面向视频监控的图像去雾方法研究及系统实现

发布时间：2018-06-11 16:01

本文选题：亮度图像 + YIQ颜色模型　；参考：《西安理工大学》2016年硕士论文

【摘要】：户外监控系统对雾霾等极端天气比较敏感,在雾霾天气条件下,户外视频质量出现严重退化,从而导致其户外视频监控性能受到严重影响。因此,为了增强户外视频监控的适用性和稳定性,研究雾霾视频图像清晰化复原技术具有重要的实际应用意义。本文的主要研究内容是以单幅图像去雾算法为基础,设计完整的适用于视频监控场景下的视频去雾方法的方案,并使用雾天户外监控系统采集的视频进行验证,具体工作如下:1、针对暗先验去雾算法产生边缘残雾和运行速度慢的问题,提出了一种基于亮度图像的快速单幅图像去雾方法。首先,对暗原色估计透射率的方法进行改进,用YIQ颜色模型的亮度分量代替暗通道,并用中值滤波代替最小值滤波,以提高计算效率,并抑制光晕现象。其次,采用效率更高的四叉树算法来求解大气光值。最后,针对室外的各种有雾图像进行实验。实验结果表明,本文方法在处理速度及去雾效果上都优于其它几种经典算法,且适用于视频去雾。2、提出一种面向视频监控的图像去雾方法。首先,根据监控视频背景基本不变的特点,利用均值滤波的背景建模方法提取通用背景图像;然后,利用上述所研究的单幅快速图像去雾方法,基于所得到的通用背景图像估计通用透射率图和大气光值;最后,利用大气散射模型复原视频的每一帧图像,并将其连续输出,得到清晰的视频。实验结果表明,监控视频的去雾方法接近实时去雾的要求。3、在上述工作的基础上,设计并实现了基于MATLAB7.9.0平台的单幅图像去雾系统和基于Visual Studio 2010和OpenCV 2.3.1开发的视频去雾系统。该系统集成了本文所提出的去雾方法,以及部分经典算法。
[Abstract]:The outdoor surveillance system is sensitive to extreme weather such as smog. Under the condition of smog, the outdoor video quality is degraded seriously, which results in its outdoor video monitoring performance being seriously affected. Therefore, in order to enhance the applicability and stability of outdoor video surveillance, it is of great practical significance to study the restoration technology of haze video image clarity. The main research content of this paper is to design a complete scheme for video de-fogging based on a single image de-fogging algorithm, and verify it by using the video collected by the fog outdoor monitoring system. The main work is as follows: 1. In order to solve the problem that dark priori defogging algorithm produces edge residual fog and slow speed, a fast single image de-fogging method based on luminance image is proposed. Firstly, the method of estimating the transmittance of dark primary color is improved. The luminance component of YIQ color model is used to replace the dark channel, and the median filter is used to replace the minimum filter in order to improve the calculation efficiency and suppress the halo phenomenon. Secondly, a more efficient quadtree algorithm is used to solve the atmospheric light value. Finally, experiments are carried out on various fog images outside. The experimental results show that the proposed method is superior to other classical algorithms in processing speed and de-fogging effect, and is suitable for video de-fogging. 2. A new image de-fogging method for video surveillance is proposed. First of all, according to the feature that the background of surveillance video is basically invariant, the background modeling method of mean filter is used to extract the general background image, and then the single fast image de-fogging method mentioned above is used. The general transmittance map and atmospheric light value are estimated based on the obtained general background image. Finally, each frame image of video is reconstructed by atmospheric scattering model, and the clear video is obtained by outputting it continuously. The experimental results show that the de-fogging method of surveillance video is close to the requirement of real-time de-fogging. Based on the above work, a single image de-fogging system based on MATLAB 7.9.0 and a video de-fogging system based on Visual Studio 2010 and OpenCV 2.3.1 are designed and implemented. The system integrates the de-fogging method proposed in this paper, as well as some classical algorithms.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】