基于渐消记忆滤波的1点RANSAC单目视觉姿态估计算法

针对1点RANSAC(Random Sample Consensus)单目视觉EKF(Extended Kalman Filter)算法中的滤波发散问题,分析了滤波发散的产生原因,提出了一种基于渐消记忆滤波的1点RANSAC单目视觉姿态估计算法。该算法通过在EKF滤波方程中引入加权因子,逐渐加大当前数据的权重,相应地减少旧数据的权重,有效地扼制了算法中的滤波发散问题。最后通过两组验证性实验验证说明了算法的有效性。实验结果表明:该算法能够有效地解决1点RANSAC单目视觉EKF算法中的滤波发散问题,具有更高的精度。第一组双轴联动实验,航向角的平均误差减小2.4158?,俯仰角平均误差减小0.1782?;第二组偏航轴大角度转动实验,摄像机航向角的估计误差一直保持在1.5?以内。

1-point random sample consensus based on fading memory filtering for attitude estimation with monocular vision

In view that 1-point random sample consensus (RANSAC) for extended Kalman filter (EKF) filtering with monocular visual has filtering divergence problem, a 1-point RANSAC based on fading memory filtering algorithm is proposed with analyzing the causes of the filtering divergence. The proposed method effectively solves the filtering divergence problem by increasing the weights of the current data and accordingly reducing the old data, with the weighted factor added in the EKF. Finally, two groups of confirmatory experiments verify the effectiveness of the method. Experiment results show that the proposed method can effectively solve the problem of filtering divergence and has higher estimation accuracy. In the first experiment, the estimation mean errors of the proposed method are decreased by 2.4158° in camera’s course angle and 0.1782° in camera’s pitch angle. In the second experiment, the estimation errors of the camera’s course angle can be kept to within 1.5°.