测量误差不确定性加权的立体视觉位姿估计目标函数

针对视觉测量过程中配合目标特征点在成像时灰度模式的各向异性且非独立同分布对位姿估计求解的影响,建立了基于特征点测量误差不确定性加权的位姿估计目标函数。利用协方差矩阵描述特征点的方向不确定性,分析成像特征点的不确定性对目标函数的作用权重,并将特征点测量误差的不确定性融入到空间共线性误差函数中,进而构造了基于成像特征点测量误差不确定性加权的新目标函数。该方法适应于特征点测量误差具有不同程度的方向不确定性的情况,最后通过广义正交迭代算法对该目标函数进行迭代优化求解。实验表明,当测量空间为2 300 mm×1 400 mm×1 400mm,采用新目标函数得到靶标重投影后图像坐标的最大误差不超过0.11pixel,立体视觉系统对标准杆的相对测量精度优于0.01%。所得的结果验证了新目标函数的位姿估计精度高、稳定性强,适用于实际的工程应用。

A objective function with measuring error uncertainty weighted for pose estimation in stereo vision

In order to investigate the effect of anisotropic and correlated non-identical gray distributions of feature points on pose estimation, a novel objective function with error uncertainty weighted of feature points was proposed. In the method, the inverse covariance matrix was utilized to describe the directional uncertainty of feature points, and weighted contribution of uncertainty to the error objective function was analysed. By transforming the uncertainty into a covariance-weighted data space, the novel objective function was constructed, which was suitable for actual applications. Finally, the optimized solution to the novel objective function was obtained via generalized orthogonal iterative algorithm. The simulation and practical experiments show that the maximum error of re-projection image coordinates of the target is less than 0.11 pixels and the measurement relative error for standard gauges is superior to 0.01% within the space 2300 mm×1400 mm×1400 mm. The results verify the high accuracy and strong robustness of the proposed approach, and should therefore have potential for engineering applications.