Traceability is of vital importance in industrial processes, allowing for improved quality procedures and efficient defect processing. Optical Character Recognition (OCR) represents a critical step in automatizing traceability through ID monitoring.
Throughout this text, a case was made for the applicability of Yolo in OCR for Metal Stamped Character IDs. The argument was supported by a comprehensive literature review and by an experiment, in which a word accuracy of 99.3% was accomplished across a test set of 600 images, a relevant metric when compared to other work.
Once the suitability of Yolo for this task was analyzed, another experiment was run aiming to assess the accuracy of a Yolo-based pipeline across data with variable quantity and quality. It was concluded that these two features are of great importance to the performance of a Yolo-based pipeline, since accuracy decreased notably when treating with scarce and variable data. Some suggestions for partially solving this problem were the standardization of image acquisition techniques and/or the application of very strong preprocessing.