One of the basic tasks of computer vision systems consists in tranforming image data (color and image brightness) into 3-D geometric data such as depth. In principle depth can be recovered either from monocular cues (shading, shape, texture, motion) or from binocular cues (stereo correspondences). We believe that the latter (stereo) is more likely to lead, within a year or two, to robust and reliable systems. In the past we contributed to the development of stereo methods and techniques.

In the future we plan to concentrate on the development and design of real-time stereo. One of the most time-consuming tasks associated with stereo is matching. We developped a correlation-based dense stereo matching technique as well as an accurate stereo triangulation method and we plan to implement a stereo device such that it operates at 20-30 frames/second. Within the OCETRE project we will develop a stereo system using the programming facilities of graphical processing units. Within a contract with Renault we will develop a stereo-based prototype system for car driving assistance.