Abstract: Focusing on the problems of lane detection and running deflection for autonomous land vehicles, a reliable approach for fast vision computing is proposed in this paper. Two 5×5 template operators of directional filters are used to extract the edges of the image. The image is binarized with an automatic threshold derived from Otsu algorithm. To reduce the computational work, the position characters of the lanes in the image are concerned to remove the horizontal contour points. Meanwhile, the constraint conditions based on the geometric characteristics are also formulated to eliminate the abundant points. After that, the lanes are detected with Hough transform. With a coordinate system of 3D space, the pinhole camera model is adopted to calculate the deviation angles and the vertical distances of the vehicle to the lanes. These parameters can be applied to estimate the lane departure time with the instant speed of the vehicle, which will be valuable for intelligent control and security early warning of the autonomous land vehicles.

Key words: computer vision, lane detection, edge detection, binarization, Hough transform

摘要： 针对汽车自主驾驶技术的车道检测和跑偏告警问题，提出了一种快速可靠的视觉计算方法。利用方向滤波算子对路面图像进行5×5模板运算，得到边缘图像；采用Otsu自动阈值算法对图像进行二值化处理，并根据车道在图像中的位置特性对边沿图像细化去点，减少后续处理运算量；在此基础上，根据车道几何特性引入约束条件，去除干扰点，并采用Hough 变换检测出车道线；依据针孔摄像机模型建立空间坐标系，用于计算汽车相对于车道线的偏转角和垂直距离，估计驶离车道的时间，为汽车自主驾驶中的安全预警及智能控制提供信息支撑。

关键词: 计算机视觉, 车道线检测, 边缘检测, 二值化, Hough变换