13 - Geometric Knowledge-Based Systems Framework for Stereo Image Matching

Abstract

One of the basic goals of computer vision is the recovery of three-dimensional (3D) structures from two-dimensional (2D) images. This process can be achieved by establishing correspondences between sequences of 2D images of a scene taken at different times or displacements or from different perspectives using either active or passive imaging techniques. In active techniques, the correspondence problem is easily solved by using artificial sources of light and illumination, which is usually an expensive process. Passive techniques, on the other hand, are much cheaper but they have to rely on existing ambient illumination to solve the correspondence problem. The process of stereo vision comprises six steps: image acquisition, camera modeling, feature extraction (acquisition), image matching, distance (depth) determination, and interpolation. Stereo image matching techniques can be classified into two categories: area-based and feature-based techniques. Area-based techniques match intensity levels in the local neighborhood of a pixel in one image with intensity levels in a corresponding neighborhood of a pixel in the other image. Feature-based techniques use higher-level features extracted from the two images as matching primitives rather than the low-intensity levels of area-based techniques.