Sistema de recuperação de informação visual em desenhos técnicos de engenharia e arquitetura: modelo conceitual, esquema de classificação e protótipo

Abstract

The objective of this thesis was the development of a conceptual model, a classification scheme and a prototype for the efficient organization and retrieval of engineering technical drawings. The need and difficulty to retrieve technical drawings more efficiently was the motivation to develop this research which can result in better services and decision making in areas such as infrastructure (sanitation, energy and road), security, health and education. The study resulted in a system that integrates concepts and techniques of information science and computer science, joining human interpretation with automatic processing and using textual and visual metadata. Prioritizing image search in a database of technical drawings, the method answers questions such as: how many fire extinguisher are there in a given place? The literature review presents concepts and techniques of organization and information retrieval of images in information science, computer science and engineering/architecture. The state of the art indicates a gap in the literature with the presence of two distinct lines of research, in which information science studies image information retrieval based in textual descriptions while in computer science the research focuses in content-based image retrieval. The method developed starts with the interpretation of the technical drawing by an indexer that extracts textual metadata and classifies the drawing according to three categories (type, process and form) which lead to a specific table of visual metadata (icons) that are automatically located, indexed and stored. Technical drawing files and their textual and visual metadata constitute the database. The user retrieves the information through the definition of the three categories (type, process and form) and the selection of a key-image, for example, the icon of a fire extinguisher. The system retrieves all the drawings that contain the key-image and presents them through the interface, highlighting the hits. To validate and verify the system, a case study at the Corpo de Bombeiros Militar de Minas Gerais was conducted and a corpus of 332 technical drawings of security projects was used. The research results shows the time of 6.2 seconds, used by an indexer to interpret, to classify and to index the technical drawing. By using the model, an 84% reduction is occurs in the processing time proving the efficiency of the integration human and computer processing. The 98.7% drawing retrieval rate through key-image definition proves the efficiency of the system. Among the theoretical and methodological contributions, is the union of two areas of knowledge which made feasible an image retrieval system based on a classification scheme, on the of key-images, the pre-processing of the drawing and the creation of a method of development of the prototype that involved an iterative process between theory and practice. The prototype is a practical contribution to the recovery of technical drawings in any field of engineering. The model developed can be adapted to other situations involving need for retrieval of images. Examples of applications for Corpo de Bombeiros Militar de Minas Gerais are greater efficiency to calculate the number of fire extinguishers, calculate the volume of water reserve, find the position of extinguishers, trace escape routes, instruct the staff during dislocation and show the strategic positioning of equipment thus achieving more agility in responding to emergency calls and increasing the possibility to find live people in fire accidents.