Learning Visual Flows: A Lie Algebraic Approach

We present a novel method for modeling dynamic visual phenomena, which consists of two key aspects. First, the in- tegral motion of constituent elements in a dynamic scene is captured by a common underlying geometric transform pro- cess. Second, a Lie algebraic representation of the trans- form process is introduced, which maps the transformation group to a vector space, and thus overcomes the difficul- ties due to the group structure. Consequently, the statis- tical learning techniques based on vector spaces can be readily applied. Moreover, we discuss the intrinsic con- nections between the Lie algebra and the Linear dynamical processes, showing that our model induces spatially vary- ing fields that can be estimated from local motions without continuous tracking. Following this, we further develop a statistical framework to robustly learn the flow models from noisy and partially corrupted observations. The proposed methodology is demonstrated on real world phenomenon,...

• Type: Conference
• Year: 2009
• Where: CVPR
• Authors: Dahua Lin, W. Eric L. Grimson, John W. Fisher III
• Keywords: Computer Vision, CVPR 2009, Learning Visual Flows