Depth information is a very important cue in 3D object recognition, and a method of extracting depth information is stereo method using cameras. In stereo vision, the most difficulty is finding corresponding points which pair two points of a real-world point in the two image planes.
This paper describes a new line matching technique to slove correspondence problem. This method, which uses a line window consisting of two modified epipolar lines(corresponding to the start point and end point of a line in the reference plane) in other plane to reduce a search region, find a matched line to be applied to the relation of cross correlation coefficient between a line in the reference plane and candidate lines in other plane. Epipolar line need not be horizontal to scan line. This algorithm was implemented at KAISION - I vision system, applied to simple objects and complex objects in circumstance which is at angle of view of about 27˚ and distance of about 2m.