Transcript presentation for the paper

Building Knowledge about Buildings
Matt Young and Eyal Amir
University of Illinois, Urbana-Champaign
The Problem
• Need a way to represent
information about
buildings.
• A wealth of information
exists in floor plans, but
what information do we
need? How to encode
it?
Our Goals
• A general framework for
representing buildings
which is:
• Simple enough to add
data quickly/automatically.
• Complete enough to
accurately represent the
structure of a building.
• Able to answer queries
regarding that structure.
Overview
•
•
•
•
•
Previous Work
Overview of Our Language
Comparison with Current Technology
Other Applications
Future Work
Previous Work - Cyc
• Contains a “Building” constant, defined as “A
specialization of both FixedStructure and
HumanShelterConstruction.”
• By following assertions through the hierarchy,
we can learn certain information about a
building such as what rooms it contains, how
many levels it has, etc.
• However, there is no structured presentation of
how things are connected together, how the
building is actually constructed.
Previous Work - IFC Data Model
• International standard for
architectural firms, CAD
developers, and
construction companies.
• Very detailed information
about building construction.
• However, also contains a
great deal of information
about processes, analysis,
CAD data, etc.
• Also, it is inconsistently
implemented.
Our Solution
• A language designed specifically to capture
only the structure of a building.
• Encoded as an ontology in OWL DL, for ease of
use with the Semantic Web, and (hopefully) full
decidability on inference.
Language – General Classes
• Classes define different
features of a Building.
• Four main classes
•
•
•
•
Building
External_Feature
Internal_Feature
Material
• Subclasses define
distinct feature types.
Language - Properties
• Properties define
relations between
features.
• Most are defined
symmetrically, for strong
connectedness.
Language - Assertions
• Assertions enforce proper construction of buildings.
• Ensure that certain properties must be filled with some
value (or possibly more than one value).
• There are no value restrictions.
Language - Specialization
• Language can be extended
with subclasses of the
general classes define to
subtypes of each feature.
e.g. House is a subtype of Building,
Bedroom is a subtype of Room.
• Subtypes are defined by
additional restrictions, some
of which may be value
restrictions.
• Subtypes can also be
inferred, but this slows down
search considerably.
Language - Limitations
• No spatial information (size, shape).
• No information about environment surrounding
building.
• Some features are difficult to encode:
• Features serving multiple purposes (e.g. A roof also
serving as a wall, such as in an A-frame).
• Features which are both external and internal.
Comparison with Current Technology
Architectural Feature
Houseplans Our language
.com
# of Bedrooms/Baths
Yes
Yes
# of Floors
Yes
Yes
Includes certain room type
Yes
Yes
Square footage
Yes
No
Has X room on Y floor
No
Yes
Has X room connected to Y room
No
Yes
# of exits from each room
No
Yes
Other Applications
• Find paths out of buildings (fire escapes).
• Complete a building floor plan given a partial
encoding of the building.
• Use a knowledge base encoded in this
language to categorize buildings given partial
information about them.
Future Work
• Adding spatial information without losing
decidability.
• Adding encoding for surrounding environment
and for objects within the building to create a
full virtual world space.
• Encoding data automatically from floor plans or
IFC models.
Conclusion
• Special thanks to Eyal for all his help and
guidance.
• Questions / Comments ?