摘要
针对智能车辆视觉导航中的车道保持问题,采用单目视觉技术检测非结构化道路上的车道线和道路边界,解决不同路况下道路检测的鲁棒性与实时性问题.首先用一种自适应阈值分割Otsu方法把道路分为道路区域和非道路区域;然后利用Otsu算法处理后的图像对Canny边缘进行滤波,在消除复杂背景边缘的同时保留可能的弱的道路边界;最后,用直线长度、平均梯度幅值、直线距离和直线角度四元组联合表示霍夫直线,采用蒙特卡罗方法对属于道路边界的霍夫直线的后验置信度进行评价,根据最大权值提取出最优道路边界线.不同场景下的非结构化道路识别实验表明:该算法能够有效克服道路缺损、光影、照度变化、水渍等不利因素的影响,平均处理时间为45 ms左右.
To prevent an intelligent vehicle from departing from its lane in vision-based navigation,a method based on monocular vision was proposed to detect a road boundary.First,the original image was segmented into the road and non-road regions by using the Otsu adaptive threshold segmentation algorithm.Subsequently,the Canny edges were filtered so that certain complicated edges in the image could be eliminated and certain weak road boundaries could be preserved simultaneously.Finally,the Hough lines detected and tracked by the Monte Carlo method were represented by the length,average gradient amplitude,distance,and orientation of the line.Also,the adopted Monte Carlo method would be able to evaluate whether or not the Hough lines belong to certain road boundaries and therefore regard the Hough line having the maximum weight as the optimum road boundary.Experiments indicate that the method can not only overcome negative influences from road flaws,shadows,changes in illumination,and water stains while spending on average only 45ms processing each frame,but also meet the requirements of robustness,real-time,and accuracy.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2011年第3期334-339,共6页
Journal of Harbin Engineering University
基金
黑龙江省教育厅科学技术研究资助项目(1154104011541050)
哈尔滨市青年科技创新人才基金资助项目(2008RFQXG067)
关键词
道路检测
蒙特卡罗方法
阈值分割
边缘检测
智能车辆
road detection
Monte Carlo method
threshold segmentation
edge detection
intelligent vehicle
