Abstract
License plate recognition technology for traffic management systems around the
importance of self-evident, it has attracted the attention of many researchers in recent
years. However, in the absence of light or even in the dark, the images collected by
most existing ordinary image acquisition equipment have problems such as low
quality, insufficient clarity, and insufficient brightness. On the other hand, the issue of
night traffic monitoring and license plate recognition in low-light environment is
becoming more and more important in today's society, which is a research topic with
practical significance.
In order to solve the above-mentioned problems in the research of license plate
recognition under low light environment, this paper proposes an effective license plate
detection and automatic recognition solution combining image restoration and neural
network method after referring to a large number of literatures and conducting many
experiments. When the low-light video is inverted, it has a high similarity to the video
captured in a hazy light environment such as a foggy environment. Analogously to the
inverted low-light image, we invert the input original low-light image and then use
atmospheric light. The model uses the defogging method combined with the filtering
operation to restore a high-quality image. Finally, a clear image with the proper
illumination is required. Because the method of mean filtering avoids the complex
calculation of estimating the atmospheric light, which provides higher quality sample
images for subsequent processing.
In order to solve the problem that the edge detection effect is not ideal in the low
illumination environment, we use the HSI color model combined with the color
positioning to further improve the success rate of the target area positioning. For the
pre-processed images, gaussian filtering is first used to eliminate the noise before
edge detection and color features are combined. The candidate regions are
preliminarily determined by using the texture structure features of the target region,
and other non-target regions are excluded by prior knowledge and SVM model to
further accurately locate the license plate. The character segmentation method is
adopted in combination with prior knowledge for the special structure of Chinese
characters (stroke and relative position), and a character segmentation method is
obtained which is more suitable for the special structure of Chinese characters.
评论0
最新资源