基于分数阶理论与最小二乘法的复杂背景车道线检测研究

发布时间：2018-05-07 16:53

本文选题：车道线检测 + 复杂背景　；参考：《长安大学》2017年硕士论文

【摘要】：随着公路交通事业的迅速发展,基于道路图像的车道线检测技术已经成为人们研究的热点和难点。在车道线检测过程中,可能会碰到很多复杂的路况,导致获取的道路图像中会出现一些复杂的背景,例如亮度较低、边缘模糊、油污污染等。在这些情况下,现有的车道线检测方法已不能满足需求,无法很好的检测出车道线。针对以上问题,本文从道路图像去噪、车道线提取和直线拟合三个主要方面对复杂背景情况下道路图像车道线检测进行深入研究,主要工作如下:(1)论文针对道路图像亮度较低、边缘模糊等特点,提出利用基于分数阶积分的算法对复杂道路图像进行去噪。不仅较好的去除了图像的噪声而且保留了图像的边缘信息,不会使图像产生模糊现象。(2)论文针对亮度较低及模糊的道路图像等,提出了一种基于最大类间交叉熵并结合改进自适应阶数的分数阶微分的车道线提取算法。通过分析图像的特征,利用最大类间交叉熵的方法计算图像最优阈值,再将像素点8邻域的平均梯度与最优阈值进行比较,来选定分数阶微分算子的阶数,进行车道线的提取。(3)论文利用改进的最小二乘法进行车道线拟合,将近视野范围内的车道线定义为直线模型,再将拟合后的直线使用消失点进行车道线连接,利用卡尔曼滤波器预测车道线的感兴趣区域减小搜索范围,提高车道线实时检测的效果。由于本文算法针对的图像一般为对比度低和边缘模糊等复杂背景图像,文中选取两幅具有代表性的图像利用本文算法进行检测,并且与参考文献中的算法进行比较分析,实验结果体现了本文算法的有效性和优越性。
[Abstract]:With the rapid development of highway traffic, lane detection technology based on road image has become a hot and difficult point. In the course of lane detection, many complicated road conditions may be encountered, resulting in some complicated background in the acquired road images, such as low brightness, blurry edges, oil pollution and so on. In these cases, the existing lane detection methods can not meet the demand, and can not detect the lane line well. Aiming at the above problems, this paper makes a deep research on road image lane line detection under complex background from three main aspects: road image denoising, lane line extraction and line fitting. The main work is as follows: 1) aiming at the low brightness of road image, Based on fractional-order integral algorithm, the edge blur is proposed to de-noise complex road images. Not only the noise of the image is removed, but also the edge information of the image is retained, so that the image will not produce blur. (2) the paper aims at the road image with low brightness and blur, etc. A lane line extraction algorithm based on maximum cross-entropy between classes and improved adaptive order is proposed. By analyzing the features of the image, using the maximum cross-entropy method to calculate the optimal threshold value of the image, and then comparing the average gradient of the pixel point 8 neighborhood with the optimal threshold value, the order of the fractional differential operator is selected. In this paper, the improved least square method is used to fit the lane line. The lane line is defined as a straight line model within the range of vision, and then the fitted line is connected with the vanishing point by using the vanishing point. The Kalman filter is used to predict the region of interest of lane line and reduce the search area and improve the real-time detection effect of lane line. Because the image in this paper is usually low contrast and edge blur, two representative images are selected to be detected by this algorithm, and compared with the algorithm in reference. The experimental results show the effectiveness and superiority of this algorithm.
【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】：U495;TP391.41

【参考文献】