Recovery of industrial objects from single perspective line drawings: A rule-based approach

Date of Award




Degree Name

Doctor of Philosophy (Ph.D.)


Electrical and Computer Engineering

First Committee Member

Tzay Y. Young, Committee Chair


This dissertation details the design and implementation of a rule-based expert system. The expert system performs three-dimensional shape recovery and orientation estimation of an object from a single perspective view, by detecting geometrical regularities in the object. The primary input to the expert system is a set of first and second order line segments, commonly referred to as a line drawing. These line segments constitute the perspective projection of the object to be reconstructed. The system assumes a man-made world abundant in geometric regularities. These man-made objects are assumed to be made up of strictly trihedrons of planar surfaces, circles, spheres, cylinders, and right circular cones. The output of the expert system is the three-dimensional wire-frame model of the object which most likely corresponds to the line drawing. The three-dimensional wire-frame models produced by the system are illustrated by displaying the orthographic projections of the front, top, and side views of the object in the camera (observer's) coordinates as well as the perspective view(s) of the object's wire-frame model (including recovered hidden lines). This rule-based expert system relies solely on geometrical considerations.The overall system consists of three major modules: the low-level preprocessing stage, the expert itself, and the graphics display routines. The expert system is rule-based, adopting forward chaining methodology, and is written in OPS5 in conjunction with Pascal and Fortran supporting routines on a VAX/VMS system. The expert part of the system contains heuristic rules which detect object regularities such as parallelism and perpendicularity of edges, right corners and parallelism of such corners, face symmetries, circles, spheres, cylinders, and right circular cones. There are also rules for recovering the focal length of the camera as well as hidden lines. Such regularities are then utilized to reconstruct the three-dimensional wire-frame model in object (world) coordinates. Then, the graphics routines display the top, front, side, and perspective wire-frame views of the object on a raster-scan CRT. The system assumes very little a priori information and associates a certainty factor to the final three-dimensional object interpretation.


Engineering, Electronics and Electrical; Computer Science

Link to Full Text