摘要
路面裂缝图像普遍存在光照不均匀,对比度低,噪声干扰严重等问题,传统图像处理方法无法有效检测路面裂缝。提出一种综合区域级和像素级特征的分步路面裂缝检测算法,将检测过程分为裂缝区域检测和区域内裂缝精确检测2部分依次进行。设计一种基于拟合图像背景的匀光算法,并结合其他预处理算法改善图像质量;而后基于图像连通域特征,采用基于多特征的滤波方法以及基于SVM的裂缝区域分类法完成裂缝区域检测;在裂缝区域内采用基于像素灰度与梯度特征的种子生长法提取裂缝目标。实验结果表明,该算法能够较好的检测和标识出复杂路面裂缝图像中的裂缝目标。
Low contrast, uneven illumination and severe noise pollution exist commonly in pavement crack images. As traditional image processing algorithms cannot detect the pavement cracks, a new split-step algorithm based on region-level and pixel-level features was proposed in this paper. The process of crack detection was divided into two steps, including crack region detection and precise crack detection in crack regions. First, a new image dodging algorithm based on fitting image background was designed and some other image preprocessing methods were utilized together to improve the quality of pavement images. Second, according to the features of connected regions, a multiple-staged filtering method and a crack region classification algorithm based on SVM were proposed to detect the crack regions. At last, a seeds growing method based on the pixel grayscale and gradient information was used to abstract cracks from the detected crack regions. The experiment results show that this algorithm can effectively detect and mark crack objects from the complex pavement images.
作者
韩锟
韩洪飞
HAN Kun;HAN Hongfei(School of Traffic and Transportation Engineering, Central South University, Changsha 410075, China)
出处
《铁道科学与工程学报》
CAS
CSCD
北大核心
2018年第5期1178-1186,共9页
Journal of Railway Science and Engineering
基金
湖南省自然科学基金资助项目(2016JJ4117)
关键词
裂缝检测
图像处理
连通区域
特征提取
种子生长
crack detection
image processing
connected region
feature extraction
seeds growth