Merging neural networks and topological models to re-engineer construction drawings

In this paper an approach is presented to digitize a drawing, to build up geometric and topologic models, to recognise construction parts and to interpret dimension lines and inscriptions. All recognized parts are transformed into a three-dimensional geometric model which provides all necessary geometric information for a product model. The recognition process of construction parts is based on a line search and topological analysis, which are not suitable for the recognition of drawing inscriptions and hand writings. Because dimension inscriptions deliver significant information about the dimensions of construction parts, a neural Kohonen network is implemented and adapted in order to recognise inscription text. Finally the gained information about dimensions is related to significant details of con-