Toward 3D uncalibrated monocular visual servo

Abstract

This work is an initial step toward combining control theory with computer vision for the case of uncalibrated monocular (or single-camera) three-dimensional manipulation. We investigate an achievable control goal using a novel image measure and derive a sampled-data control system that results in quick convergence of the end-effector trajectory to a depth-invariant velocity subspace in the camera coordinate frame. We also discuss how this uncalibrated behavior can be used to perform online calibration.