Self-Structural Syntax-Directed Pattern Recognition of Dimensioning Components in Engineering Drawings

A mechanical engineering drawing is a set of orthogonal 2D views of a 3D object. To establish vision-based communication among CAD/CAM systems and to implement 3D reconstruction, drawings need to be preprocessed to exploit the information conveyed by the annotation before separating it from the geometry. A rationale for a machine drawing understanding system is first argued. A context-free dimensioning grammar provides a basis for a self-structural, syntax-directed pattern recognition scheme. It employs learning the characteristic parameters of arrowheads and text from a detected sample, and using them to detect the entire population. Syntactic considerations that significantly decrease the search space help predict the spatial location and orientation of potential components. The underlying ideas of the scheme may be applicable to a broad scope of tasks involving intelligent recognition.