Knowledge-based understanding of engineering drawings.

摘要

A knowledge-based three-phase system to achieve intelligent interpretation and comprehensive understanding of engineering drawings drawn to ANSI drafting standards has been developed for use with the automatic input of paper drawings to CAD/CAM systems and with CAD-based vision systems. Because separating object lines from dimensioning lines is a critical problem in drawing interpretation, the primary objective of this thesis is to develop techniques to identify and separate dimensioning information from object (or physical) line information. A new rule-based text/graphics segmentation algorithm to separate text in horizontal and vertical orientation from graphics, and a model-based arrowhead detection method to detect arrowheads in any orientation, have been developed. Arrowhead tracking and search methods are used to extract leaders, tails, and witness lines from segmented images containing graphics only. Text blocks and feature control frames which are extracted from the segmented images are then associated with their corresponding leaders to obtain complete dimension sets. Detection of dimension sets results in the separation of dimensioning lines from object lines. A two-stage supervised algorithm has been developed to extract section lines, and a dashed-line detection algorithm has been developed to detect centerlines and hidden lines. These methods of extracting object lines result in intelligent interpretation of geometric objects.;The ultimate objective is not only to provide compact vectorized encoding of paper drawings in electronic format, but also to generate intelligent interpretation of geometric objects. Algorithms in this three-phase system, when combined with graphics recognition and 3-D interpretation algorithms, form a complete system for intelligent interpretation of paper-based drawings from 2-D multiple views to 3-D objects.