Semester

Fall

Date of Graduation

2013

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Mechanical and Aerospace Engineering

Committee Chair

Marcello R. Napolitano

Committee Co-Chair

Thomas Evans

Committee Member

Powsiri Klinkhachorn

Committee Member

Xin Li

Committee Member

Giacomo Marani

Committee Member

Marcello R. Napolitano

Committee Member

Jacky Prucz

Abstract

When capturing a non-cooperative satellite during an on-orbit satellite servicing mission, the position and orientation (pose) of the satellite with respect to the servicing vessel is required in order to guide the robotic arm of the vessel towards the satellite. The main objective of this research is the development of a machine vision-based pose estimation system for capturing a non-cooperative satellite. The proposed system finds the satellite pose using three types of natural geometric features: circles, lines and points, and it merges data from two monocular cameras and three different algorithms (one for each type of geometric feature) to increase the robustness of the pose estimation. It is assumed that the satellite has an interface ring (which is used to attach a satellite to the launch vehicle) and that the cameras are mounted on the robot end effector which contains the capture tool to grapple the satellite. The three algorithms are based on a feature extraction and detection scheme to provide the detected geometric features on the camera images that belong to the satellite, which its geometry is assumed to be known. Since the projection of a circle on the image plane is an ellipse, an ellipse detection system is used to find the 3D-coordinates of the center of the interface ring and its normal vector using its corresponding detected ellipse on the image plane. The sensor and data fusion is performed in two steps. In the first step, a pose solver system finds pose using the conjugate gradient method to optimize a cost function and to reduce the re-projection error of the detected features, which reduces the pose estimation error. In the second step, an extended Kalman filter merges data from the pose solver and the ellipse detection system, and gives the final estimated pose. The inputs of the pose estimation system are the camera images and the outputs are the position and orientation of the satellite with respect to the end-effector where the cameras are mounted. Virtual and real simulations using a full-scale realistic satellite-mockup and a 7DOF robotic manipulator were performed to evaluate the system performance. Two different lighting conditions and three scenarios each with a different set of features were used. Tracking of the satellite was performed successfully. The total translation error is between 25 mm and 50 mm and the total rotation error is between 2 deg and 3 deg when the target is at 0.7 m from the end effector.

Download

Share

COinS