基于同名像点位置偏差的特征点分类方法

Feature-point classification based on position changes of correspondences

通过处理序列图像获取目标的结构和运动参数是实现空间飞行器近距离自主导航的重要技术,在进行参数计算之前,必须区分目标特征点和背景特征点。本文首先分析了在小视场角观测情况下,序列图像帧间同名像点位置偏差与对应空间特征点位置的关系,然后提出了基于帧间同名像点位置偏差的特征点分类方法,该方法计算简单,采用滑动多帧多门限判决技术提高了分类的可靠性,仿真实验表明了方法的有效性。

Solving structure-from-motion problem from image sequences is one of the important methods to implement automatic navigation for space vehicle in close quarters. Before the parameters are computed, the feature-points must be classified to target or background. The paper firstly analyzed the relation between position changes of correspondences in two close frames and the position of 3D points under small angle of views. Then a method based on the position changes is proposed to classify the points. The method is simple and reliable in that it applies the technology of gliding-multi-frame and multi-threshold. And the experiments make it clear that the method is effective.