During on-orbit servicing of non-cooperative satellites, the relative position of the satellite with respect to the servicing vehicle is needed. Non-cooperative satellites do not have devices to assist the pose estimation (such as markers or reflector), so a system that uses the natural features of the satellite to determine its pose needs to be developed.
A machine vision-based system was developed for estimating the pose (position and orientation) of objects with circular features. Images of the object (target) are obtained using a camera. Next, an ellipse detector selects the extracted ellipse that best matches the target’s circular features on the image. The pose of the object is obtained using the geometric properties of the detected ellipse. A particular application is the pose estimation of a satellite during an on-orbit satellite servicing mission. In a conceptual servicing mission, the system detects the interface ring of a satellite during the capture phase. When the servicing vehicle is a few meters from the satellite, a robotic arm with a camera and a capture tool is used to grasp the satellite.
An interface ring is a feature common to several satellites, and is used to attach to the launch vehicle. Another convenience of using the interface ring is that it can also be used as a grasping point due to its structural resistance, which may not be true for other circular features on the satellite such as thrusters and heat shields. The ring detection is difficult since it can potentially be partially out of the camera view due to its size (diameter larger than one meter) and the fact that the satellite is covered with MLI (multi- layer insulation), which is highly reflective and has textures that can create false features that may confuse a machine vision-based system.
The system is intended to be used in the capture phase of an on-orbit satellite servicing mission, where the servicing vessel is at close proximity to the satellite (less than 20 meters), and a robotic arm with a capture tool is deployed to grasp the satellite. The camera of the machine vision-based pose estimation system is mounted on the robotic arm end-effector. The pose estimation system features a monocular camera with a wide-angle lens having a horizontal field of view of 125°, low radial distortion (3% barrel), and a maximum resolution of 1280 × 960 pixels.
The main advantage of the system is its robustness to occlusion, i.e., the pose is estimated at adequate accuracy and frame rate even when the interface ring is occluded by other features of the satellite, or when the ring is partially out of view. The pose error changes with the camera target’s relative position. This behavior can be attributed to the nonlinearity of the vision system (lens distortions, discretization of the ellipse shape by the image pixels). Another source of error is the ellipse detection process in which the detected ellipse can be different from the actual ring in the image (mix inner and outer diameters).
This work was done by Andres Velasquez Escandon, Thomas Evans, Giacomo Marani, and Marcello Napolitano of West Virginia University for Goddard Space Flight Center. GSC-16971-1