Combining Stereovision Matching Constraints for Solving the Correspondence Problem

A major portion of the research efforts of the computer vision community has been directed toward the study of the three-dimensional (3-D) structure of objects using machine analysis of images (Scharstein & Szeliski, 2002). We can view the problem of stereo analysis as consisting of the following steps: image acquisition, camera modelling, feature acquisition, image matching, depth determination and interpolation. The key step is that of image matching, that is, the process of identifying the corresponding points in two images that are cast by the same physical point in 3-D space (Barnard & Fishler, 1982). This chapter is devoted solely to this problem. A correspondence needs to be established between features from two images that correspond to some physical feature in space. Then, provided that the position of centres of projection, the focal length, the orientation of the optical axes, and the sampling interval of each camera are known, the depth can be established by triangulation. The stereo correspondence problem can be defined in terms of finding pairs of true matches, namely, pairs of features in two images that are generated by the same physical entity in space. These true matches generally satisfy some constraints (Tang et al., 2002): 1. Epipolar, given two features, one in an image and a second in the other one in the stereoscopic pair, if we follow a given line, established by the system geometry, these two features must lie on this line, which is the epipolar. 2. Similarity, matched features have similar local properties or attributes. 3. Smoothness, disparity values in a given neighbourhood change smoothly, except at a few depth discontinuities. 4. Ordering, the relative position among two features in an image is preserved in the other one for the corresponding matches. 5. Uniqueness, each feature in one image should be matched to a unique feature in the other image. A review of the state-of-art in stereovision matching allows us to distinguish two sorts of techniques broadly used in this discipline: area-based and feature-based. Area-based stereo techniques use correlation between brightness (intensity) patterns in the local neighbourhood of a pixel in one image with brightness patterns in the local neighbourhood of the other image (Scharstein & Szeliski, 2002; Herrera et al., 2009a,b,c; Herrera, 2010; Klaus et al., 2006). Feature-based methods use sets of pixels with similar attributes, normally, either pixels belonging to edges (Grimson, 1985; Ruichek & Postaire, 1996; Tang et al., 2002),