Online Learning in Visual Tracking: Qinxun Bai (PhD Oral Exam)

  • Starts: 2:00 pm on Wednesday, September 4, 2013
  • Ends: 3:00 pm on Wednesday, September 4, 2013
Abstract: Visual tracking is one the fundamental problems in computer vision. A typical tracking system consists of several main components, including motion modeling, appearance modeling and model update. While prior knowledge about the object to be tracked may or may not be available when designing the tracker, it is widely desirable for a tracker to be able to adapt its appearance model during tracking to take into account the possible changes caused by object pose, lighting condition, distractors and occlusions. This study will focus on how the problem of appearance modeling for visual tracking, especially under the tracking-by-detection framework, can be formulated as an online classification and learning problem. Two representative methods from the vision community and one closely related work from the machine learning community will be discussed in detail. Commitee Members: Stan Sclaroff Margrit Betke George Kollios
MCS 148

Back to Calendar