VEIL Knowledge Base

A prototype knowledge base containing domain concepts was created for use in VEIL. The type of knowledge encoded in this domain model ranged from visual objects such as roads, buildings and vehicles to abstract concepts such as groups. Particular instances of these domain concepts are linked to geometric objects in the Radius site model.

Visual Entity Hierarchy

The visual entity hierarchy shows those domain concepts that are associated with visible parts of an image.

The concepts in the visual-entity hierarchy include those that can be easily distinguished visually as well as integrating collateral information from non-image sources. An example of visually distinguishable concepts are the subconcepts of structure (i.e., bleachers, fuel-tank and building). The subconcepts of building (particularly headquarters and barracks) are not easily distinguished by visual inspection and represent the integration of other non-image information.

Group Hierarchy

The group hierarchy organizes abstract concepts concerned with the aggregation of visual objects.

The Group concept is divided into subconcepts along multiple dimensions. Some subconcepts (i.e., in-garrison and deployed-unit) are defined functionally. Other subconcepts (i.e., vehicle-group) are defined based on the type of group member. [A vehicle group is defined to be any group, all of whose members are of type vehicle] . Vehicle groups are also subdivided based on unit size (i.e., platoon, company, etc.).

VEIL Common Knowledge Base

The knowledge base used in the VEIL demonstration is available for Web browsing. This knowledge base contains the definitions that form the common domain terminology for use with the Radius Model Board 2 images. Information about the site that does not change with each image is stored in this knowledge base as well. The Loom Source for the knowledge base is also available as a text file. Note that the definitions in the knowledge base make reference to functions in the CME package that are defined elsewhere.

Links to Radius Site Model Objects

Loom instances (such as a particular building) are linked to RADIUS site model objects (such as a composite object containing three cube objects). This link allows us to write Loom relations that can reference the underlying geometric model in order to do geometric reasoning. We have implemented functions to compute the following geometric functions and linked them to Loom relations:

Using the location function, we then implemented compass directions exploiting the RADIUS coordinate system's alignment to standard geographic coordinates. The meaning of the geometric relations is defined from the point of view of a reference point. Note that there are two alternative definitions of the four primary directions (North, South, East, West). The primary definitions define half planes. Secondary definitions distinguished by the suffix "Q" define quadrants.

Information on "Near-To" not finished yet