Title:
Sparse Key Point Representations of Dynamic and Amorphous Structure
Abstract:
Recovering actionable, low-dimensional representations of embedded structure within large sensor datasets is a problem that has been extensively studied in video, imagery, and higher dimensional remote sensing datasets. Scale Invariant Feature Transform (SIFT) and related key point descriptor schemes comprise one class of approach that has seen resurgence of interest. We discuss adaptation of such methods to situations in which the structures of interest are highly dynamic, evolving and amorphous; examples include finding and tracking significant coherent features within massive-scale physics simulations and in machine vision problems in robotic handling of unfixed, irregular objects.
|