Visual detection of polygonal retroreflective markers for pose determination of uncontrolled space targets

The need to remove space junk from crowded orbital regimes has fostered the interest towards the development of technologies enabling autonomous active debris removal missions. Due to the lack of a dedicated inter-satellite link, approach and docking/berthing operations must rely on visual-based solutions to ensure accurate relative navigation capabilities. In this respect, while the current debris population is composed of uncooperative targets, future satellites are expected to be equipped with fiducial markers to ease the relative navigation function of an autonomous chaser if a servicing/removal mission is requested. These markers shall be placed on all the available target faces to ensure the capability to handle also tumbling satellites. In this framework, this work deals with the design and testing of a relative navigation module for proximity operations towards passively cooperative space targets focusing on the image processing pipeline. The module features a monocular camera and an active, low-power, wide-beam laser, both operating in the infrared band. This emitter is used to illuminate markers made of highly reflective material, having various polygonal geometries on different target faces so that markers’ identification is based on their shape discrimination. The proposed solution does not require an a-priori pose information knowledge. First, numerical tests are executed in a dedicated simulation environment including a synthetic image generator based on the open-source software Blender. A large variability of pose conditions is reproduced in terms of both target distance and observation angles. Then, a campaign of experimental tests is carried out employing a prototype of the relative navigation module made with commercial-off-the-shelf components. Numerical results demonstrate that the module can detect and discriminate 6-cm size markers with various shape from 15 m distance up to docking/berthing, while experimental tests, performed up to distances of 8 m, confirm the capability to detect the shape of 5-cm size markers. The algorithm proves to be also robust to observation angles up to 65°, both in numerical and experimental validations, showing mm-level detection errors.