Abstract

In the recent past, trinocular matching [1]-[3] has been proposed as a new technique for achieving local matching in stereo analysis. However, the use of a third image for stereo analysis requires additional computations like preprocessing an extra image (which usually involves expensive 2-D convolutions). Understanding the trade-off between the contribution of a third camera in stereo correspondence and the additional computations needed for the third image is of paramount importance for the design of a real-time stereo-based vision system. Contemporary literature does not contain detailed analysis of the gain in local matching performance of trinocular stereo vis-a-vis binocular stereo. Comparing the two systems based upon accurate ground truth verification of computed disparity has also not been presented. This paper analyzes quantitatively the improvement in local matches when a third image is used.

© 1989 Optical Society of America