Active stereo vision with concurrent detection and identification of occlusion and specular highlights

Abstract

An important aim of computer vision is to recover precise whole-field three-dimensional (3D) depth maps from 2D images. A different objective is to provide relative depth information about specific region-of-interest in the scene. This thesis focuses on the latter aim. It proposes a new approach which, in addition to performing vergence, is capable of detecting and identifying occlusion and viewpoint-dependent specular highlights concurrently during the vergence process. This results in numerous advantages such as facilitating the exploration of better camera viewing position and direction which alleviates the problems created by the presence of occlusion and viewpoint-dependent specular highlights.