This paper describes a feature based stereo matching method which enables matching of occlusion boundaries using normalized light intensity.
In many feature based stereo methods, property constraint is used to reduce the number of matching candidates. This constraint, however, can not be used at occlusion boundaries in case that background intensity of an occlusion boundary in one image is different from that in the other image, because the difference makes the different property values.
A foreground object can be found in two images. Thus, an occlusion boundary can be matched as the boundary of the foreground object, if intensity of either region separated by the boundary in one image is similar to that in the other image. However, an object can be observed as the different intensities in both images by influences of lighting effects and the different camera characteristics. To solve this problem, at first, light intensity is normalized so that intensities derived from matching edges, which are obtained from texture edges, may be identical between images. Next, edges at occlusion boundaries are matched with using the normalized intensity.
抄録全体を表示