基于亮度对比度和暗原色先验原理的白天道路图像能见度检测方法

A Visibility Detection Method Using Daytime Road Images Based on Brightness Contrast and Dark Channel Prior Principle

低能见度是对道路通行影响最为严重的气象要素之一。随着数字摄像技术和图像识别技术的发展以及气象和交通部门间信息共享工作的开展,利用高速公路沿线摄像头视频数据快速识别能见度成为提高能见度时空监测精度的重要手段。本文提出了一种基于亮度对比度和暗原色先验原理的白天道路图像能见度检测方法。首先根据霍夫变换直线检测方法提取道路兴趣域,然后根据亮度对比度方法检测人眼可分辨最远像素点,将其作为目标点,最后基于暗原色先验原理求取目标点的透射率,并根据能见度与消光系数的关系公式求取图像能见度值。利用安徽省京台高速吴玗北段和宁绩高速宁国互通段视频图像资料和邻近交通气象站能见度监测资料,采用绝对误差和能见度等级误差对能见度检测效果进行检验。结果表明,本方法对能见度的变化较为敏感,能见度等级的检测效果较好,准确度可达95%,对开展公路交通视频图像能见度识别工作具有较好借鉴应用意义。

Low visibility is one of the most serious meteorological factors affecting road traffic.Along with the technology development of digital camera and image recognition and the information shared among the meteorological and traffic departments,using camera video data along the expressway to rapid recognize the visibility becomes an important method to improve the precision of visibility temporal and spatial monitoring.In this paper,we propose a daytime road image visibility detection method based on the brightness contrast and dark channel prior principle:(1)to extract the interested road domain according to the Hough transform detection method,(2)by using the brightness contrast method,the farthest pixel point that the human eye can distinguish is used as the target point,(3)to evaluate the transmittance of the target point based on the dark channel prior principle,(4)to figure out the image visibility value by the relationship between visibility and extinction coefficient.The test indicators of this paper are visibility absolute error and visibility level error.The test data is derived from the video image data of the Wuyu section of Jingtai Expressway in Anhui Province and the Ningguo-Hutong section of Ningji Expressway in Anhui province,and the visibility data set of the traffic weather stations near the expressway in Anhui Province.The results indicate that the method can accurately reflect the trend of visibility change,and the detection accuracy can be as high as 95%,which can be a good reference for the image visibility recognition.