Authors
Young Im Cho*, Akmaljon Palvanov
Department of Computer Engineering, Gachon University, 1342 Seongnam-daero,
Sujeong-gu, Seongnam-si, Gyeonggi-do, South Korea
*Corresponding author. Email: [email protected]
Corresponding Author
Young Im Cho
Received 31 October 2018, Accepted 15 December 2018, Available Online 25
June 2019.
DOI
https://doi.org/10.2991/jrnal.k.190531.003
Keywords
Atmospheric visibility; convolutional neural networks; CCTV; graphic user
interface; recognition
Abstract
Due to the recent improvement in computer performance and computational
tools, deep convolutional neural networks (CNNs) have been established
as powerful class of models in various problems such as image classification,
recognition, and object detection. In this study, we address two fundamentally
dissimilar classification tasks: (i) visibility estimation and (ii) food
recognition on a basis of CNNs. For each task, we propose two different
data-driven approaches focusing on to reduce computation time and cost.
Both models use camera imagery as inputs and works in real-time. The first
proposed method is designed to estimate visibility using our new collected
dataset, which consist of Closed-circuit Television (CCTV) camera images
captured in various weather conditions, especially in dense fog and low-cloud.
Unlikely, the second model designed to recognize dishes using artificially
generated images. We collected a limited number of images from the web
and artificially extended the dataset using data augmentation techniques
for boosting the performance of the model. Both purposing models show high
classification accuracy, requiring less computation power and time. This
paper describes the complexity of both tasks and also other essential details.
Copyright
© 2019 The Authors. Published by ALife Robotics Corp. Ltd.
Open Access
This is an open access article distributed under the CC BY-NC 4.0 license
(http://creativecommons.org/licenses/by-nc/4.0/).