Structural Recognition of Hand Drawn Floor Plans

Josep Lladós, Enric Martí
Unitat de Processament d'Imatges i Intel.ligència Artificial
Departament d'Informàtica - Edifici C - Universitat Autònoma de Barcelona.
08193 Bellaterra (Barcelona)
e-mail: josep@upisun1.uab.es, enric@upisun1.uab.es

Abstract:

A system to recognize hand drawn architectural drawings in a CAD environment has been developed. In this paper we focus on its high level interpretation module. To interpret a floor plan, the system must identify several building elements, whose description is stored in a library of patterns, as well as their spatial relationships. We propose a structural approach based on subgraph isomorphism techniques to obtain a high-level interpretation of the document. The vectorized input document and the patterns to be recognized are represented by attributed graphs. Discrete relaxation techniques (AC4 algorithm) have been applied to develop the matching algorithm. The process has been divided in three steps: node labeling, local consistency and global consistency verification. The hand drawn creation causes disturbed line drawings with several accuracy errors, which must be taken into account. Here we have identified them and the AC4 algorithm has been adapted to manage them.
 


Get a postscript version of the paper