Network virtualization for real-time processing of object detection using deep learning
- Authors
- Kim, Dae-Young; Park, Ji-Hoon; Lee, Youngchan; Kim, Seokhoon
- Issue Date
- 2021
- Publisher
- Springer Nature
- Keywords
- Network virtualization; Real-time processing; Object detection; Deep learning
- Citation
- Multimedia Tools and Applications
- Journal Title
- Multimedia Tools and Applications
- URI
- https://scholarworks.bwise.kr/sch/handle/2021.sw.sch/2222
- DOI
- 10.1007/s11042-020-09603-0
- ISSN
- 1380-7501
1573-7721
- Abstract
- These days, networked cameras are used in various applications using deep learning. In particular, as the deep learning technology for image processing develops, image-based application services using networked camera are expanding. Object detections are the representative application in the image-based applications. Images from the networked camera are transmitted to a deep learning machine, which performs object detection using a deep neural network (DNN) algorithm. For real-time processing of the object detection, lightweight of the image learning steps is needed. Both preprocessing of training sets and lightweight learning models can reduce computing loads for image learning. However, it is most important to receive video frames from the network camera without delay. In this paper, we provide a way for the learning machine to receive video frames with minimal delay. The proposed method is a kind of network virtualization for image-based object detection. It monitors network the status of available network interfaces in networked cameras. When a camera transmit video frames, the virtualized module determines the appropriate network interface to reduce delay. The performance of the proposed method is evaluated in the image-based object detection system using deep learning.
- Files in This Item
- There are no files associated with this item.
- Appears in
Collections - College of Engineering > Department of Computer Software Engineering > 1. Journal Articles
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.