Speaker:  Dr. Clint Slatton
Topic:  Information-theoretic segmentation of sampled topographic elevations
Room: NEB 201
Time: 5:00pm-6:00pm

Abstract:
Pulsed laser systems carried onboard aircraft have the ability to yield
high-resolution three-dimensional sampling of natural terrain with good
geometric fidelity.  When lasers with small beam divergence are used, a
small fraction (<=10%) of the pulses penetrates the vegetation canopy and
provides elevations of the bare surface.  However, these pulse returns must
be separated from the rest before being interpolated to form a dense
estimate of the bare surface.  A two-step hierarchical segmentation
procedure is employed to first locate major terrain classes and then to
segment the raw point data.  The surface and vegetation density of natural
terrain are both non-stationary random processes, so decision boundaries for
the segmentation are determined from the data adaptively.  Casting the
detection problem in a probabilistic framework allows for principled
solutions with measurable uncertainty.  Examples of surface detection using
airborne measurements over a dense forest canopy are presented.