Robust features extraction for on board monocular based spacecraft pose acquisition

This paper presents the design, implementation, and validation of a robust feature extraction architecture for real-time on-board monocular vision-based pose initialization of a target spacecraft in application to on-orbit servicing and formation flying. The proposed computer vision algorithm is designed to detect the most significant features of an uncooperative target spacecraft in a sequence of two-dimensional input images that are collected on board the chaser spacecraft. A novel approach based on the fusion of multiple and parallel processing streams is proposed to filter a minimum number of extracted true point features, even in case of unfavourable illumination conditions and in presence of Earth in the background. These are then combined into relevant polyline structures that characterize the true geometrical shape of the target spacecraft.