Fe ature -base d lase r data
simplificatio n fo r lo w po lygo n
visualizatio n
Pamela Fox, pfox@usc.edu
Shirish Doshi, shirishd@usc.edu
Suya You, suyay@usc.edu
Ulrich Neumann, uneumann@imsc.usc.edu
Integrated Media Systems Center
University of Southern California

Overview
• Objective
• Related Work
• Our Approach
– Pre-processing: Polygon Reduction
– Simplification into Primitives
• Results
• Extension to outdoor/mixed scenes
• Future Work
• Conclusion

Objective
• Simplified representation of
scanned laser (range) data from
indoor scenes
• Problems:
Too many triangles,
too many holes,
not enough memory!
• Original application for
RFID tracking

Applications
• urban planning
• virtual reality based geographics
information system (VRGIS)
• environment monitoring
• military missions
• Video game mods

Related Work – Polygon Reduction
• Tim Garthwaite, Jason Reposa. “Mesh
Decimation.”
• Mathieu Desbruin. “Variational Shape
Approximation.” University of Southern
California

Related Work – Indoor Scenes
• Rui Wang, David Luebke. “Efficient
Reconstruction of Indoor Scenes with
Color” Department of Computer Science,
University of Virginia
• R.T. Whitaker, J. Gregor, P.F. Chen .
“Indoor Scene Reconstruction from Sets of
Noisy Range Images” University of
Tennessee

Related Work – Laser Data Simplification
• “Urban Site Modeling from LiDAR” – Suya
You, Jinhui Hu, Ulrich Neumann, Pamela
Fox
• Designed for buildings on land, laser data
taken from aerial view, less dimensions

Our Approach – 2 Steps
• Polygon Reduction for
Corner/Edge (Feature) retention
• Further simplification of pre-processed
scenes into geometric primitives based on
indoor constraints and user-selected
features

1) Pre-processing: Polygon Reduction
• Standard Region planar
area
Growing Algorithm
• Retain neighbors near
corners & edges so
we can use corner-
finding algorithm in
next step
non-planar area
and neighbors

2) Simplification into Primitives
• Indoor Constraints:
• Standard components of indoor scene:
– floor, ceiling, walls
• User firsts selects ground, ceiling, walls
– Least squares plane fitting algorithm, points
•When creating other objects in
room, user can project to ground,
ceiling, walls, or any pre-created
object (e.g. a table)
•This means you often need
only to select points on the
frontal view of objects.

2) Simplification into Primitives
• Primitive Creation:
– User selects minimally necessary
points
– Depends on type of primitive and
available data
• Corner locating:
– If user can’t determine exact corner,
they can select a nearby corner and
let program use K-means clustering
algorithm to determine most likely
corner

Results
• 99.9% lossy “compression”
– 520,000 to 1,200 triangles
• Texturing, lighting

Extension to Outdoor/Mixed Scenes
• The user is not forced to tell the program a
ground or ceiling or walls – so they can bypass
those steps and simply use the primitive creation
tools. So you can use the program to make
simplified representations of any scenes that can
be reduced to geometric primitives or customized
preset objects.
• For example, if you’re reducing a scene with
trees, there could be two preset tree primitives –
evergreen, deciduous. Then when you see a tree
in the laser data you’d select the height, the end
of the greenery, radius, and project it to the
ground.

Future Work
• add on increased automation – like
advanced primitive fitting where the user
would have to select less points and the
program would search the surrounding
points to fit it.
• Other extensions could use intensity data
as well to help with user or program
segmentation – currently all triangles
colored the same.

Conclusion
• Fast loading and visualization
• Semiautomated method
• Built off existing graphics software