Publication detail
Estimations of Shape and Direction of an Air Jet Using Neural Networks.
RICHTER, J. ŠŤASTNÝ, J. JEDELSKÝ, J.
Czech title
Estimations of Shape and Direction of an Air Jet Using Neural Networks.
English title
Estimations of Shape and Direction of an Air Jet Using Neural Networks.
Type
conference paper
Language
en
Original abstract
Analysis of airflow properties is an important step during validation of functionality of air distribution systems in a closed environment such as vents in car cabins. Optical visualization methods, based on imaging of the airflow visualization using smoke or fog, are often applied in such cases. The aim of this work is in an automation of processing of such images captured during visualization. It can be accomplished, besides special mathematical methods, using neural networks. We have employed a multilayer perceptron network for a detection of fog-containing areas in airflow images. Network learning was used and documented here for a recognition of the fog presence in individual pixels of the image based on colour intensities of the pixel neighbourhood. The fog detection was used for estimation of the jet shape. Hopfield network, which allows to relate the jet with one of the four basic flow directions, was applied consequently. The information about jet direction is important for further image processing functions that take into account the distance from the jet source.
Czech abstract
Analysis of airflow properties is an important step during validation of functionality of air distribution systems in a closed environment such as vents in car cabins. Optical visualization methods, based on imaging of the airflow visualization using smoke or fog, are often applied in such cases. The aim of this work is in an automation of processing of such images captured during visualization. It can be accomplished, besides special mathematical methods, using neural networks. We have employed a multilayer perceptron network for a detection of fog-containing areas in airflow images. Network learning was used and documented here for a recognition of the fog presence in individual pixels of the image based on colour intensities of the pixel neighbourhood. The fog detection was used for estimation of the jet shape. Hopfield network, which allows to relate the jet with one of the four basic flow directions, was applied consequently. The information about jet direction is important for further image processing functions that take into account the distance from the jet source.
English abstract
Analysis of airflow properties is an important step during validation of functionality of air distribution systems in a closed environment such as vents in car cabins. Optical visualization methods, based on imaging of the airflow visualization using smoke or fog, are often applied in such cases. The aim of this work is in an automation of processing of such images captured during visualization. It can be accomplished, besides special mathematical methods, using neural networks. We have employed a multilayer perceptron network for a detection of fog-containing areas in airflow images. Network learning was used and documented here for a recognition of the fog presence in individual pixels of the image based on colour intensities of the pixel neighbourhood. The fog detection was used for estimation of the jet shape. Hopfield network, which allows to relate the jet with one of the four basic flow directions, was applied consequently. The information about jet direction is important for further image processing functions that take into account the distance from the jet source.
Keywords in English
Neural network, perceptron, Hopfield network, fog detection, direction detection
RIV year
2013
Released
26.06.2013
Location
Brno
ISBN
978-80-214-4755-4
Book
MENDEL 2013
Pages from–to
221–226
Pages count
6
BIBTEX
@inproceedings{BUT101457,
author="Jan {Richter} and Jiří {Šťastný} and Jan {Jedelský},
title="Estimations of Shape and Direction of an Air Jet Using Neural Networks.",
booktitle="MENDEL 2013",
year="2013",
month="June",
pages="221--226",
address="Brno",
isbn="978-80-214-4755-4"
}