![](bg1.png)
A Method of Weather Recognition based on Outdoor Images
Qian Li, Yi Kong and Shi-ming Xia
College of Meteorology and Oceanography, PLA University of Science and Technology, NanJing, China
Keywords: Outdoor Image, Weather Recognition, Power Spectrum Slop, SVM, Decision Tree.
Abstract: To improve the quality of video surveillance in outdoor and automatic acquire of the weather situations, a
method to recognize weather phenomenon based on outdoor images is presented. There are three features of
our method: firstly, the features, such as the power spectrum slope, contrast, noise and saturation and so on
are extracted, after analysing the effect of weather situations on image; secondly, a decision tree is
constructed in accordance with the distance between the features; thirdly, when every SVM classifier on the
non-leaf node of the decision tree is constructed, some features are selected by assigning the weight. The
experiment results prove that the proposed method can effectively recognize the weather situations in
outdoor.
1 INTRODUCTION
In many outdoor applications for computer vision,
the “bad” weather situations, such as haze, fog, rain,
hail and snow, are involved. And it is urgent to
detect and recognize the various outdoor weather
situations, especially the severe ones. Meanwhile,
the observation of weather situations in meteorology
is still mainly rely on manual, and weather situation
is not exactly the same even within every small
region. Therefore, automatic recognition of the
outdoor weather situation based on image or video
data gets more extensive attention in recent years.
According to the duration and extent of
influence to the video or image, weather situations
can be divided into static or steady weather
situations category and dynamic weather situations
category (Garg, 2004).In Static weather situations
such as sunny, cloudy, fog, smoke, haze and so on,
there is some or more stable particles in the
atmosphere to attenuate and refract the ambient
light, so the impact on image quality of these
phenomena is relatively more stable, mainly for the
blur degradation. Dynamic weather situations, such
as rain, snow, dust storm, hail and so on, make
ambient light attenuation and refraction for the
movement of unstable particles in the atmosphere,
and the image quality degradations caused in these
situations are mainly motion blur, point noise and
movement trace noise. Because of the differences in
imaging process, for example, the influence of the
size of rain and snow, the degradation effect will be
different. So identifying and studying different
dynamic weather phenomena in different
environments and situations is one of the difficulties
in current research.
This paper presents an approach to identify and
classification of weather situations to use existing
surveillance cameras to improve the recognition rate
of outdoor image and resolve the problem of
automatic weather observation. We construct
classifiers with the structure of decision tree by
features extracted from the sample outdoor images
and acquire accurate weather situations classification
results to the images captured by video camera.
2 OUR METHODS OVERVIEW
Weather recognition is a brand-new subject and only
a few of previous work has addressed this issue.
Narasimhan (Narasimhan, 2002; Narasimhan, 2003)
improved the image quality through the
establishment of the physical optics model of the
atmosphere in the fog and rain and other inclement
weather, however their research was mainly based
on the premise of the known current weather, and
did not classify the image automatically. Roser
(Roser, 2008) recognized clear, light rain and heavy
rain weather that exists in the image of driver
assistance systems based on HSI color space
histograms. Yan (Yan, 2009) analyzed the gradient
510
Li Q., Kong Y. and Xia S..
A Method of Weather Recognition based on Outdoor Images.
DOI: 10.5220/0004724005100516
In Proceedings of the 9th International Conference on Computer Vision Theory and Applications (VISAPP-2014), pages 510-516
ISBN: 978-989-758-004-8
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)