基于车道线虚线角点检测的行车安全视距测算模型

开展行车视距调查对于营运期公路安全评价至关重要，这对车载条件下行车视距检测提出了要求。针对现有基于车道线图像特征点所构建的视距模型精确度不高的问题，提出了1种以车道线虚线角点为关键特征的行车安全视距测算模型。在车载设备获取的图像预处理基础上，采用轮廓跟踪法对车道线虚线轮廓进行提取，通过设定轮廓尖锐度阈值以实现对车道线虚线角点的初步筛选；使用最大、最小距离法对候选角点进行聚类分类，将每类中尖锐度最大的点判定为真实角点；此外，结合车道线虚线图像梯形特征实现对角点的精确提取；根据成像原理的坐标转换关系，通过解构角点在世界坐标和像素坐标之间的映射关系以求解二者之间的转换矩阵，得到实际道路环境的行车安全视距测算模型；将模型所测算的行车视距与运行车速所需的行车视距进行对比，实现对不同道路线形下行车视距的评价。通过实车实验对所提行车视距测算模型进行动态和静态检测精度验证。结果表明:该模型在静态条件下的行车视距检测误差小于7%，低于采用车道线特征点提取方法检测的误差；在动态车载条件可实现行车视距的连续检测，表明在该模型能适应动态条件对行车视距的检测。该模型可实时动态检测行车视距，为营运期公路安全评价提供支撑。 

A Model for Estimating Driving Sight Distances Based on Corner Point of Broken Line of Roadway

A study of driving sight distances is critical for safety evaluation of highways, which makes it ideal for estimating driving sight distances using in-vehicle equipment. To address the low accuracy of the existing sight distance models using the feature points of lane marking images, a model for estimating driving sight distances with the dotted corner points as the important feature is proposed. Based on the images preprocessing obtained by an in-vehicle equipment, the contour tracking method is used to extract the contour of line markings, so that the initial screening of the corner points can be extracted by setting a threshold sharpness of the contour. After using the maximum and minimum distance methods to cluster and classify candidate corner points, the points with the largest sharpness in each category is determined as the"true"corner points. In addition, the accurate extraction of the diagonal points is achieved by using the trapezoidal features of the dashed line image of the lane marking. By considering the relationship between the global mapping coordinates and the pixel coordinates of the corner points, the transformation matrix between the two coordinates is settled and the estimation model of driving sight distance is developed. By comparing with estimated sight distance with the required distance at a given operation speed, the evaluation of driving sight distance of the alignment of current road segment is implemented. The dynamic and static detection accuracy of the proposed sight distance estimation model is verified by a field experiment. Study results show that the estimation errors under the static condition are less than 7%, which is lower than the traditional methods. In addition, under dynamic conditions, the errors of driving sight distance are consistent with the results of static conditions, indicating that the proposed estimation model has a good performance under dynamic conditions. Comprehensively, the model can be used to support safety evaluation of highway design and operation.