Abstract: The future intelligent vehicle navigation system has to be accurate, steady and reasonably priced. To achieve this goal, many fusion models have been built（GNSS/DR, GNSS/INS, GNSS/MM）. Although some have been proved successfully used in different kinds of environment, they still have shortages, especially in areas where the accuracy of GNSS is jeopardized. This paper researches on a method which can be able to meliorate the GNSS localization precision by stereo-vision utilizing the geographical information of landmarks in partial areas. Random Hough transformation is used to detect landmark, SIFT and K-means algorithm are used to match the landmarks. Stereo-disparity figures out the vector between vehicle and landmark, then it builds an aided model to calculate vehicle’s position. The contents include using experimental vehicle to collect the real-time data in a challenged environment. Finally, after analyzing the error of stereo-vision and the error of GNSS, it proves that the proposed method makes an obvious improvement on GNSS locali-zation results in the areas where landmarks are visible.

Key words: intelligent vehicle navigation, challenged environment, Global Navigation Satellite System（GNSS）, stereo-vision aided localization system

摘要： 具有精确、稳定的定位结果以及合理的价格是未来的智能车辆导航系统的发展趋势。为了达到这个目标，人们建立了多种组合导航模型（GNSS/DR，GNSS/INS，GNSS/MM）。尽管这些模型已在多种不同环境中成功应用，但它们仍有许多缺陷，尤其是在全球卫星导航系统（GNSS）定位精度受到威胁的区域。研究了一种通过双目视觉利用路标的地理位置信息对GNSS定位精度进行局部改良的方法。随机霍夫变换用于路标检测，SIFT算法与K均值算法将用于路标的匹配识别。双目视差计算智能车与路标之间的向量，从而建立辅助定位模型计算车辆的位置。利用实验车在一处复杂环境区域进行实时数据采集，通过计算出的双目视觉定位误差与GNSS定位误差对比分析，验证了该方法在路标可见范围内对GNSS定位结果有明显改善。

关键词: 智能车辆导航, 复杂环境, 全球卫星导航系统, 双目视觉辅助定位系统