LEMS and Electrical Science Seminar
Abstract: In this talk, I'll describe our work on accurate calibration of an active binocular vision system, which is an important and fundamental issue for active stereo vision. In addition to the proposal and demonstration of a four-stage calibration process, there are three major contributions in this work. First, we propose an MFL (Motorized-Focus Lens) camera model which assumes constant nominal extrinsic parameters. Second, we develop a calibration method for the MFL camera model , which separates estimation of the image center and effective focal length from estimation of the camera orientation and position. This separation has been proved to be crucial; otherwise, estimates of camera parameters would be extremely noise-sensitive. Thirdly, we show that, once the parameters of the MFL camera model is calibrated, a nonlinear recursive least-square estimator can be used to refine all the 39 kinematic parameters. Real experiments have shown that the proposed method can achieve accuracy of one pixel prediction error and 0.2 pixel epipolar error, even when all the joints, including the left and right focus motors, are moved simultaneously. This accuracy is good enough for many 3D vision applications, and we are currently applying this result to active reconstruction of 3D environment for virtual reality.
Biography: Yi-Ping Hung received the B.S.E.E. degree from National Taiwan University in 1982, the Sc.M. degree in Electrical Engineering, the Sc.M. degree in Applied Mathematics, and the Ph.D. degree in Electrical Engineering, all from Brown University, in 1987, 1988, and 1990, respectively. He conducted research on computer vision at the Laboratory for Engineering Man/Machine Systems (LEMS), Brown University, for more than five years. He has been employed by Philips Laboratories, Briarcliff Manor, N.Y., during the summers of 1987 and 1988, working on robot vision. In 1990, he joined the Institute of Information Science in Academia Sinica, where he is now an associate research fellow and the deputy director of the institute. He is also an associate professor in the Department of Computer Science and Information Engineering at National Taiwan University. He has published more than fifty technical papers, and has two Ph.D. and ten master students graduated under his supervision. Currently, he serves as a member in the executive committee of the NTU Center of Excellence for Research in Computer Systems, of Taiwan Association of Artificial Intelligence, and of the Chinese Image Processing and Pattern Recognition Society. His research interests include computer vision, image processing, pattern recognition, virtual reality, and robotics.
Brown Applied Mathematics Dissertation Defense Information
Brown Applied Mathematics Pattern Theory and Vision Seminar
Brown University, Divison of Applied Mathematics | |
Abstract: A probability density proportional to exp(-U(x)) is given. A diffusion with -gradient U as the drift is often used to approximate this underlying probability. Instead of restricting to this particular drift, a family of diffusions with this probability as their common equilbrium is considered. Questions such as which diffusion has the fastest convergence, and what is the optimal convergence rate, are studied. For Gaussian diffusions, the situation is fairly clear, but the nongaussian case is far from being understood.
These and related problems in probability and analysis will be discussed.
Special PDE Seminar
Brown Electrical Engineering Dissertation Defense
Abstract: Implicit polynomials are among the most effective representations for complex object recognition because of their stability, robustness and invariant characteristics. This talk will first present a general overview and then focus on several key aspects of implicit polynomial technology for representing and recognizing complicated 2D and 3D shapes subject to partial occlusion and missing data. New concepts and results for fast, robust, repeatable fitting of implicit polynomials to data, invariantly representing and recognizing complicated shapes, object signature curves and invariant patches/parts, PIMs (Polynomial Interpolated Measures) and orthogonal decomposition, closed-form pose estimation, are some of the highlights of the talk. With these, we lay down a foundation that enables a technology based on implicit polynomial curves and surfaces for various object representation/recognition applications. I will also discuss some areas of content-based querying/indexing of multimedia (image/video) databases, where object shape and geometric structures are the essential factors. A prototype image-query-by-sketch system using this representation is introduced. It is built upon the Java technology and allows queries over the web. We report some preliminary results. From these results, we see that our implicit polynomial feature-based method generally performs much better than a pixel-based method that has been adopted by a number of well-known content-based image indexing and retrieval systems. Special attention is also given to the user interface part of such applications. Key insight into how geometric structures should be stored is gleaned from studying how users interactively describe and refine the shapes used to perform the queries.
Brown Applied Mathematics Dissertation Defense Information
<--- 1997 Index