Doctoral Thesis: Nonparametric Directional Perception

SHARE:

Event Speaker: 

Julian Straub

Event Location: 

32-G449 (Kiva)

Event Date/Time: 

Friday, March 24, 2017 - 1:00pm

Abstract:

Artificial perception systems, like autonomous cars and augmented
reality headsets, rely on dense 3D sensing technology such as RGB-D
cameras and LiDAR scanners.  Due to the structural simplicity of
man-made environments, understanding and leveraging not only the 3D
data but also the local orientations of the constituent surfaces, has
huge potential.  From an indoor scene to large-scale urban
environments, a large fraction of the surfaces can be described by just
a few planes with even fewer different normal directions.  This
sparsity is evident in the surface normal distributions, which exhibit
a small number of concentrated clusters.  In this work, I draw a
rigorous connection between surface normal distributions and 3D
structure, and explore this connection in light of different
environmental assumptions to further 3D perception.  Specifically, I
propose the concepts of the Manhattan Frame and the unconstrained
directional segmentation. These capture, in the space of surface
normals, scenes composed of multiple Manhattan Worlds and more general
Stata Center Worlds, in which the orthogonality assumption of the
Manhattan World is not applicable.  This exploration is theoretically
founded in Bayesian nonparametric models, which capture two key
properties of the 3D sensing process of an artificial perception
system: (1) the inherent sequential nature of data acquisition and (2)
that the required model complexity grows with the amount of observed
data. The inference algorithms I derive herein inherently exploit and
respect these properties.  The fundamental insights gleaned from the
connection between surface normal distributions and 3D structure lead
to practical advances in scene segmentation, drift-free rotation
estimation, global point cloud registration and real-time
direction-aware 3D reconstruction to aid artificial perception systems.
 
Thesis Supervisors: John W. Fisher III, John J. Leonard