Date of Graduation
Statler College of Engineering and Mineral Resources
Mechanical and Aerospace Engineering
The primary objective of this research was to use three different types of features (corners, lines, and ellipses) for the purpose of satellite grasping with a machine vision-based pose estimation system. The corner system is used to track sharp corners or small features (holes or bolt) in the satellite; the lines system tracks sharp edges while the ellipse system tracks circular features in the satellite. The corner and line system provided 6 degrees of freedom (DOF) pose (rotation matrix and translation vector) of the satellite with respect to the camera frame, while the ellipse system provided 5 DOF pose (normal vector and center position) of the circular feature with respect to the camera frame. Satellite grasping is required for on-orbit satellite servicing and refueling. Three machine vision estimation systems (base on line, corner, and ellipse extraction) were studied and compared using a simulation environment. The corner extraction system was based on the Shi-Tomasi method; the line extraction system was based on the Hough transform; while the ellipse system is based on the fast ellipse extractor. Each system tracks its corresponding most prominent feature of the satellite. In order to evaluate the performance of each position estimation system, six maneuvers, three in translation (xyz) and three in rotation (roll pitch yaw), three different initial positions, and three different levels of Gaussian noise were considered in the virtual environment. Also, a virtual and real approach using a robotic manipulator sequence was performed in order to predict how each system could perform in a real application. Each system was compared using the mean and variance of the translational and rotational position estimation error. The virtual environment features a CAD model of a satellite created using SolidWorks which contained three common satellite features; that is a square plate, a marman ring, and a thruster. The corner and line pose estimation systems increased accuracy and precision as the distance decreases allowing for up to 2 centimeters of accuracy in translation. However, under heavy noise situations the corner position estimation system lost tracking and could not recover, while the line position estimation system did not lose track. The ellipse position estimation system was more robust, allowing the system to automatically recover, if tracking was lost, with accuracy up to 4 centimeters. During both approach sequences the ellipse system was the most robust, being able to track the satellite consistently. The corner system could not track the system throughout the approach in real or virtual approaches and the line system could track the satellite during the virtual approach sequence.
Luckett, Jeremy A., "Comparison of Three Machine Vision Pose Estimation Systems Based on Corner, Line, and Ellipse Extraction for Satellite Grasping" (2012). Graduate Theses, Dissertations, and Problem Reports. 3334.