2000-2001 MICS Student Research
By Kevin Davis
Taking 3D Transformations into Account in Stereo Matching
Advisor: Dr. Jeff McKinstry
Given two pictures of the same scene taken from slightly different points of view (i.e. a stereogram), it is possible to determine the distance from the camera for each point in the scene that is visible in both pictures assuming that a unique match for that point can be found. This is the basis for human binocular vision, or stereo vision. The task of finding corresponding points in the two pictures is know as the correspondence problem and is the problem we tried to solve. A common approach to stereo matching is to take a small patch from the left eye image and slide it along the right eye image looking for the best match. it has been pointed out by others that this assumes that the object is perpendicular to the viewing axis; thus any surface that is not will not provide a perfect match. This leads to the possibility of false matches. We overcome this problem by assuming many possible depth profiles for the surface that generated a given image patch. The depth profile that gives the best matching score will be more accurate because the assumption of fronto-parallel objects is dropped. We demonstrate the results of this method with computer simulations using artificial Stereograms created with three-dimensional computer graphic software.