报告题目：Salient Motion Detection using Potential Surface
A challenge in motion segmentation is that different motions are often mixed up and interdependent in the real data. Since the 2D representation of the dependent motions is obscure, it leads the segmentation difficult. In this talk, we will discuss a motion segmentation and recovery method for complex scenarios. Using an invariant interpretation of the motion vector, we first transform the 2D motion vector field into the 3D potential surface. In which, different motions are placed onto different layers so that they can be segmented much easier. By applying the surface fitting, the potential surfaces of global and local motions are then estimated. Finally, the recovered motions are obtained by projecting the segmented potential surfaces back to motion field. Using potential surface makes our method able to deal with both independent and dependent, rigid and non-rigid motions without prior knowledge. We will also explore its application in video alignment and action recognition.
Xuefeng Liang is an associate professor in the Department of intelligence Science of Informatics, Kyoto University. He was awarded a Ph.D. degree in information science from Japan Advanced Institute of Science and Technology, Japan, in 2006. After that, he worked on the vision system in robotics as a postdoc in Ubiquitous Functions Research Group, ISRI, National Institute of Advanced Industrial Science and Technology, Tsukuba, Japan. In 2008, he moved to London and worked on vision perception study at both Vision Lab, Queen Mary University of London and Department of Psychology, University College London UK. In 2010, he was appointed associate professor in Graduate School of Informatics at Kyoto University. His interests include computer vision, pattern recognition, image processing, and computational geometry.