Machine Drawing Understanding System (MDUS) - Solid Reconstruction from Engineering Drawings

Oct 23, 2026

Machine drawing reconstruction explains how orthogonal multiview projections can be interpreted to recover a 3D object structure from 2D line drawings. Prior work addresses solid reconstruction and “fleshing out projections,” often using constraint-based reasoning on geometry primitives. The text highlights a major limitation: real engineering drawings include annotation entities (dimensions, tolerances, specifications) that are treated as noise by conventional algorithms. A Machine Drawing Understanding System (MDUS) separates geometry from annotations and uses annotation information to enhance reconstruction, enabling practical processing of paper and plotted drawings and supporting information exchange contexts like IGES.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Solid reconstruction from orthogonal multiview drawings

Interpretation and “fleshing out projections”

Constraint-based reconstruction approaches

Key limitation: annotation entities treated as noise

MDUS preprocessing to separate geometry and annotation

Dimension-sets and drafting standards hierarchy

Product data exchange context (IGES)

A machine drawing consists of a description of one or moreprincipal orthogonal views(projections)of an object.Theseprojections can be used to reconstruct the 3D structure of theobject in a variety of ways.The problem of solid reconstructionfrom engineering drawings deals with a subclass of the classof multiview line drawings,where the object world is notrestricted to a finite set of prototypes.The interpretation ofmultiview line drawings has been investigated quite intensively [1,7,8,12,13,14,15,16,18].For specific wireframe projec-tions this is known as the“fleshing out projections”problem[19].Gigus and Malik [8]have developed an algorithm forcomputing the aspect graph for line drawings of polyhedralobjects.They provide several references for surveys of currentapproaches for solving the 3D object recognition problem.Sugihara[18]presents a computational mechanism for theinterpretation of line drawings by enabling a machine toreconstruct 3D object structures from their pictures drawn ona 2D plane.The objects he considers are polyhedra,and theline drawings are single-view objects bounded by planar faces.One potential application suggested by Sugihara for hismechanism is flexible human-machine communication.Sinceit is tedious work for designers to convert their thoughts intonumerical forms,all they will have to do is draw pictures oftheir designed objects and give a small number of additionaldata,such as lengths of edges and angles between faces.Thequestion of how this data is to be provided is not discussed. The subclass of line drawings weare concerned with,mechanicalengineering drawings,usually con-sists of top,front and side views.Various approaches to the recon-struction problem have been pub-lished in the literature.Broadly,they can all be viewed as applica-tions of constraints to various 2Dentities.Wesley and Markowski[19]have developed an algorithm tofind all solid polyhedral objectswith a given set of two dimensionalprojections.These projections maycontain depth information in theform of dashed and solid lines,mayrepresent cross sections,and maybe overall or detailed views.Haralick and Queeney [9]applyconstraints to the 2D primitive enti-ties,which are the lines on the 2Dprojections.Aldefeld[1]starts fromclosed loops of lines and appliesrules or constraints to them.Thisapproach,although less general,may frequently be practical.Preiss[12]also casts the problem into oneof constraint propagation.Hs con-trol structure arrives either at aconsistent constraint net or at a conclusion that there is no consis-tent net,thus enabling a certain val-idation of the set of 2D projections.This algorithm handles also cylin-drical faces,which are not treatedin [19]. As noted [19],quite apart fromthe mathematical interest of these“fleshing out projections”algo-rithms,they provide a basis for au-tomatic conversion of a set of 2Dprojections typically found in me-chanical engineering drawings intosolid volumetric representations ofobjects.There is,however,a majorproblem with these and other simi-lar algorithms that currently limitstheir applicability.The algorithmsassume that there exists a well-defined,noise-free set of 2D pro-jections,readily available to be usedas input.Thus they overlook thepresence of annotation entities inpractically all engineering draw-ings.These entities are aimed atproviding the drawing reader withinformation about the object whichcannot be expressed by its geometryentities.While geometry lines de-scribe contours of facets in the ob-ject's projections,annotation addssuch information as dimensionsand tolerances,production specifi-cations,administrative instructionsand so forth.Reconstruction algo-rithms perceive this type of infor-mation as ‘noise.'While obtaining a'noise-free'set of projections maybe relatively easy for models of ob-jects whose projections were origi-nally created using a CAD/CAMsystem,it is far from being trivialwhen the source is a paper drawing,produced either manually or by acomputer dri

FAQs

What does machine drawing reconstruction aim to achieve from 2D drawings?

It reconstructs the 3D structure of an object from one or more orthogonal projections contained in multiview line drawings.

Why do many existing reconstruction algorithms have limited applicability to real engineering drawings?

They assume a noise-free set of 2D projections, but practical engineering drawings include annotation entities like dimensions and tolerances that algorithms often misinterpret as noise.

What role does the Machine Drawing Understanding System (MDUS) play?

MDUS performs preprocessing to separate geometry entities from annotation entities and uses annotation information to enhance the geometric reconstruction before removing annotations.

Oct 23, 2026

Upload by I feel Different