摘要
为了提高非结构化道路识别算法的有效性,提出了一种道路分割的新方法,建立了道路区域和非道路区域混合高斯彩色模型,根据像素隶属于彩色模型的概率进行基于彩色信息道路分割.利用彩色分割的结果对提取的图像边缘进行有效约束,抑止大量非道路边沿所产生的图像边缘.并且将彩色分割结果和道路图像的边缘信息融合,利用道路图像边缘信息对真实道路边沿定位的精确性和彩色信息对道路区域分割的适应性,通过动态规划算法求解出真实的道路边沿.实验结果表明,提出的新方法可以有效地分割出道路区域,对各种路况具有良好的适应性.
A new detection method for unstructured off-road scenes was proposed to improve the effectiveness of lane segmentation algorithm. The color of road and off-road regions were modeled by Gaussian mixtures which were trained in a supervised manner. For a new image, color segmentation was first applied based on the posterior probability. Sobel edge detector was then used to locate candidate edgelets along road boundaries. By combining both segmentation and edge information, how likely each edgelet was along the actual road boundaries could be evaluated. Dynamic programming was used to find an optimal concatenation of edgelets that form road boundaries. Experimental results show that this method could segment the lane boundary effectively, and is robust against noise, shadows and illumination variations.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2006年第1期29-32,共4页
Journal of Zhejiang University：Engineering Science
关键词
视觉导航
移动机器人
道路分割
vision-based navigation
mobile robot
lane segmentation
