基于单目视觉的公路视距检测技术研究

发布时间：2018-02-21 15:03

本文关键词： 交通安全视距检测单目视觉车道线提取试验验证　出处：《长安大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着我国公路交通的迅速发展,交通安全问题日益突出,大量研究表明,公路视距不足,驾驶人获取道路危险信息滞后,是导致道路交通事故频繁发生最为主要的原因之一。所以,保证驾驶人在道路交通运输过程中获得足够的视距是保障道路交通安全的首要任务。因此,如何及时有效地检测公路视距,改进公路安全设计,提醒驾驶人合理控制车速,避免道路交通事故发生,提升公路运营的安全保障能力将是改善道路交通安全问题的重要途径。为了获得良好的公路视距检测结果,本文对图像处理及单目视觉测距进行了相关研究,提出了基于车道线特征点坐标转换关系的公路视距检测算法,主要研究内容如下:(1)首先,针对传统的最大类间方差法阈值分割造成的车道线区域错误分割问题,利用车道线灰度分布曲线通过多次试验确定车道线的灰度分布范围,将车道线灰度范围设置为两个阈值进行双阈值分割。(2)其次,针对真实道路环境中可能出现的障碍物边缘信息的干扰以及由于光照不均、车道线污损等原因导致的车道线边缘模糊问题,本文采用了基于Sobel竖直梯度卷积模板进行边缘增强处理,改善了车道线边缘的清晰度,提高了车道线提取的效果,然后针对车道线初步提取后存在的线条干扰问题,本文采用了基于车道线宽度特征的车道线行扫描方法进行二次提取。(3)最后,根据视距的相关知识,将视距的检测问题转化成车道线消失点与车辆之间的车道线长度的检测问题。分析了摄像机成像过程中坐标系之间的转换关系并采用张正友法进行摄像机标定得到摄像机内参数,通过几何投影及坐标转换关系,推导出了基于车道线特征点坐标转换关系的视距检测模型,并进行了静态实车试验。研究表明:本文提出的基于单目视觉的自由流状态下的公路视距检测算法可以较好地检测出直线以及曲线等各类道路线形的公路视距,平均公路视距检测误差为5.2%。研究结果可以应用于运营公路的安全评价以及行车过程中对驾驶人的安全提醒。
[Abstract]:With the rapid development of highway traffic in China, traffic safety problems become increasingly prominent. A large number of studies show that the road visual distance is insufficient, and the road dangerous information for drivers is lagging behind. It is one of the most important reasons that cause the road traffic accident to happen frequently. Therefore, it is the first task to ensure the road traffic safety to ensure that the driver obtains the sufficient visual distance in the course of the road traffic transportation. How to detect the highway visual distance in time and effectively, improve the highway safety design, remind the driver to control the speed reasonably, avoid the road traffic accident, To improve the safety and security ability of highway operation will be an important way to improve road traffic safety. In order to obtain good results of road visual range detection, this paper studies image processing and monocular vision ranging. In this paper, a highway line-of-sight detection algorithm based on the relationship between lane feature points and coordinate transformation is proposed. The main research contents are as follows: firstly, aiming at the problem of lane error segmentation caused by the threshold segmentation of the traditional maximum inter-class variance method, a new algorithm is proposed. The grayscale distribution range of driveway line is determined by many experiments, and the grayscale range of driveway line is set to two thresholds for double threshold segmentation. In view of the interference of obstacle edge information which may appear in real road environment and the ambiguity of lane edge caused by uneven illumination and lane line fouling etc. In this paper, the vertical gradient convolution template based on Sobel is used for edge enhancement, which improves the clarity of lane line edge and improves the effect of lane line extraction. In this paper, the lane line scanning method based on lane width feature is used to extract the lane line twice. Finally, according to the relevant knowledge of the distance of sight, The detection problem of visual distance is transformed into the detection of lane line vanishing point and lane line length between vehicles. The transformation relationship between coordinate system in camera imaging process is analyzed and calibrated by means of Zhang Zhengyou method. Parameters inside the camera, Based on the geometric projection and coordinate transformation relationship, a line-of-sight detection model based on the relationship between lane feature points and coordinate transformation is derived. The research shows that the proposed road visual distance detection algorithm based on monocular vision in free flow state can better detect the road line-of-sight distance of all kinds of road lines, such as straight lines and curves. The detection error of average road sight distance is 5.2. The results can be used to evaluate the safety of operating roads and to remind drivers of safety in the course of driving.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U492.8

【相似文献】