Optimal Feature Design for Vision-guided Manipulation

Optimal Feature Design for Vision-Guided Manipulation Azad Shademan, Master of Applied Science in Electrical and Computer Engineering Department, Ryerson University, 2005 Numerous industrial applications use vision-guided manipulation, where cameras are used to generate the feedback control signal. Current vision algorithms select a set of image features to estimate pose in real-time. Object design has received little attention in this context; more importantly, improper designs could lead to task failure. The focus of this thesis is on optimal industrial design of features. The goal is to construct the theory of optimal design for vision-guided manipulation. The problem is posed as a multi-objective optimization problem within an axiomatic-design theoretic framework. The visual and directional motion resolvability objectives are specified for a given 6-dimensional camera trajectory. Simulation results verify that the redesigned object satisfies the objectives. The practical implementation is attempted by camera calibration, pose estimation, and experiments on a real industrial object under a known camera trajectory.