Detección de pose usando Deep Learning en ambientes industriales
Date
2022-12
Journal Title
Journal ISSN
Volume Title
Publisher
AJEA- Actas de Jornadas y Eventos Académicos de UTN
Abstract
Cada vez es más común aplicar visión artificial en distintos entornos industriales, para establecer la ubicación, forma y calidad de los objetos. En trabajos anteriores se analizaron algoritmos de machine learning que permitieron detectar objetos y su presentación (frente/dorso). La presente investigación analiza la aplicación de una red convolucional profunda conocida como U-Net, para la segmentación de imágenes. El objetivo es lograr obtener la pose de los objetos, completando así la información de pose (ubicación) para utilizar métodos de bin picking (recogida aleatoria de contenedores) que completen el proceso de embalaje del objeto.
It is becoming more and more common to apply machine vision in different application domains, to establish the location, quality, formation of objects. In previous works, machine learning algorithms were analyzed to detect objects and their presentation (front/back). The present research analyzes the application of a deep convolutional network defined under the U-Net algorithm, for image segmentation. The objective is to obtain the pose of the objects, thus completing the pose information (location) in order to use bin picking methods to complete the object packing process.
It is becoming more and more common to apply machine vision in different application domains, to establish the location, quality, formation of objects. In previous works, machine learning algorithms were analyzed to detect objects and their presentation (front/back). The present research analyzes the application of a deep convolutional network defined under the U-Net algorithm, for image segmentation. The objective is to obtain the pose of the objects, thus completing the pose information (location) in order to use bin picking methods to complete the object packing process.
Description
Keywords
redes convolucionales, pose, U-Net, convolutional networks
Citation
Jornadas de Ciencia y Tecnología 2022
Collections
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwised noted, this item's license is described as openAccess