Desenvolvimento de uma solução de visão computacional para contagem de garrafas tampadas em uma linha de produção fabril.

Abstract

This final project presents the development of a computer vision solution for auto mated counting of capped bottles in an industrial beverage production line. The proposal aims to replace traditional methods, such as those based on optical sensors, promoting greater efficiency and automation in the counting process. In this context, the proposed system consists of a Logitech C920s webcam and an NVIDIA Jetson Nano, combined with color segmentation and thresholding techniques to process images in real-time, with an optimized region of interest to focus exclusively on relevant objects. Validated in a factory environment, the prototype demonstrated efficient performance, achieving high precision and recall, with an F1-score of 99.8%. The main contribution of this project lies in demonstrating the feasibility of integrating accessible and scalable computer vision technologies into manufacturing environments, improving quality control and operational efficiency while reducing errors and waste. This solution provides an approach to indus trial automation and paves the way for future applications and improvements in various manufacturing sectors.