道路标线Beamlet快速精确分割算法  

Quick and precise road marking segmentation algorithm based on Beamlet

作　　者：徐志刚[1] 赵祥模[1] 杨澜[1] 韦娜[1] 张立成[1] 

机构地区：[1]长安大学道路交通智能检测与装备工程技术研究中心,陕西西安710064

出　　处：《长安大学学报（自然科学版）》2013年第5期101-108,130,共9页Journal of Chang’an University(Natural Science Edition)

基　　金：国家自然科学基金项目(51278058;60902075);陕西省自然科学基金项目(S2013JC9397);博士后特别资助项目(201003660);中央高校基本科研业务费专项资金项目(CHD2009JC114;CHD2010ZY007;CHD2010JC056)

摘　　要：为了提高道路标线识别的精度和速度,提出了一种基于Beamlet的道路标线快速精确分割算法。首先采用双线性插值算法将1 024×1 024像素大小的原始路面标线图像缩小为64×64像素大小的低分辨率图像,然后采用Sobel算子检测出图像中的阶跃边缘,在对文献中Beamlet基的生成方式和直线目标检测规则进行了改进的基础上,实现了道路标线直线边缘的快速检测。最后融合道路标线的直线边缘与区域灰度信息,将道路标线目标精确地分割出来。研究结果表明:采用改进的Beamlet算法能够准确地检测出路面图像中的标线边缘,对光照、阴影、裂缝、沥青补丁等噪声具有良好的抗噪性能,其处理时间分别为通用Beamlet算法和Hough变换算法的60.6%和7.7%;融合边缘和灰度的分割算法能精确地分割出标线区域,经过1 500张路面图像测试,其准检率(99.0%)明显高于动态阈值与全局阈值结合算法的准检率(73.8%);算法的检测效率和检测精度可满足道路检测车采集图像实时处理的需要。In order to improve the accuracy and efficiency of the identification algorithm for road marking, a quick and precise road marking segmentation algorithm based on Beamlet was pro- posed. Firstly, with bilinear interpolation algorithm, the original image with the resolution of 1 024×1 024 pixels was resized into a smaller one with the resolution of 64×64 pixels, then the Sobel operator was used to detect the step edges in the image. Secondly, a new Beamlet algorithm was utilized to improve the producing method of the Beamlets and the rules for the detection of the line objects, so as to fulfill the quick detection of line edges of road marking. Finally, the edge and grayscale information were fused to mark the road marking target accurately. The re- sults show that the improved Beamlet algorithm proposed in this paper can accurately detect the edge of the road markings, which has a good anti-noise performance to the uneven illumination, shadows, cracks, patches and other noises. The processing time of the proposed algorithm is60.6% and 7.7% of the general Beamlet algorithm and Hough Transform. The fused edges and grayscale can segment the road marking area precisely. After a test on 1 500 pavement images, the precision of the proposed algorithm（99.0%） is significantly higher than that of the combina- tive threshold algorithm（73.8%）. The accuracy and test efficiency of the proposed algorithm can meet the needs of the real-time processing of the images acquired on field. 2 tabs, 9 figs, 16 refs.

关 键 词：交通工程 路面养护 道路标线 图像分割 BEAMLET 边缘检测 融合分割 

分 类 号：U491.52[交通运输工程—交通运输规划与管理]