基于纹理边界检测的航天器椭圆特征提取方法

为确定失效航天器等非合作目标的相对位姿，提出一种通过纹理边界检测的椭圆特征提取方法。该方法假设椭圆特征是航天器表面两种不同纹理的边界，利用上一时刻相对位姿信息，将对接环离散几何模型投影到像平面，并沿各离散点的法向方向通过概率方法检测纹理边界点。利用随机抽样一致(RANSAC)方法剔除边界点中的粗大误差，进而拟合出椭圆参数。纹理特征对光照变化具有鲁棒性，因此该方法能够在变光照、星体表面反光不均匀等复杂情况下快速准确地提取图像中的椭圆特征。本文以对接环图像特征提取为例进行仿真校验，分析了算法参数和噪声对提取椭圆精度和时间的影响。利用真实图像与基于梯度边缘的椭圆提取方法进行对比，结果表明，所提出的算法具有较高的精度和速度。

Elliptical Feature Extraction on Spacecraft Based on Texture Boundary Detection

In order to determine the relative position and attitude of a non-cooperative target, such as a failed spacecraft, an algorithm based on the texture boundaries is developed for ellipse extraction. It is assumed that the elliptical feature represents the boundary of the two different texture regions on the surface of the spacecraft. With the relative position and attitude information from last time, a discrete geometric model of the target is projected onto the image plane. The texture boundary points are detected by the method of probability along the normal direction of each discrete point. Then the points with large errors are eliminated by the random sample consensus (RANSAC). Subsequently, the parameters of the extracted elliptical feature are obtained. As the texture feature is robust to changes in illumination, the proposed algorithm can extract the elliptical features rapidly and accurately in the image in the complex cases, such as the changes in illumination and unevenness of the surface reflection on the spacecraft. Considering the feature extraction of a docking ring as an example, the simulation is performed. The influence of the parameters and noise on the accuracy of the proposed algorithm is investigated, and the cost time of the result is analyzed. The proposed algorithm and the algorithm based on the gradient information are compared in real images. The results show that the proposed algorithm has high accuracy and low computational cost.