Automatically finding sub-regions for video tracking
Zhidong Li (NICTA)
NICTA SML SEMINARDATE: 2013-09-12
TIME: 11:15:00 - 12:15:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Tracking target in video is difficult when the target is non-rigid, such as human. One kind of methods is to build a target model based on previous knowledge. However, without the previous knowledge, a general tracker is required. Some methods only track the whole region of the target. Apart from this, sub-region based tracking is popular. We will first discuss a biologically inspired framework of video tracking based on salient proto-objects. The salient proto-objects are used as sub-regions. As detecting salient proto-objects is limited by the number of proto-objects which must be manually set, we then consider the infinite mixture model to relax the assumption. And we believe that combing the Infinite mixture model with the saliency proto-objects can extend the application to more flexible tracking problems.
