摘 要

Abstract III

1 绪论 4

1.1 课题研究背景及意义 4

1.2 车道线识别研究现状......................................................................................................4

1.3 论文主要内容 5

2 道路图像的增强 6

2.3.2 自适应滤波 11

3 道路图像边缘检测 13

3.1 图像边缘的定义 13

3.2 边缘检测概述 13

3.3 经典边缘检测算法 14

4 基于多尺度几何分析的道路图像边缘检测 18

4.1 基于多尺度的Beamlet 19

4.2 几种基于Beamlet变换的算法 21

4.3 基于Beamlet算法的图像边缘检测 21

4.4 道路图像的阈值分割 22

4.5 本章小结 23

5 道路车道线的提取 23

5.1 Hough变换 23

5.2 道路车道线的提取 24

5.3 整体算法流程 25

5.4 实验仿真结果与分析 26

5.5 本章小结 31

6 总结与展望 32

参考文献 33

致 谢 35

基于Beamlet算法的车道线识别算法研究

摘 要

现如今，智能车辆导航领域的研究热点之一是自动驾驶技术，车道线识别更是这当中的一项重要技术，噪声、成像的质量会严重干扰道路图像识别结果。本文重点讨论车道识别技术面临的一些重要问题，提出了改进道路图像，以及边缘检测等，着重介绍了目前正应用于车道相关处理和识别技术之中的一种Beamlet算法。

因为噪声的种类繁多，处理手段也不尽相同。所以要选择对应的噪声消除方案来处理，中值滤波有效地去雨雪天气造成的椒盐噪声；高斯白噪声用自适应滤波处理；图像模糊的问题选直方图均衡算法。

介绍一种基于Beamlet变换的边缘检测算法，与传统算法相比。通过这种算法检测出到车道线更清晰、连续。考虑到道路图像的实际情况，在选择阈值分割算法时决定最大类间方差法。

并把霍夫变换来作为获取道路边界线的首选方式，这种方法有较好的抗噪性和鲁棒性，因而提高了识别的准确性。

为了模拟各种实际检测时可能产生的问题，道路图像中加入了各种噪声以及模糊处理后进行了具体的实验仿真，来对比检测道路的准确性，结果证实文章里提及的算法更加有效可靠，能够提高检测的精度。同时也为后续的工作提供了有价值的信息。

关键词：道路标线；Beamlet变换；边缘检测；多尺度；霍夫变换

Research on lane detection algorithm based on Beamlet

Abstract

Now, automatic driving technology is one of the hotspots in the field of intelligent vehicle navigation. Lane detection is one of the most significant skills in this part. The quality of noise and imaging will seriously interfere with the result of road image recognition. This paper mainly discusses some important problems in lane recognition technology, and puts forward some algorithms to improve road image and edge detection, and mainly discusses the lane recognition technology based on Beamlet.

In this paper, in the face of a variety of possible noise in the road image, the corresponding noise elimination scheme is chosen to deal with the noise. The median filter can effectively remove the salt and pepper noise caused by the rain and snow weather. As for Gauss white noise, the best filtering effect is adaptive filtering, and the image model can be solved by the improved direct square graph equalization algorithm. The problem of the paste.

Compared with the classic image edge detection algorithm, an edge detection algorithm based on Beamlettransform is proposed. The algorithm corresponds to the representation of optimal image features, and the lane lines detected are more clear and continuous. After comparing several traditional threshold segmentation methods, taking into account the characteristics of road images, the maximum variance method is the best.

When the road boundary line is extracted, a Hough transform method is used to combine the image space with the Hough space. This method can make up the shortcoming of too many pseudo straight lines due to excessive accumulation, and has better noise resistance and robustness, thus improving the accuracy of recognition.

By adding noise and fuzzification of the collected road images, the problems that may be encountered in lane recognition are simulated, and the accuracy of road detection is improved according to the results of simulation experiments. The validity and reliability is provided to the post sequence research work.

Key Word: Road Marking；Beamlet Transform；Edge Detection；Multiscale ；Hough Transform

1 绪论

1.1 课题研究背景及意义

1.1.1 课题研究背景