Development of Correspondence Field and Its Application to Effective Depth Estimation in Stereo Camera Systems

Stereo camera systems are still the most widely used apparatus for estimating 3D or depth information of a scene due to their low-cost. Estimation of depth using a stereo camera requires first estimating the disparity map using stereo matching algorithms and calculating depth via triangulation based on the camera arrangement (their locations and orientations with respect to the scene). In almost all cases, the arrangement is determined based on human experience since there lacks an effective theoretical tool to guide the design of the camera arrangement. This thesis presents the development of a novel tool, called correspondence field (CF), and its application to optimize the stereo camera arrangement for depth estimation.