Comparação de algoritmos de visão computacional tradicional e aprendizado de máquina aplicados na automatização de uma mesa de pinball
Loading...
Date
2023-06
Authors
relationships.isAdvisorOf
Ahlert, Edson Moacir
relationships.isRefereeOf
Borba, Fabricio Hartmann
Meyer, Vinicius
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
O aprendizado de máquina já é realidade a muito tempo na criação de ferramentas de visão computacional e recentemente tem se popularizado com a criação de carros autônomos que utilizam redes neurais convolucionais (RNC) para realizar a classificação e detecção de objetos em imagens, porém isso não necessariamente significa que as técnicas de visão computacional tradicionais se tornaram obsoletas e foram substituídas pelas RNC. Neste trabalho serão aplicados 5 métodos na tarefa de automatização de uma mesa de pinball, sendo 3 deles métodos tradicionais que são algoritmos matemáticos como transformada de Hough, diferença absoluta e segmentação por cor. E outros 2 métodos de aprendizado de máquina adaptados ao seu cenário de hardware com a biblioteca YOLO que utiliza RNC para ler e extrair informações de imagens com a proposta de ter alta velocidade de processamento sem abrir muita mão da precisão e da acurácia. Nos resultados deste trabalho é possível entender melhor as vantagens e desvantagens de cada abordagem por meio dos resultados de precisão, acurácia e tempo de processamento de cada solução.
Machine learning has been a reality for a long time in the creation of computer vision tools and has recently become popular with the creation of autonomous cars that use convolutional neural networks (CNN) to perform the classification and detection of objects in images, but this does not necessarily means that traditional computer vision techniques have become obsolete and have been replaced by CNN. In this work, 5 methods will be applied in the task of automating a pinball table, 3 of them being traditional methods that are mathematical algorithms such as Hough transform, absolute difference and color segmentation. And 2 other machine learning methods adapted to your hardware scenario with the YOLO library that uses RNC to read and extract information from images with the proposal to have high processing speed without giving up much precision and accuracy. In the results of this work it is possible to better understand the advantages and disadvantages of each approach through the results of precision, accuracy and processing time of each solution.
Machine learning has been a reality for a long time in the creation of computer vision tools and has recently become popular with the creation of autonomous cars that use convolutional neural networks (CNN) to perform the classification and detection of objects in images, but this does not necessarily means that traditional computer vision techniques have become obsolete and have been replaced by CNN. In this work, 5 methods will be applied in the task of automating a pinball table, 3 of them being traditional methods that are mathematical algorithms such as Hough transform, absolute difference and color segmentation. And 2 other machine learning methods adapted to your hardware scenario with the YOLO library that uses RNC to read and extract information from images with the proposal to have high processing speed without giving up much precision and accuracy. In the results of this work it is possible to better understand the advantages and disadvantages of each approach through the results of precision, accuracy and processing time of each solution.
Description
Keywords
Visão computacional; Aprendizado de máquina; Métodos tradicionais; Computer vision; Machine learning; Traditional methods
Citation
ASSIS, Lucas De. Comparação de algoritmos de visão computacional tradicional e aprendizado de máquina aplicados na automatização de uma mesa de pinball. 2023. Monografia (Graduação em Engenharia de Software) – Universidade do Vale do Taquari - Univates, Lajeado, 30 jun. 2023. Disponível em: http://hdl.handle.net/10737/3696.