducing the improved Sage-Husa noise estimator.According to the recognition quantity of feature points,the occlusion is divided into partial occlusion and severe occlusion.The partial occlusion subsystem is estimated by local filtering after repairing the missing observation points according to prior information.The severe occlusion subsystem does not participate in the fusion,and it is initialized with the current global estimation result.The threshold of distinguishing occlusion is obtained by simulation,and the experimental results show that the proposed method can improve the estimation accuracy and robustness of the system under occlusion.