Title：Simultaneous Correspondences and Motion Estimation in Geometric Computer Vision
In this talk, I will introduce our recent efforts of estimating correspondences and motion simultaneously in geometric computer vision. Particularly I will discuss two of our ongoing research projects on this topic in details.
Motion segmentation can be addressed as a subspace clustering problem, assuming that the trajectories of interest points are known. However, establishing point correspondences is in itself a challenging task. Existing approaches tackle the correspondence estimation and motion segmentation problems separately. We introduce an approach to performing motion segmentation without any prior knowledge of point correspondences. We formulate this problem in terms of Partial Permutation Matrices (PPMs) and aim to match feature descriptors while simultaneously encouraging point locations to satisfy subspace constraints. This lets us handle outliers in both point locations and feature appearance. The resulting optimization problem can be solved via the Alternating Direction Method of Multipliers (ADMM), where each subproblem has an efficient solution. In non-rigid structure from motion, given point correspondences across multiple images, it has been shown that 3D non-rigid structure can be recovered through factorization techniques. We present a unified framework to simultaneously solve for point correspondences and non-rigid structure by using the PPMs, aiming at establishing point correspondences and enforcing low-rank constraint in the deformable shape. Our new formulation can handle outliers and missing data elegantly.
Yuchao Dai is currently an ARC DECRA Fellow with the Research School of Engineering at the Australian National University, Canberra. He received the B.E. degree, M.E degree and Ph.D. degree all in signal and information processing under supervision of Prof. Mingyi He from Northwestern Polytechnical University, Xi’an, China, in 2005, 2008 and 2012, respectively. He was a visiting student at ANU from Oct. 2008 to Oct. 2009 under the supervision of Prof. Richard Hartley and Dr. Hongdong Li with the support of the China Scholarship Council. His research interests include structure from motion, multi-view geometry, human-computer interaction, compressive sensing and optimization. He has published papers in both top-ranked journals and prestigious conferences such as IEEE TPAMI, IJCV, ICCV, CVPR and ECCV. He won the best paper award in CVPR 2012 (the first one in mainland China). His recent work aims at developing dense non-rigid structure recovery from monocular video sequences, with a view to support analysis and understanding of complex dynamic scene.
戴玉超博士现为澳大利亚国立大学工程研究院ARC DECRA学者。师从西北工业大学何明一教授乐彩网app下载，他分别于2005、2008和2012年获得信号与信息处理学科学士、硕士和博士学位。2008年至2009年受国家留学基金委资助赴澳大利亚国立大学联合培养乐彩网app下载，对方导师为几何计算机视觉的奠基者Richard Hartley教授。他的研究方向包括结构与运动恢复乐彩网app下载、多视角几何、人机交互乐彩网app下载、压缩感知和最优化等乐彩网app下载。他先后在计算机视觉领域的顶级期刊和会议如IEEE模式分析与机器智能（TPAMI）、国际计算机视觉期刊（IJCV）乐彩网app下载、国际计算机视觉大会（ICCV）、IEEE计算机视觉与模式识别会议（CVPR）和欧洲计算机视觉会议（ECCV）等发表论文多篇。他与何明一教授和澳大利亚国立大学Hongdong Li副教授合作完成在非刚性结构与运动恢复方面的研究工作获得CVPR 2012最佳论文奖（大陆高校28年来首次获得该奖项）乐彩网app下载。近期他的研究工作致力于通过单目视频序列进行复杂动态场景的分析和理解乐彩网app下载。