| dc.description.abstract | Over the past decade, Traffic Surveillance System (TSS) has seen significant development,
from the manual control to automatic performance. Many researchers have developed
algorithms to cope with a wide range of scenarios such as overcast, sunny weather that created
shadows, rainy days that result in mirror-reflection on the road, or nighttime when low lighting
conditions limit the visual range.
Most of the methods strive to operate well in all cases. Indeed, they just perform correctly
in some specific case, their result in other cases are not good. Therefore, determining suitable
methods for each case is the crucial thing in any system.
One of the most challenging problems is the scene determination in a highly dynamic
outdoor environment in real-world applications. As also pointed out in recent surveys, there
have been weak studies on a mechanism for scene recognition and adapting appropriate
algorithms for that scene. Therefore, this research presents a scene recognition algorithm for
running all-day surveillance, which covers four types of outdoor environment.
The proposed method detects and classifies outdoor surveillance scenes into four common
types: overcast, clear sky, rain, and nighttime. The major contributions are to help diminish
hand-operated adjustment and increase the speed of responding to the change of alfresco
environment in the practical system. To obtain highly reliable results, authors combine the
histogram features on RGB color space with the probabilistic model on CIE-Lab color space
and input them into a feed-forward neural network. Early experiments have suggested
promising results on real-world video data.
Keywords: scene recognition, traffic surveillance system, probabilistic model, artificial
neural network | en_US |
