Parallel epipolar stereo matching
Herbert Jahn · 2002
Stereo matching is an important step in 3D reconstruction which is not yet solved satisfactorily. Efforts are necessary to enhance the matching quality and to reduce the processing time. To speed up the processing a parallel algorithm for coarse binocular epipolar image matching is presented. Data from epipolar image line pairs are fed into a special parallel processing structure which fulfills only simple operations such as convolution, averaging, and search for maxima. A priori information on possible maximum disparities is used too. A special smoothing algorithm can be applied to enhance the quality of the disparities. The method is to be applied as a first (coarse) approximation of 3D reconstruction using airborne digital stereo pushbroom cameras but it can also be used to fast evaluation of data from other stereo systems. The quality of the method is demonstrated with the often used Pentagon image pair.