A new method of video object extraction is proposed to accurately obtain the object of interest from actively acquired videos. Traditional video object extraction techniques often operate under the assumption of homogeneous object motion and extract various parts of the video that are motion consistent as objects. In contrast, the proposed active video object extraction (AVOE) paradigm assumes that the object of interest is being actively tracked by a camera moving in 3D and classifies the possible motions of the camera that result in the 2D motion patterns as recovered from 2D image sequences. Consequently, the AVOE method is able to extract only objects of interest from active videos while ignoring other less important objects. We formalize the AVOE process using notions from Gestalt psychology. We define a new Gestalt factor called "shift and hold" which acts as a bridge between 2D Gestalt groupings and 3D object perception.; We also propose a novel cooperative method for efficient dense 2D motion estimation as part of the AVOE framework. Using motion fields recovered from successive frames of the video, we propose a core algorithm to perform 2D object extraction. In addition, we also propose a linear programming based boundary adjustment algorithm that takes into account the strength and orientation of candidate boundary pixels to refine object outlines extracted by the core algorithm.; More effective indexing and retrieval techniques can be devised if the extracted objects are not limited only to their 2D views but can be intelligently integrated to form 3D object models. In this way, objects can be searched and retrieved using their 3D shapes in addition to the 2D image based features. In order to address this need for 3D object models, we also describe the design and implementation of an active video object extraction and 3D reconstruction system as part of this thesis.
展开▼
