Keywords: neural networks, topological models, re-engineering, construction drawings, identification of constructions parts, correction of identified part dimension.

In general, the digitising of paper-based drawings in CAD-systems is conducted manually. Important points and lines are selected manually and the related coordinates are saved to the digital drawing model. This is a very time-consuming job, which does not allow any semantic interpretations. Dosch et al [1] developed a system for graphic analysis and pattern recognition in the area of architecture with a great range of functions. Especially for the constructional engineering this system does not fulfil practical requirements, because drawing objects in construction drawings differ a lot regarding to their form, extent and orientation. Therefore in this contribution a procedure with line identification is preferred. Different methods as "orthogonal zig-zag" [2] or "spare pixel tracing" [3] are used for vectorisation in commercial systems. In this contribution a procedure with medial axis set up of the burning algorithm by Lindquist and Lee [4] is introduced. This burning algorithm is the basic foundation of different algorithm for line recognition in pixel-based drawings [5].

Unfortunately this approach based on topological information is not sufficient for the recognition of inscriptions. To overcome these difficulties of inaccuracies the inscriptions of drawings have to be interpreted, which is realized by merging the information obtained by a neural network and the topological model. Therefore a Kohonen network [6] is adapted to recognize standard lettering as well as handwriting in drawings [7]. The information gained from the Kohonen network and the topological model is merged in order to identify single characters, to combine single characters to inscriptions and finally to relate the inscriptions to dimension lines and construction parts. The gained inscription information is essential for the check-up and correction of the construction part dimensions identified by the topological line search process. Finally all recognised parts are transformed into a three-dimensional geometric model which provides all necessary geometric information for a product model.

In this contribution theoretical basics and practical applications of merging neural networks and topological models and of the re-engineering process are presented. A case study of an existing building demonstrates the usability and efficiency on the outlined approach.

The testing of the implemented software tool with a ground floor plan of a real building has shown, that the identification of construction elements and their dimensions and the transfer into a product model is generally possible. Yet hitherto the appropriate criteria for the identification have to be adjusted to the particular drawing. As neuronal networks have been shown very effective within the identification of dimension values, succeeding consideration should involve neural networks within the identification of construction elements. Though construction elements always have the same criteria of recognition, they are seldom exact identically. Therefore, the application of neuronal networks in combination with information gained from geometric and topological models shows a great promise.

Furthermore, the user interface of the software tool has to be adapted to the requirements and conditions of the every day practice. For the different input values sensible standard values have to be provided or appropriate algorithms for automatic determination of the input values have to be implemented. Particularly, dealing with recognized inconsistencies between drawing, dimension and inscription requires fine tuning of user interaction.

1

P. Dosch, K. Tombre, C. Ah-Soon, G. Masini, "A complete system for the analysis of architectural drawings", International Journal on Document Analysis and Recognition, Vol. 3 (2000), pp. 102-116 doi:10.1007/PL00010901

2

J. Chai, D. Dori, "Orthogonal zig-zag: An efficient method for extracting bars in engineering drawings", in: C. Arcelli, L.P. Cordella, G. Sanniti di Baja (eds.): Visual Form, New York, Plenum, 1992

3

F.E. Wolter, "Cut locus and medial axis in the euclidean space and on surfaces", GDV, University of Hannover, Germany. Laboratory for Engineering Man / Machine Systems, Brown University, Rhode Island, USA, 2000, www.lems.brown.edu/vision/Presentations/index.html.

4

W.B. Lindquist, S.-M. Lee, "Medial Axis Analysis of Void Structure in Threedimensional Tomographic Images of Porous Media", Journal of Geophysical Research, Vol. 101 (1996), No. B4, pp. 8297-8310 doi:10.1029/95JB03039

5

V. Berkhahn, C. Esch, "Re-Engineering of Objects in Construction Drawings", in R. Amor (ed.): Construction IT Bridging the Distance, CIB Report: Publication 284, Proc. Of CIB W78's 20th International Conference on Information Technology for Construction, Auckland, 2003.

6

J. Heaton, "Programming Neural Networks in Java", www.jeffheaton.com, 2004

7

S. Komorowski, "Rechnergestützte Bestandsaufnahme, Identifizierung vom Bauteildimensionen in papierbasierten Baubestandsplänen", Studienarbeit, Institut für Bautechnik und Holzbau and Institute of Computer Science in Civil Engineering, University of Hannover

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £80 +P&P)