基于图像处理的车道线检测算法研究

发布时间：2018-01-12 23:14

本文关键词：基于图像处理的车道线检测算法研究　出处：《河南工业大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着机动车数量的不断增加，交通事故的发生量逐年上升，造成了巨大的财产损失和人员伤亡。近些年来，国内外学者们相继提出了一些具有重大意义的车道线检测算法，但是普遍存在两方面的问题。第一，当路面存在较多或较强干扰信息时，算法鲁棒性不够强。第二，在这种干扰信息存在的条件下，在保证系统鲁棒性的同时实时性又不能满足系统需求。为了解决以上两个问题，本文提出了一种快速、有效的车道线检测算法。为了消除图像在采集阶段引入的噪声干扰和一些冗余信息，本文首先对采集到的道路图像进行感兴趣区域划分、灰度化和图像增强处理。然后，，根据车道线具有的线性边缘特征，采用Gabor滤波处理；为了避免在复杂路况下图像分割时阈值难以确定的问题，针对车道线的特征设计了几种Haar矩形特征模板来提取车道线特征，再结合Adaboost算法训练分类器，通过训练最终得到判别函数来对特征点进行分类。为了避免Adaboost算法的缺陷，针对车道线的分类问题，设计了针对车道线的改进Adaboost算法，在计算过程中采用多种快速处理方法，在不降低鲁棒性的条件下保证了系统实时性。最后，使用改进的可并行运算的Hough变换方法提取车道线的参数信息并模拟出车道线。经过在多种路况下实验表明，本文采用的特征提取算法能比传统算法准确率高10%左右，并且本文的改进Hough变换方法比传统Hough变换节约90%以上的时间消耗。在多种路况下均能快速、准确识别出图像中的道路标识线，能够满足车道线识别算法对实时性和鲁棒性的要求。
[Abstract]:With the increasing number of vehicles, traffic accidents happen to rise year by year, caused a great loss of property and casualties. In recent years, domestic and foreign scholars have proposed some significant lane detection algorithm, but generally there are two problems. First, when the road exists more or stronger interference information, the robustness is not strong enough. Second, in the presence of interference information, to ensure the system robustness and real-time and can not meet the requirement of the system.
In order to solve the above two problems, this paper presents a fast and effective lane detection algorithm. In order to eliminate the image noise introduced in the acquisition stage and some redundant information, firstly, the road images collected the interested region, grayscale and image enhancement processing. Then, according to the characteristic of linear edge the lane line, using Gabor filter; in order to avoid the complex conditions of image segmentation threshold is difficult to determine, according to the characteristics of the lane design several Haar rectangular feature templates to extract the lane feature, combined with the Adaboost algorithm to train the classifier through training, finally get the discriminant function to classify the feature points in order to avoid. The shortcomings of Adaboost algorithm, aiming at the problem of classification of lane, the improved Adaboost algorithm for lane design, using a variety of in the process of calculation The fast processing method ensures the real-time performance of the system without decreasing the robustness. Finally, the improved parallel operation Hough transform is used to extract the lane information and simulate the lane line.
The experiments show that in a variety of conditions, the algorithm used to extract features than the traditional algorithm has high accuracy rate of about 10%, improved Hough transform method and the consumption than traditional Hough transform to save more than 90% of the time. In a variety of conditions can quickly, accurately identify the image of road marking line, to meet the requirements lane recognition algorithm of real-time and robustness.

【学位授予单位】：河南工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;U463.6;TP391.41

【参考文献】