视觉导航中基于图像的路边识别算法研究

【作者】 张晓峰；

【导师】 陆建峰；

【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】 近年来，智能车辆系统成为各国潜心研究的一个新领域，其中的视觉导航系统需要实时检测行驶环境，确定道路的边界，使得智能车辆能够在无人驾驶或操纵的情况下自主安全平稳的行驶。 本文主要研究了视觉导航中道路边界识别问题，按照图像预处理→道路区域分割→道路边缘检测→道路边缘提取→道路边界表示这个流程来对图像进行处理。对于道路区域分割采用最优阈值分割，分离出道路区域。对于道路边界的检测，本论文改进了Robert算子得到了连续且很窄的边缘，边缘检测效果比较理想。在道路边界的提取中，结合阈值分割后的图像和边缘检测后的图像，用链码跟踪方法提取最可能的道路边界曲线。对于道路边界表示问题，建立了一种道路模型，并提出了一种基于最优分段线段的道路边界拟合算法。通过对红外线夜视图以及仿真图等序列图像的实验，表明本文所提出的方法可以满足实用需要。更多还原

【Abstract】 Recent years, Intelligent Vehicle System has become a new field which attracts more attention over the world, where Vision Navigation System should detect the running environment and locate the boundary of the road in real-time so as to guarantee the intelligent vehicle to drive autonomously, steadily and safely without manual operation.This paper primarily studies the problem about recognizing the edges of the road for navigation and adopts following operations sequence: Image preprocessing → Road region segmentation → Road edge detection → Road edge extraction → Road edge representation. For road segmentation, optimal threshold is used to extract the road region. For road edge detection, this paper improves Roberts operator so as to obtain continuous and narrow edges, the result after improvement is satisfied. Chain code tracing is applied to the segmented image and edge detected image to extract the most possible curve. One road model is built for road boundary, and a optimal fitting with multiple line segment is proposed. The whole system is tested on several series of infrared image and simulated image and achieves satisfactory results.更多还原