基于车行视程与大气光快速估值的车载视频去雾算法

发布时间：2018-04-29 01:21

本文选题：车行视程 + 大气光估值　；参考：《工程科学与技术》2017年03期

【摘要】：交通场景中的视频图像去雾处理,是一个实时性极强的不确定反问题,针对雾霾天气下车载视频图像退化严重的现象,分析了交通环境中雾气浓度对车前物景可视度的关系,提出了大气能见度与车行视觉距离之间的关系模型,讨论了降质图像增强的理论与方法,建立了雾霾条件下车行可视距离的线性回归公式和基于大气能见度的透射率快速估值模型。同时,研究了雾霾物像暗通道的基本特征,提出了利用引导滤波对图像实现边缘平滑、细节增强以及利用盒式滤波保持边缘信息,降低时间复杂度;直接利用灰度图像获取大气光像素矩阵的估值方法,建立了基于大气光的快速估计模型,解决了暗原色先验理论方法的"非天空区"假设及其时间复杂度难以适应车载视频图像处理的问题。根据上述提出的透射率估计方法和天空光的估值模型,本文提出了一种鲁棒性好,实时性强的雾霾视频图像去雾的新算法,并完成了基于该方法的视频图像恢复处理流程设计,构造了车载雾霾视频图像恢复处理的综合验证平台,通过信息熵和图像边缘检测的方法,对本文提出的方法与目前已有的几种流行的去雾方法进行了图像恢复质量的比较,结果表明,本文提出的方法在反映图像细节和清晰化等方面都取得了良好的处理效果。
[Abstract]:The de-fogging of video images in traffic scene is a highly real-time uncertain inverse problem. In view of the serious degradation of vehicle video images in haze weather, the relationship between fog concentration in traffic environment and the viewability of vehicle front scene is analyzed. The relationship model between visibility and visual distance is proposed, and the theory and method of image enhancement are discussed. A linear regression formula for the visible distance of a vehicle under haze and a fast transmissivity estimation model based on atmospheric visibility are established. At the same time, the basic characteristics of the dark channel of haze image are studied, and the image edge smoothing, detail enhancement and the use of box filter to keep edge information are proposed to reduce the time complexity. A fast estimation model based on atmospheric light is established by directly using gray image to obtain the estimation method of atmospheric light pixel matrix. The "non-sky region" hypothesis of dark priori theory and its time complexity are difficult to adapt to vehicle video image processing. According to the above proposed transmittance estimation method and the sky light estimation model, this paper proposes a new algorithm for removing fog from haze video images with good robustness and real time performance, and designs the video image restoration processing flow based on this method. A comprehensive verification platform for vehicle haze video image recovery processing is constructed. By using information entropy and image edge detection methods, the image restoration quality is compared with several popular de-fogging methods in this paper. The results show that the method presented in this paper has achieved good processing results in terms of image detail and clarity.
【作者单位】：四川大学制造科学与工程学院;
【基金】：四川省科技支撑计划资助项目(2014Z0007;2010GZ0171)
【分类号】：TP391.41;U463.6

【相似文献】