From - Mon Aug 20 09:57:12 2001 Date: Tue, 24 Jul 2001 13:30:37 -0400 (EDT) From: Amitabh Varshney Subject: CFP: Visual Computer Special Issue on Computational Video CALL FOR PAPERS THE VISUAL COMPUTER SPECIAL ISSUE on Computational Video Guest Editors: Y. Aloimonos and A. Varshney Center for Automation Research University of Maryland at College Park SCHEDULE: Submission of Manuscripts: October 30, 2001 Review and decision notification: December 1, 2001 Revised camera-ready copies: December 30, 2001 Publication: early 2002 The past few years have witnessed a number of fundamental developments in the processing of perceptual data. This, in conjunction with advances in technology, has led to a dramatic increase in a variety of applications ranging from video manipulation to virtual reality. More specifically, application of geometry, statistics, control theory and physics to visual input opens new avenues and creates new methodologies for a variety of important, novel areas. Examples include video editing, teleimmersion and virtual reality, three-dimensional video, synthetic worlds and the synergistic mixture of graphics with computer vision. In video editing one needs to alter the content of the video by deleting or inserting particular objects; this requires recovering and maintaining relationships between different coordinate systems. If the camera that captured the video segment of interest was moving, then only if the movement of the camera can be determined will we be able to place a new object in the video. This is a problem that has been heavily studied in computer vision. Similarly, if there are independently moving objects in the video that need to be manipulated, they must be detected and localized. This is the motion segmentation problem, another topic of great interest in computer vision. In virtual reality and synthetic world models one needs to synthesize views of a realistic scene; this can only happen through the synergy of graphics and vision by using models of the real world in artificial environments. Virtual environments immerse their viewers into totally synthetic worlds disembodied from the viewers' senses. Augmented reality, on the other hand, extends a viewer's real-world sensing by projecting graphical models into the viewer's field of view. These models must be aligned with the viewer's image of the real world. This registration problem can be solved in a close-range indoor environment by using magnetic or infrared devices that track the viewer's head pose. Outdoors, however, the only universal methods of localizing the viewer, or in general of performing registration tasks, are based on visual and inertial sensors. Using a video-based head-mounted display has the advantage of using the same images for computing pose that will be superimposed on the virtual objects. A potential problem with this approach is the need for intrinsic camera calibration. But outdoor applications where the devices must be wearable cannot guarantee the preservation of an off-line a priori calibration. Thus one has to treat the registration and calibration problems simultaneously, a problem of great current interest in computer vision. Creating avatars amounts to applying control theory techniques to realistic models of animate objects. This requires both realistic models and knowledge of the control dynamics that will produce a particular action. These problems are currently dealt with in various ways, but it is generally accepted that robust solutions will be obtained by applying computer vision methodologies to develop shape and action descriptions. Creating 3D video that can be seen from any viewpoint amounts to relating multiple video streams. In order to visualize a video sequence from any viewpoint it would appear that one needs to collect it from all possible viewpoints. This is, however, impossible; a general solution requires the development of appropriate shape descriptions of the scene depicted in the video. This is a version of the well-known structure from motion problem in computer vision. All these applications require understanding the relationships between the analysis and synthesis of perceptual data. The field of computer vision is devoted to the analysis problem -- from visual data, recover models of the world. The field of graphics is devoted to the synthesis problem -- develop models that give rise to realistic visual data. The new emerging paradigm makes the distinction between these disciplines quite fuzzy, as models extracted by analysis can be used for synthesis. Indeed, analysis and synthesis are becoming inseparable. Although the applications described above, and many others, appear to arise in different communities and to have an engineering flavor, there exists a scientific theme that ties them together. The techniques that underlie the eventual solution of these problems constitute, basically, tools for visual imagination. These tools will allow us to simulate large numbers of processes; they are tools that will help us visualize and imagine. In this special issue of VISUAL COMPUTER entitled "Computational Video" we solicit submissions on any of the topics described above. It is our hope that the papers in the journal will help define the new emerging themes in the confluence of computer graphics and computer vision. Please send by October 30, 2001 three copies of a manuscript prepared according to the journal's guidelines (about 5000 words) to: Yiannis Aloimonos Amitabh Varshney yiannis@cs.umd.edu, varshney@cs.umd.edu Center for Automation Reseach A. V. Williams Bldg University of Maryland at College Park College Park, MD 20742