| Paper drawings are difficult to update, and carry higher costs in storage, distribution and revision than their corresponding CAD files. It is necessary to convert them into suitable computer formatted files in order to improve the information assets of many companies. In this work, a system for the automatic interpretation of mechanical engineering drawings has been designed, implemented and tested. The proposed system is a prototype that attempts to solve the conversion problem by using a modular architecture in which each module extracts a particular primitive present in the Singular-runs Graph Representation (SGR) of the engineering drawing. The SGR was developed as an alternative data structure for the conversion of drawings. It is a compressed version of the drawing and it can be computed directly from the run-length representation of the binary image. Primitive recognition algorithms based on this data structure are faster than other approaches as it was seen during the performance evaluation phase of the vectorization algorithm based on the SGR that was developed as part of this research. The modular architecture of the conversion system and its use of an ontology for the representation of knowledge about the mechanical engineering drawings make the system easy to scale up to add more complex recognition capabilities. This was seen through the a posteriori integration of a module for the recognition of schematic representation of screws in the drawings. Finally, the SGR-based vectorization algorithm, and the modules of the system were tested following a rigorous experimental protocol to characterize its performance. |
