摘要
针对现有基于视觉的坑槽检测方法仅侧重识别、提取精度不高的缺点,提出了一种综合利用坑槽灰度特征和纹理特征的坑槽识别及提取方法。图像二值化后,根据形状特征和标准偏差对坑槽进行定性识别和初始提取;对检测出含有坑槽的图像,利用灰度共生矩阵提取纹理特征,并利用主元分析法剔除冗余纹理特征,通过模糊C-均值聚类算法将属于坑槽病害的纹理区域聚为一类,再叠加定性识别的初始提取结果,经形态学处理即可准确提取坑槽病害区域。该方法定性识别的召回率为90.0%,精确率为87.1%,准确率为92.2%,所提取的坑槽区域与原图像坑槽区域重叠度在80%以上的图像比率为70.4%,在70%以上的比率为85.2%。结果表明,该方法对于存在裂缝、碎石、积水等复杂状况的路面,都可取得良好的坑槽识别和提取效果。
Aiming at the present pavement pothole detection method only focused on pothole recognition,but could not extract the pothole accurately,this paper proposed a recognition and extraction method,which combined gray feature and texture feature of the pothole together.After image binaryzation,the algorithm applied shape feature and standard deviation to perform pothole recognition and the initial extraction.After pothole recognition,it calculated the texture feature vector of the detected pothole image by GLCM,and eliminated the redundant texture feature by PCA.Then it clustered the pothole texture by FCM algorithm,and integrated the clustered result with the initial extraction result of pothole recognition.At last,the algorithm extracted the pothole region through morphological operation.The recall rate of pothole recognition was 90.0%,the accurate rate was 87.1%,and the accuracy rate was 92.2%.The overlap rate between the extracted pothole region and the pavement image’s pothole region was estimated,and the image number ratio of overlap rate more than 80%was 70.4%,and the ratio of overlap rate more than 70%was 85.2%.The experiments show that this method can obtain favorable results for pothole recognition and extraction although the image contains cracks,gravel or water etc.
作者
王朋辉
胡永彪
田明锐
戴勇
Wang Penghui;Hu Yongbiao;Tian Mingrui;Dai Yong(National Engineering Laboratory for Highway Maintenance Equipment,Chang’an University,Xi’an 710064,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第5期1596-1600,共5页
Application Research of Computers
基金
中央高校基本科研业务费专项资金资助项目(310825165028
310825153313)
关键词
坑槽识别及提取
模糊C-均值聚类
纹理特征
主元分析
形态学
pothole recognition and extraction
fuzzy C-means clustering
texture feature
principal component analysis
morphological operation