基于局部大气光评估的形态学去雾算法研究及应用

发布时间：2018-04-25 10:36

本文选题：暗通道 + 局部大气光　；参考：《兰州交通大学》2017年硕士论文

【摘要】：随着科技的进步,视频监控系统已得到广泛应用,为现代社会的生产和生活带来方便并提供安全保障。例如,银行、大型商场、公司等场所的安保监控系统,可以了解室内情况,顾客行为等,为顾客的人身和财产安全作保障;在交通监控方面,监控范围广,交管部门能够在第一时间发现问题,解决问题。而当户外天气被雾气笼罩时,人的视觉会出现朦胧感,摄像监控设备也会受到天气的影响,所拍摄到的画面不清晰。这样所得到的图像会使图像信息特征提取困难、图像使用不精确,从而降低图像的使用价值,也会直接影响到视频监控系统的实用性。交通道路特别是高速公路上在雾天更容易发生交通事故,交通监控显示的画面模糊,给交管部门及时掌握交通状况带来很大不便。因此,对雾天条件下所拍摄的有雾图像实施去雾处理,提高图像质量是很有必要的。He等人的暗通道先验去雾算法是去雾领域中的经典算法,但该算法如果滤波窗口选择太小,则会导致暗通道先验理论的失效,窗口过大,会导致有雾图像的暗通道在边缘区域评估过小,透射率图出现误差;大气光的粗略估计,也会导致恢复图像细节不突出。基于暗通道先验算法中滤波窗口选择的不足和大气光粗略评估所导致的细节不突出的问题,本文的主要工作有:(1)本文提出形态学暗通道去雾算法,修正暗通道窗口过大导致的过渡腐蚀现象。(2)本文采用形态学亮通道算法确定局部大气光的算法,修正暗通道窗口过小则会导致的暗通道先验理论的失效的现象,恢复图像的天空区域细节,改善图像近景视觉效果。对于道路场景图像的处理,目前,车辆视觉导航已广泛应用于智能交通、安全辅助驾驶等领域。道路交通信息,如道路车道线、交通标志牌等信息是车辆视觉导航系统发挥作用的前提和基础。本文对去雾算法的应用主要有以下几个方面:(1)去雾算法在雾天环境下交通场景图像中的应用。(2)去雾算法在道路车道线的特征提取中的应用。(3)去雾算法在车牌信息监测的应用。通过对几种经典去雾算法进行分析以及视觉效果的对比,发现本文算法具有更好的实时性。
[Abstract]:With the development of science and technology, video surveillance system has been widely used, which brings convenience and security for the production and life of modern society. For example, security monitoring systems in banks, large shopping malls, companies, and other places can understand indoor conditions, customer behavior, and so on, so as to ensure the personal and property safety of customers. In terms of traffic monitoring, the scope of monitoring is wide. The traffic control department can find the problem and solve the problem in the first time. When the outdoor weather is shrouded in fog, people's vision will appear hazy, camera monitoring equipment will also be affected by the weather, and the picture taken is not clear. The obtained image will make it difficult to extract the features of the image information, and the use of the image is imprecise, thus reducing the use value of the image and directly affecting the practicability of the video surveillance system. Traffic accidents are more likely to occur in foggy days on traffic roads, especially on expressways, and the pictures of traffic monitoring and display are blurred, which brings great inconvenience to traffic control departments in time to grasp traffic conditions. Therefore, it is necessary to defog the fogged images taken under fog conditions and improve the image quality. The prior dark channel de-fogging algorithm proposed by he et al is a classical algorithm in the field of fog removal, but if the filter window is too small, the filter window is too small. It will lead to the failure of the prior theory of dark channel, too large window, so that the dark channel with fog image will be evaluated too small in the edge region, and the transmittance map will appear error, and the rough estimation of atmospheric light will also lead to the inconspicuous details of the restoration image. Based on the deficiency of filtering window selection in dark channel priori algorithm and the lack of detail due to rough assessment of atmospheric light, the main work of this paper is: 1) in this paper, we propose a morphological dark channel de-fogging algorithm. In this paper, the morphological bright channel algorithm is used to determine the local atmospheric light, and the failure of the dark channel priori theory will be corrected if the dark channel window is too small. Restore the image of the sky area details, improve the image close-range visual effect. For road scene image processing, vehicle visual navigation has been widely used in the fields of intelligent transportation, safety assisted driving and so on. Road traffic information, such as road lane, traffic signs and so on, is the premise and foundation of vehicle visual navigation system. In this paper, the application of de-fogging algorithm is as follows: 1) the application of de-fogging algorithm in traffic scene image in fog environment.) the application of de-fogging algorithm in feature extraction of road lane line.) the application of de-fogging algorithm in license plate information monitoring. By analyzing several classical de-fogging algorithms and comparing the visual effects, it is found that the proposed algorithm has better real-time performance.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TN948.6

【参考文献】