![](bg1.png)
An Image Data Learning Method by Discriminating Multiple ROIs Data
Patterns for Extracting Weather Information
Jiwan Lee, Sunghoon Jung, Kijin Kim, Minhwan Kim and Bonghee Hong
Department of Electrical and Computer Engineering, Pusan National University, Busan, South Korea
Keywords:
Image Data Analysis, Weather Information, Time-series Data Analysis, Learning, Clustering, CCTV Images.
Abstract:
In order to generate weather information about rainfall and foggy visibility through analysis of CCTV images,
the analysis on the changing patterns of time-series image data is a new approach to generating weather in-
formation from CCTV images. This paper demonstrates a method to generate optimum ROIs for extracting
subtle weather image changes caused by fog and rainfall. It suggests the optimum ROI size and distance inter-
val between ROIs through experiments. Finally, a clustering-based method for extracting weather information
is proposed that has different data pattern difference between ROIs as a learning model, which is based on the
suggested optimum ROI size and interval.
1 INTRODUCTION
Weather information on dangerous roads such as rain-
fall on the forward side of the moving car while driv-
ing, precipitation and foggy visibility is highly valu-
able for the safety of drivers. This study suggests
that the extracted date pattern data based on image
changes is a viable option. Furthermore, applying pat-
tern graphs as a clustering learning model will facili-
tate the extraction of rainfall amount and foggy visi-
bility to be computed from real-time CCTV images.
The related studies that extracted weather infor-
mation from CCTV images(Beung Raul Park and
Lim, 2007; Beung Raul Park, 2007), or those that
measured foggy visibility(Bong-Keun Kim and Lee,
2008) were mainly based on discrimination of the
converted values of hue, saturation, and brightness.
However, these existing methods do not measure the
distance of foggy visibility and precipitation by an-
alyzing the time-series data of CCTV images. An-
other problem present in the existing method is the
reduced accuracy caused by the averaging of the con-
verted values for entire CCTV images, as it selects
entire CCTV images as Region Of Interest (ROI).
In order to extract time-series data changes from
CCTV images, the problem of configuring the time
interval between frames and selecting a target area
within an image for change detection should be ad-
dressed. The configuration of the time interval for
change detection may be determined through a rel-
atively easy experiment, whereas there are various
options such as an entire screen, specific ROI, and
multiple ROIs for the selection of target detection
area. Time-series data pattern appears differently de-
pending on the image characteristics influenced by
selection of target regions, in other words, based on
whether the target area is the sky, forest, or roads.
This paper suggests a method for identifying ROIs
and determining the number of ROIs that demon-
strate time-series images changes following weather
changes through real data analysis and experiments.
In this study, experiments were conducted to find out
what the optimal number of ROIs within a specific
CCTV images is. By considering different change
patterns for each weather condition concerning road
area and forest area, the size and number of optimal
ROI size and interval were determined in order to dis-
criminate slightly image change patterns.
The image data differences between multiple
ROIs result in different pattern graphs that would be
used for determining weather information. Pattern
graph changes reflecting subtle image changes be-
tween multiple ROIs were rendered into a clustering
learning model to demonstrate the discrimination of
weather information.
This paper consists of the following contents:
Chapter 2 introduces related studies. Chapter 3 ex-
plains the data used for this study and the data anal-
ysis results. Chapter 4 describes the method for se-
lecting multiple ROIs using CCTV images. Chapter
5 suggests the technique for producing weather in-
formation through clustering patterns. The Chapter
438
Lee, J., Jung, S., Kim, K., Kim, M. and Hong, B.
An Image Data Learning Method by Discriminating Multiple ROIs Data Patterns for Extracting Weather Information.
DOI: 10.5220/0006380404380443
In Proceedings of the 2nd International Conference on Internet of Things, Big Data and Security (IoTBDS 2017), pages 438-443
ISBN: 978-989-758-245-5
Copyright © 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved