Files in this item



application/pdfMalik_Rahul.pdf (33MB)
(no description provided)PDF


Title:An Activity-Based Framework for Automated Placement and Configuration of Multiple Stereo Cameras
Author(s):Malik, Rahul
Director of Research:Nahrstedt, Klara
Doctoral Committee Chair(s):Campbell, Roy H.
Doctoral Committee Member(s):Nahrstedt, Klara; Bajcsy, Peter; Huang, Thomas S.; Dorai, Chitra
Department / Program:Computer Science
Discipline:Computer Science
Degree Granting Institution:University of Illinois at Urbana-Champaign
multiple stereo camera placement
color constancy across multiple cameras
Abstract:With the advent of virtual spaces, there has been a need to integrate physical worlds with virtual spaces. The integration can be achieved by real-time 3D imaging at several geographically distributed locations followed by fusion of information from multiple virtual and physical spaces. Systems that enable such information fusions are called tele-immersive and need to be easily deployed at many geographically distributed locations. One of the key requirements of these systems is to achieve high quality 3D and appearance information about physical spaces. Performance of several applications such as face recognition, activity classification, human tracking, etc. improves significantly from it. The quality of information depends on placement of stereo cameras and color constancy of scene appearance from multiple imaging devices. Thus, an important challenge in the deployment is the optimal placement of stereo cameras and color adjustments. Placement of stereo cameras directly impacts 3D reconstruction error and spatial resolution achieved (the quality of 3D information). In addition, color adjustments are needed to produce artifact-free and visually appealing rendered results (the quality of appearance information). The color adjustments include calibrating inter-camera color responses and matching measured colors to actual colors. This dissertation addresses two problems of improving the quality of 3D video obtained from stereo cameras. First, we consider the placement of multiple stereo cameras in order to minimize 3D reconstruction error and satisfy the resolution requirements specified by the user. In order for the user to specify the resolution requirements in space, we present a activity model framework for the user to concretely describe the minimum resolution required. Based on this resolution specification from the user for different activities, given by the activity model, this dissertation optimizes the placement of stereo cameras. Placement of stereo cameras differs significantly from the placement of 2D cameras in terms of objectives and applications. This dissertation finds the best placement for stereo cameras according to by minimizing the uncertainty in stereo depth estimation as well as satisfying different constraints by the user such as minimum resolution required, 360o placement, room geometry, etc. Using this, we show the improvements achieved according to our placement as opposed to other ad-hoc placement techniques, such as nearest, farthest and random. We also present the effects of different type as well as number of activity models upon the camera placement. The second problem which this thesis addresses is that of color constancy across multiple cameras. The problem of color constancy is important as consistent inter-camera color responses are needed to produce artifact-free and visually appealing rendered results. Also, the determined color should be similar to actual color. This is a complex problem as different cameras have different color responses, even with same hardware settings. We use Gretag MacBeth Color Checker chart to achieve similar response for different cameras. In this thesis, we present algorithms for determining the best hardware register value settings as well as additional software based transforms to achieve more fine-grained refinement. In addition to these, this thesis also presents three user studies to better understand the parameters that most impact the performance. In first study, we study the impact of communication media on remote collaboration. We compare 3D video tele-immersive, 2D video Skype and face-to-face for collaboration in a remote product development scenario. In second study, we seek to understand the importance of various tele-immersive system parameters for virtual-physical interaction. The task is that of moving a virtual ball to a basketball basket, and it requires localization, orientation and motion coordination for a human in tele-immersive environment to complete the task. Finally, in third study, we study the effect of cues and view presentation styles upon the ability to regain proprioceptive abilities. We consider wheelchair basketball players maneuvering the wheelchair as the activity and seek to compare the effectiveness of 3D tele-immersion over other cues and view-presentation styles.
Issue Date:2011-01-21
Rights Information:Copyright 2010 Rahul Malik
Date Available in IDEALS:2011-01-21
Date Deposited:2010-12

This item appears in the following Collection(s)

Item Statistics