Monday May 16 4:00 - 4:50 PM Kelley 1001
William Brendel Postdoctoral Researcher School of EECS Oregon State University >From Multitarget Tracking to Event Recognition in Videos This talk will address two fundamental problems in computer vision---namely, multitarget tracking and event recognition in videos. These problems are challenging, because uncertainty may arise from many aspects, including imaging quality (e.g., motion blur), and scene complexity (e.g., dynamic cluttered backgrounds, occlusions). We show that these challenges can be successfully addressed by: (1) using a multiscale, volumetric video representation, and (2) taking into account temporal constraints between events from domain knowledge. Two alternative approaches to multitarget tracking will be presented. The first approach seeks to transitively link object detections across consecutive video frames by finding the maximum independent set of a graph of all object detections. The second approach hierarchically partitions the space-time volume of a video into tracks of objects, producing a segmentation graph of that video. The resulting tracks encode rich contextual cues between sal ient video parts in space and time, and thus facilitate event recognition and segmentation. For event recognition, we use a structural probabilistic model. The talk will explain how to efficiently and robustly conduct learning and inference of this graph model on videos represented by segmentation graphs. Biography William Brendel received his Engineering Master at CPE Lyon, France in 2005. He worked for Infineon Technologies in Munich, Germany between 2003-2004 and for Laika Studios House in Portland, OR in 2007. He recently passed his PhD defense in computer science in the School of EECS at the Oregon State University. His research mostly focuses on graphics, computer vision and machine learning problems. _______________________________________________ Colloquium mailing list [email protected] https://secure.engr.oregonstate.edu/mailman/listinfo/colloquium
