In the stereo vision system, one of the most important problems is to obtain the corresponding points between the left and right images, because this process has direct effects upon the measurement accuracy. In this paper, a stereo matching algorithm based on isodensity lines and a new method for interval corresponding estimation is proposed. The proposed method consists of two stages. In the first stage, important corresponding points (high confidence points) are sought using the similarity of isodensity lines extracted from the left and right images, then one-to-one correspondence for the pixels existing in the intervals derived from the important points is obtained in the second stage. From the computer simulation using dummy head, adequate results were obtained even for the forehead where isodensity lines can not be extracted in conventional method.
View full abstract