Geometric Methods for Controlled Active Vision.

摘要

In this just completed research program, we developed several new directions for our work in controlled active vision. We have developed a general framework for geometric observer-like structures based on non- parametric implicit (level set) curve descriptions of dynamically varying shapes. Special emphasis was given to the geometric nature of the dynamical system as well as the key issue of robustness. In particular, we formulated an approach to the problem of information transport and filtering from a measurement curve to an estimated curve. In this framework, we may naturally incorporate several different tools such as particle filtering and segmentation methods.