Reconhecimento de caracteres estampados em tarugos de aço

Abstract

This dissertation presents the implementation and analysis of tools that aims to achieve characters recognition. This tool promotes discussions on its viability application of a real problem at Gerdau Açominas. The problem consists in the identification of the Production Order number (PO) stamped on steel objects (billets). The recognition goal is the complete automation of the BIL (Billets Inspection Line). Achieving theidentification of the PO, the recognition system must be capable of modifying the parametric variables of the inspection line in order to separate the billets for each customer, preventing problems as, for example, mixture of parts and inspections with wrong parameters. The study begins considering a necessary conditioning and components (hardware) to make possible the gathering of information (images) in aproper manner. The collection of images is made inside the limitations imposed by the process and viable improvements in accordance with the context are suggested. Segmentation, feature extraction and lassification algorithms are analyzed. In the face of new challenges that it presents, alternative approaches and adjustments aiming an optimized performance of the strategies are suggested. The reason of commitment between efficiency and time of processing are discussed, since image processing methods tend to have high computational cost. Diverse collected images of the real process and the performance of each stage of processing are analyzed. Results of the methodologies turned out unsatisfactory; however a discussion is carried through regarding the quality of the image and the format of the characters that are stamping the billets. Critical points intending the change in the process are mentioned and modifications in the strategies are suggested in order to better fit the problem.