融合图切割和聚类算法的鲁棒自适应道路跟踪  

Novel robust and self-adaptive road following algorithm based on graph cuts and clustering

作　　者：周文晖[1,2] 林丽莉[3] 武二永[1] 

机构地区：[1]杭州电子科技大学计算机学院,杭州310018 [2]浙江省综合信息网技术重点实验室浙江大学,杭州310013 [3]浙江工商大学信息与电子工程学院,杭州310035

出　　处：《仪器仪表学报》2009年第11期2366-2371,共6页Chinese Journal of Scientific Instrument

基　　金：浙江省自然科学基金(Y1080967);浙江省重大科技专项(优先主题)社会发展项目(2008C13076);浙江省教育厅科研计划(Y200701698);浙江省教育厅科研计划资助项目(Y200702656);浙江省综合信息网技术重点实验室开放课题资助项目

摘　　要：维道路跟踪是移动机器人视觉导航的关键任务之一。由于室外道路环境的复杂性,使得鲁棒连续的基于二维图像序列的道路跟踪仍然是个挑战性任务。本文提出一种基于改进图模型的自适应道路跟踪算法,利用基于边缘置信度的均值偏移算法,将图像划分为具有准确边界的若干同质区域,以这些区域为结点构建改进图模型,然后根据道路/非路模型统计信息,采用Graph Cut方法获得最终的二值图。该算法将Graph Cut和均值偏移方法有效融合,以克服各自缺点,并通过道路/非路模型自更新使得该算法可有效适应室外环境下复杂场景变化。实验结果表明,该算法在复杂道路环境下具有很好的性能,且适合快速运算的应用要求。Two dimension road following is a crucial task of vision navigation for mobile robots. Because road environments are usually complex, robust and continuous road following based on two-dimension image sequence is still a challenging task. This paper proposes a self-adaptive road following algorithm based on an improved graph model. Firstly, the mean shift algorithm embedded with edge confidence is used to partition the images into homogenous regions with precise boundary, and an improved graph model is constructed with these regions. According to the statistic information of the road/non-road models, the Graph Cut （GC） algorithm is applied to achieve the final binary images. This algorithm can overcome some difficult problems of mean shift and GC by effectively combining these two methods. Moreover, the road/non-road model self-update property makes this algorithm adapt to complex scene changes in outdoor environment. Experiment results indicate that the proposed method possesses excellent performance in complex environments and meets the requirements of fast computing.

关 键 词：道路跟踪 图切割 均值偏移 边缘置信度 

分 类 号：TN911.73[电子电信—通信与信息系统]