【作者】 吕欣；

【导师】 樊春玲；

【作者基本信息】 青岛科技大学 ， 检测技术与自动化装置， 2010， 硕士

【摘要】 近年来,视觉辅助导航已成为智能车辆导航领域的研究热点之一,其中道路检测是视觉辅助导航中的关键技术,道路图像的检测结果会受到噪声质量及成像质量的严重影响。本文以道路检测技术中将会遇到的几个关键问题作为研究重点,对道路图像进行增强、边缘检测等算法的处理,主要研究了基于图像处理的道路检测技术。本文根据道路图像的噪声类型,针对不同的噪声采取相应地去噪方案,提出一种改进的直方图均衡算法,可以有效地解决图像的模糊问题;而对于雨雪天气造成的椒盐噪声,利用中值滤波能够有效地去除其噪声;对于高斯白噪声,则采用自适应滤波的滤波效果最好。在图像边缘检测中,通过对经典边缘检测算法的具体分析与比较,提出一种基于多尺度几何分析的Beamlet变换的边缘检测算法。该算法符合最优图像特征的表示方法,能够充分利用图像的几何正则性,检测出的道路边缘更加连续清晰,使道路的边缘特征得到较好的保护。在阈值分割环节,通过对几种传统阈值分割方法的比较,根据道路图像特点,确定使用最大类间方差法进行阈值分割。在进行道路边界线提取时,采用一种基于直线点密度的霍夫变换方法,将图像空间和霍夫空间有机地统一起来,该方法能够避免由于过度积累而导致伪直线过多的缺点,同时具有较好的鲁棒性和抗噪性,从而提高道路检测的准确性。文中通过对采集的道路图像添加椒盐噪声、高斯白噪声以及模糊化的处理,模拟实时道路检测中可能会遇到的影响图像质量的问题,仿真实验结果论证本文所提出的算法能够准确地检测出道路图像中的车道线,提高道路检测的识别率,证明文中算法的有效性和可靠性,为后续的研究工作提供有用的道路信息。更多还原

【Abstract】 Recently, road detection technique is a very important part of vision navigation, which is the key technique of intelligent vehicle guidance. The detection result is seriously affected by the quality of noise and image. Several key problems in the road detection technique are chosen as the emphasis of this paper, and then modify the road image quality by a series of changes in image enhancement, edge detection and so on. This paper analyses and researches the methods of road detection technique based image process.This paper analysis the type of noise, put forward the right denoise method for different types of noise, and then proposed an improvement in histogram equalization, which can effectively solve the blurring problems in the image. For certain degree influence regarding the path recognition under the sleet condition to use spiced salt noise to carry on the simulation to it. The discovery uses automatic filter the effect to white Gaussian noise is better.Through the specific analysis and comparison of various edge detection algorithm, the image edge detection algorithm based on multiscale geometry analyse is used to extract the edge outline. The new edge detection algorithm can make use of the geometrical regularity better and conform to best practices in image picture attribution. By the simulate experiments, the method processes lubricity and series in the edge, the texture of road edge is protected better. In image segmentation, this paper study server traditional threshold segmentation algorithm, and use maximum classes square error in this paper.In road boundary line extraction, use Hough transform based on Line Point Density, which can solve the accumulating problem of classical Hough transform. This method also have nice anti noise performance and robustness, and can effectively extract lines form road image. Add spiced salt noise, white Gaussian noise and fuzzy treatment methods in the real-time simulation of the road likely to be encountered in the quality of the image, then pretreatment for road detection. The proof shows that the algorithm in this paper can effectively improve the recognition rate in road detection, accurately extract lines form road image, and proved the efficiency and the reliability of this approach. It also provided useful road information for follow up study.更多还原