Abstract: Road detection is a key issue in the field of intelligent vehicle.Learning-based methods always divide the road image into several regions，then classify them as road or non-road.But due to the complexity of real world environment，there are uncertain regions with both road and non-road areas，and this confusion makes the classification on these regions meaningless.To address the above problem，a new road detection algorithm based on the re-classification on uncertain regions（RCUR） is presented.RCUR detects uncertain regions automatically，and then segments the uncertain regions into several sub-regions by taking advantage of complementary of different segmenting algorithms.Through combination and classification on sub-regions，RCUR distinguishes the road area and non-road area in uncertain regions effectively.Experiments demonstrate that the proposed algorithm adapts itself well to the diversity of roads，improves the accuracy in road detection and decreases the negative impact caused by noises.

Key words: road detection, uncertain regions, Re-classification on Uncertain Regions（RCUR）

摘要： 路面检测是智能汽车领域的一个重要研究课题。基于学习的方法将获取的汽车前方的图像划分为一些区域，然后分别将这些区域分类为路面区域或非路面区域。由于现实场景的复杂性，存在一些既包含路面又包含非路面的不确定区域，只是将其分类为路面区域或非路面区域是不合理的。针对上述问题，提出了一种新的基于分割的路面检测算法，其核心是不确定区域再分类算法RCUR（Re-classification on Uncertain Regions）。该算法检测出不确定区域后，利用不同分割算法的互补性将不确定区域分割为若干子区域，通过对子区域的组合、分类可以有效地区分出不确定区域中的路面与非路面部分。实验表明该算法能够在现实场景中适应路面的多样性，提高路面检测的正确率，降低噪声对路面检测结果的影响。

关键词: 路面检测, 不确定区域, 不确定区域再分类算法（RCUR）