基于车道线检测与图像拐点的道路能见度估计

为了解决传统的能见度仪价格昂贵、采样有限，以及现有的一些视频测量手段需人工标记物、稳定性差等问题，基于车道线检测与图像拐点提出一种通过固定摄像机识别雾天天气并计算道路能见度的算法。与以往研究不同，在交通模型增加了均质雾天因素。该算法主要分为三步：首先，计算场景活动图，利用区域搜索算法（ASA）结合纹理特征提取待识别区域，如果在待识别区域内像素自顶向下以双曲线形式变化则判断当前天气为雾天，同时计算区域内图像亮度曲线的拐点；其次，基于可伸缩窗算法检测车道线，提取车道线端点并标定摄像机；最后，结合图像拐点以及摄像机参数计算大气消光系数，根据国际气象组织给出的能见度定义计算能见度。通过三种场景下的能见度检测，实验结果表明，该算法与人眼观测效果一致，准确率高于86%，检测误差在20m以内，鲁棒性好。

Visibility estimation on road based on lane detection and image inflection

The traditional visibility meters are expensive, their sampling is limited, and some of the existing video measurement methods need artificial markers and are of poor stability. In order to solve these problems, a new algorithm for weather recognition and traffic visibility estimation through fixed camera was proposed based on lane detection and image inflection. Different from previous research, our traffic model added homogenous fog factor in traffic scenes. The algorithm consisted of three steps. Firstly, calculate the scene activity map. With the help of the Area Search Algorithm (ASA) combined with texture features, extract area for identifying. The current weather condition is foggy if the pixels from top to bottom in the extracted area change in hyperbolic fashion. At the same time calculate inflection point of image brightness curve in the extracted area. Secondly, detect traffic lane based on the retractable window algorithm, extract the lane’s endpoint and calibrate the fixed camera. Finally, according to the visibility definition, calculate traffic scene visibility by International Meteological Organization based on monocular camera model and light propagation model in fog weather condition. Through experiments of visibility estimation for three different scenes, the experimental results show that the algorithm is consistent with human eye’s observation and the accuracy rate is up to 86% while the inspection error is within 20m.