Stereoscopic image visualization for telerobotics. Experiments with active binocular Cameras

Abstract

The ability to perceive a three dimensional view of a remote workspace is an essential factor for teleoperation performance. In this chapter we present a novel class of 3D video interfaces that actively control the image acquisition system geometry to improve the teleoperation process. We describe a prototype system composed by two main components: (i) a visualization system called 3D-VIP, which receives two video input streams from a binocular camera system and combines them into a single stereoscopic video image displayed on a computer screen, and (ii) an automatic vergence control system that commands the movements of the binocular camera system to minimize the disparity of observed objects, thus facilitating depth perception. Automatic vergence control is based on the estimation of dominant disparity in the stereo image pair. It employs a foveal image representation, whose image resolution decreases toward image periphery, to focus attention on central objects and attenuate the influence of background clutter. Foveal images have smaller sizes than usual cartesian images which allows very fast sample rates and adaptation to working conditions. We present results on real teleoperation experiments which show that active systems compare favorably to conventional ones with respect to working range. Given the increasing availability of high quality cameras and motorized systems, the integration of automatic vergence control with stereoscopic video devices will pave the way for a new advanced generation of human interfaces in teleoperated systems. © 2007 Springer-Verlag Berlin Heidelberg.