Dynamic integration of depth cues for surface reconstruction from stereo images

Abstract

This thesis describes a computational method for surface estimation from stereo images. The images are obtained with a dynamically controlled camera system, having imaging parameters which are selected on the basis of an evolving composite description of the surfaces in the environment. This follows the paradigm of active vision, which implies a feedback mechanism to select physical sensory parameters in order to improve the quality of the derived results.

In addition to stereo disparity, camera vergence and focus are used as sources of depth information. Through the integration of these cues, a global surface map of the visual field is synthesized by systematically scanning the scene, combining estimates of adjacent, local surface patches, each acquired by an intermediate camera configuration and having a small depth range. The method provides for the local optimization of calibrated values for imaging parameters. This approach, when coupled with dynamic image acquisition and analysis, results in a powerful mechanism for autonomous surface reconstruction.