What do I call "drawing interpretation"?

As terminology for the different levels of processing of drawings has not yet been standardized, I would like to give a short definition for what I call "drawing interpretation". The picture below shows the processing sequence I have in mind, when I talk about "drawing interpretation".

Definition: "drawing interpretation" is the recognition of domain specific objects and meaningful relations between these objects in a drawing.

In this definition "drawing interpretation" starts with graphical primitives (which are the result of vectorization), it is capable of recognizing objects of the application domain as well as geometric inducted relations. Operations that go beyond this kind of geometric reasoning are assumed to build a separate processing step (utilization).

Why should "drawing interpretation" be adaptable?

Drawings are an important source of information in engineering domains like mechanical engineering, electronic circuit design, architecture, and cartography. If we referred to these drawings as unassembled collections of line and arc segments only, we would certainly miss the meaning which the drawer intended to imply. Meaningful entities of drawings are mainly symbols, object contours, connections, and geometric relations between the afore mentioned entities. Therefore, today's CAD-Systems have changed from a line-based design process to a feature-based modelling of domain objects. Despite of this convenient new design paradigm, there might still be many old drawings in the archives that haven't reached this high level of abstraction, e.g. paper drawings and old CAD drawings. To incorporate these "flat" drawings into modern design processes, the lines and arcs will have to be categorized into symbols and meaningful geometric relations.

Considering commercial tools for processing of paper drawings, there are a couple of solutions for scanning and vectorization available. Vectorization extracts geometric primitives (lines, circular arcs, filled areas, circles, arrows, etc.) from scanned raster drawings. As the current vectorization techniques are mostly domain-independent, vectorization tools can be used for all kinds of drawings. Although these vectorization techniques seem to be comparatively economical solutions (offering one tool for all domains) a more sophisticated interpretation of the primitives will unavoidably require the use of domain knowledge. Unfortunately, the symbols and meaningful geometric relations are varying significantly between domains as different as for example mechanical engineering and architecture. What adds to this variety is the great discrepancy between the drafting conventions by different companies. Therefore, today's lack of commercial available interpretation tools can be explained by the basic problem that each installation has to be adapted to the symbols and graphic conventions of the customer. I feel that such interpretation tools will be seen on the market, when the ease and speed of the tool's adaption will reach a certain level.

A knowledge-based approach to "drawing interpretation"

Aiming at a general solution to the generic problem of drawing interpretation, we developed a prototype of an adaptable drawing interpretation kernel (ADIK), that can be specialized to a domain-specific categorization tool by a declarative knowledge base (see the list of selected publications ). In principle, ADIK can be characterized by three major processing steps. At first, ADIK reads the graphical primitives (lines, arcs, etc.) delivered by a vectorization tool. Then, ADIK compares the primitives with the knowledge base, that contains the descriptions of domain-specific symbols and graphic conventions. At last, the recognized objects and the primitives that build up the objects are stored for further processing in CAD systems or drawing databases.

ADIK categorizes a drawing by terms of a domain-independent kernel and declarative drawing specifications. The Figure below gives an overview of the main components of the adaptable drawing interpretation kernel ADIK and its interaction with a domain-specific knowledge base.

ADIK provides a set of predefined geometric features and specification elements. The knowledge base contains descriptions of graphical objects and geometric rules of the target domain based on these elements. ADIK uses these descriptions to build generic objects and to relate them in a decomposition hierarchy as specified in the knowledge base. The geometric specification of each graphical object in the knowledge base is compiled into a set of constraint functions which are attached to the corresponding generic objects. Finally, the taxonomic and decompositional relations between the graphical objects are used to determine a control strategy for drawing interpretation.