基于线段融合的空间非合作目标稳健特征提取算法

相对位姿测量是空间非合作目标态势感知的主要内容，在位姿测量中，需要先对目标图像进行特征提取，而特征提取的精度和鲁棒性直接影响位姿测量性能。为了提高空间非合作目标特征提取的鲁棒性，本文给出一种基于线段融合的特征提取算法。该算法首先采用基于梯度的滤波器来消除空间目标图像的背景干扰，然后采用LSD直线检测算法、Hough Lines直线检测算法和Shi-Tomasi角点检测算法提取三组特征点，再用K-D空间划分树以及K最近邻搜索算法融合这三组特征点，保留包含显著特征的较少数量特征点，进一步组合成线段结构，并对线段进行融合，以此来提取反映目标整体几何框架的信息，从而提升稳健性。仿真实验和半物理仿真实验测试结果表明，本文提出的基于线段融合的特征提取方法在空间目标特征提取中具有更好的稳健性。

A robust feature extraction algorithm for spatial non-cooperative targets based on line segment fusion

Relative pose measurement is the main content of spatial non-cooperative target situational awareness. In pose measurement, feature extraction of target image is required first, and the accuracy and robustness of feature extraction directly affect the performance of pose measurement. In order to improve the robustness of spatial target feature extraction, this paper presents a feature extraction algorithm which integrates multi-processing streams and utilizes the whole geometric frame of the target. Firstly, the gradient filter is used to eliminate the background interference of the spatial target image. Then, LSD line detection algorithm, Hough Lines line detection algorithm and Shi-Tomasi corner detection algorithm are used to extract three groups of feature points. Then, k-D spatial partition tree and K-nearest neighbor search algorithm are used to fuse the three groups of feature points. The minimum number of feature points containing the most significant features is extracted, and finally the filtered feature points are combined into a broken line structure, so as to introduce the correlation information between feature points and better characterize satellite features. The semi-physical simulation results show that the proposed feature extraction method has better robustness.