Resumo:
Modeling and Simulation is one of the most widely used methods in the field of Operations Research. It has been evolving daily, moving from models with a limited lifespan and that do not exchange data regularly with the real world, to more dynamic, constantly used approaches that allow real-time or near real-time data exchange between real and virtual systems. This new type of model, with continuous use and intense data exchange between the real and virtual worlds, is called a Digital Twin. This approach has been used most extensively in the manufacturing sector, however, its use has expanded to other highly relevant sectors of the economy, such as the retail sector. Several operations within this sector can benefit from the application of Digital Twins, especially when combined with Artificial Intelligence-based techniques, such as Computer Vision. Inventory control, for example, stands out as one of the most fundamental and strategic operations in this sector, but it has rarely benefited from modern approaches such as Digital Twins and Computer Vision to provide more automated and data-driven support to its stakeholders. To contribute to the literature and fill this gap, this work aims to present a computational model to support shelf inventory management in retail establishments, integrating modern Computer Vision techniques with the dynamism provided by Digital Twins. It seeks to map, monitor, and predict product consumption in establishments by calculating an emptying rate that takes into account empty spaces (gaps) found on their shelves. Furthermore, it aims to integrate free software tools and easily accessible devices (cameras), being modular, expandable, and adaptable to the specific products of each establishment. Practical tests conducted sought to demonstrate the model's robustness in different scenarios with variations in zoom, lighting, and layouts, its ability to detect human occlusions at different levels and movements, and its practical utility in consumption and replenishment scenarios by offering both descriptive and predictive feedback in near-real time, thus acting as a data-driven alternative to aid decision-making.