为了实现多圆特征识别和姿态估计,将航天器常见的几何特征如旋转体(SOR)应用于姿态估计,提出了一种基于椭圆归类的单目视觉姿态估计方法。在图像中采用基于弧段的椭圆检测方法检测目标上的椭圆特征;提出一种基于SOR空间圆平行性和垂直性约束的椭圆归类方法,得到合理的椭圆特征;利用这些特征估计航天器和摄像机之间的姿态。实验结果表明:该方法具有较好的椭圆归类效果和姿态估计精度,对于含有0~16%的椒盐噪声的仿真图,归类精确率不低于97%;实物实验中,角度误差不超过1°,深度方向(小于10 m)的测量误差不超过80 mm,其他方向的测量误差不超过15 mm。

A novel ellipse classification and relative pose estimation algorithm is presented based on surface of revolution (SOR) for multiple-ellipse images. Firstly, the contour following method is performed on images to detect ellipses. Secondly, the meaningful ellipses are obtained based on parallel constraint and vertical constraint. Finally, relative pose between the target and the camera is calculated by the meaningful ellipses. Experimental results indicate that the method performs well in ellipse classification and pose estimation. The precision of classification is higher than 97% for synthetic images corrupted by 0-16% salt-and-pepper noise. The absolute error of the pose angle is less than 1°, and the absolute errors along the depth direction and other directions are less than 80 mm and 15 mm, respectively, when the measurement distance is less than 10 m.