Publication Details
A Study of Real-time Computer Vision Tasks in 5G-enhanced Environment
Klepárník Petr, Ing., Ph.D. (DCGM)
Kapinus Michal, Ing., Ph.D. (DCGM)
Dobeš Petr, Ing. (DCGM)
Smrž Pavel, doc. RNDr., Ph.D. (DCGM)
Computer vision is an integral part of many robotic applications which are often executed on limited hardware. In this paper, we study applications which execute the vision tasks remotely in a cloud or edge environment instead of local execution. Such a structure of applications (called Network Application within the 5G-ERA project) helps to overcome many limitations of robotic platforms stemming from less powerful hardware. The one requirement is that each robot must have a sufficient network connection. In the future, we believe, this will be solved easily using new generations of networks which will offer low latencies and high bandwidth for connected clients. In this paper, we discuss the particular case of vision tasks and show their properties in a use case of visual collision warning and train detection systems for autonomous robots. In particular, we compare local execution with the execution in a cloud-based system using the automatically deployed container. The Network Application along with experimental data is available as an open source on GitHub.
@inproceedings{BUT188927,
author="Roman {Juránek} and Petr {Klepárník} and Michal {Kapinus} and Petr {Dobeš} and Pavel {Smrž}",
title="A Study of Real-time Computer Vision Tasks in 5G-enhanced Environment",
booktitle="EuCNC & 6G Summit Proceedings",
year="2023",
pages="1--5",
address="Gothenburg",
url="https://www.fit.vut.cz/research/publication/13233/"
}