Active vision based stereo vision

Abstract

In this paper, a new stereo method is introduced which, unlike to the traditional ones, does not involve image points correspondence. By controlling a pan-tilt-translation camera platform to move along an axis perpendicular to the camera optical axis, and equidistantly take multiple (M) images during the camera movement, it is shown that M image points coming from a same object point will lie on a straight line in the virtual image, which is constructed by combining the M Epipolar lines of the object point in the image order, and that the depth of this object point will be proportional to the slope value of the straight line. Thus under the active vision paradigm[1], the problem of determining the depth of an object point is converted into a much easier one of calculating the slope value of a straight line in virtual image. The experiments showed that the proposed stereo vision method is robust, and is especially adequate for qualitative vision applications.

This work was supported by the Chinese National Foundation of Sciences and the Chinese National High Technology Program