中文摘要：

现有的交通摄像机标定算法大多基于车道线长度、车辆尺寸等先验信息，由于在非结构化道路中往往不存在车道线，使得标定算法具有局限性．为了改进摄像机标定在非结构化道路中的适用性，结合摄像机线性模型与均质雾天，提出一种只包含路面以及运动车辆的摄像机动态标定算法．首先生成并更新背景和场景活动图，提取路面、天空的纹理特征，利用区域搜索算法得到感兴趣区域，并根据感兴趣区域的像素变化规律判断当前天气是否为均质雾天；其次根据暗原色先验原理计算场景透射率，将结果映射到[0，255]作为图像显示；最后结合均质雾天光线传输模型、摄像机线性模型和暗原色先验原理导出标定方程，选取路面上具有特定透射率的8个点生成2个一次方程、1个二次方程和1个三角方程，依次标定摄像机参数，将视频多帧图像标定的参数值平均得到准确值．与角点检测法、摄像机6点标定法以及基于消失点的标定算法进行对比的实验结果表明，该算法是有效的且满足视频的实时性处理要求．

英文摘要：

Most existing traffic camera calibration algorithms work on the basis of rune length size prior information; however in unstructured road there are often no traffic lanes. In order to improve the applicability of camera calibration algorithms in unstructured road, a dynamic camera calibration algorithm combining with the camera linear model and homogeneous fog is presented in this paper; this algorithm only contains road surface and moving vehicles. Three main steps are included in the algorithm. Firstly, the algorithm generate and update scenes background and scene activity diagram, extract texture features of the road and sky, search interesting region by the area search algorithm （ASA） and judge whether it is homogeneous fog or not according to the pixel changing form in the interesting region; Secondly, calculate scene transmittance for each point in the scene according to dark channel prior and map it toto display as an image. Finally, deduce camera calibration equation by combining with light transmission model, linear camera model and dark channel prior; select eight road surface points with special transmittance on the road surface to generate two linear equations, one quadratic equation and a trigonometric equation to calibrate camera parameters in turn, get exact camera parameters value through multi-frame average. At the end of this paper, calibration results are given. Comparisons with the conner detection method, six point calibration algorithm and calibration algorithm based on vanishing point verify the effectiveness of this algorithm; meanwhile this proposed algorithm can satisfy real time video processing.