基于红外图像道路识别与道路跟踪

【作者】 周红娟；

【导师】 任明武；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2004， 硕士

【摘要】 智能车辆(Ⅳ)系统是近年来各国潜心研究的一个新领域，其中的视觉系需要实时检测行驶环境，确定道路的边界，使得智能车辆能够全天候的在无人驾驶或操纵的情况下自主安全平稳的行驶。采用红外摄像机获取图像是该视觉系统常用的一种方法。因此，红外图像的道路识别成为智能车辆系统的一个研究重点。 本文主要针对红外道路图像的特点，对如何有效的提取出道路区域作了一定的研究。主要从三个方面研究了红外道路图像的处理算法。从点域、邻域分别探讨和实现了对道路图像的灰度拉伸、噪声平滑及突出边缘的锐化处理，使用合理的方法去掉了过亮和过暗点的影响，给出了红外道路图像的直方图预处理技术以及基于直方图分析和灰度统计的二值化算法。在对多种边缘检测方法，包括传统的Sobel算子、Laplacian边缘检测分析后，提出了一种带状边缘的检测方法。 另外，本文将广义Hough变换引入到道路边缘曲线拟合中，分别探讨了直线和抛物线的道路模型的曲线拟合。由于道路的复杂性，本文对道路作了一定的假设和推理，并将这些知识同各种算法结合起来，取得了一定的效果。更多还原

【Abstract】 Intelligent Vehicle (IV) System is a new field in which many countries research pay attention to in recent few years. The vision system in it needs real-time checking the environment and confirming the boundary of the road, so that the intelligent vehicle rides automatically, steadily and safely without manual operation. Using infrared camera to acquire images is a common method of the vision system. So road recognition of infrared image becomes a research emphasis of Intelligent Vehicle System.The paper primarily explicates some algorithms about infrared image processing and that how to recognize roads effectively. The paper researches some algorithms in three ways. Firstly, in image pretreatment. The paper respectively discusses some methods of image processing about gray level stretch, image smoothness and image sharpness from point domain and neighborhood domain. Secondly, in threshold analysis, the paper gives the methods of histogram pretreatment and automatic threshold selection based on histogram and gray level statistic. Finally, in edge detection, the paper gives some traditional methods of edge detection, at the same time, it propose a new method based on string-edge detection.The paper cites "generalized-Hough transform" to perform curve fitting, that is, draw out the edge of the road, as well as possible corresponding to different road model, such as "beeline model" and "parabola model". Because of the complexity of roads, the paper makes certain hypothesis and deduction on the roads, and integrates the knowledge into kinds of algorithms, so that some results of road recognition have a good effect.更多还原