基于像素-亚像素级形态分析的路面三维图像裂缝自动识别算法

Automatic Crack Detection Algorithm from 3D Pavement Images Based on Shape Analysis at Pixel-Subpixel Level

为了准确有效地检测路面裂缝,为路面性能评估、路面养护管理、路面结构和材料设计提供数据支撑,针对1 mm/像素路面三维图像提出了基于像素-亚像素级形态分析的裂缝自动识别算法。首先,应用Canny算法和区域生长算法检测候选裂缝目标并进行融合处理,得到融合分割图像;然后,提取并重构像素级与亚像素级图像骨架;最后,融合像素-亚像素级骨架图像,综合利用形态学算子和轮廓长度、圆度、扁平率等连通域形态特征提取裂缝目标。基于150张路面三维图像(992像素×992像素)对笔者算法和另外5种既有算法进行测试,结果显示,笔者算法获得了较高的准确率(均值90.45%)和召回率(均值96.49%),F均值由高至低分别为:笔者算法(90.72%)、种子并行生长算法(39.65%)、GAVILáN算法(33.46%)、各向异性测度算法(30.32%)、Canny检测(25.85%)和OTSU分割法(5.85%)。算法适用性分析表明,笔者算法较适用于细小裂缝图像识别,种子并行生长算法、GAVILáN算法和各向异性测度算法有利于宽而明显的裂缝识别,而Canny和OTSU通常可作为裂缝识别算法中的一个图像处理环节。

In order to detect pavement crack accurately and effectively as well as provide data basis for pavement performance evaluation,pavement maintenance and management,pavement structure and material design,an automatic pavement crack detection algorithm was proposed based on shape analysis at pixel-subpixel level aiming at 1mm/pixel 3D pavement images.Firstly,candidate crack targets were detected and fused by Canny algorithm and Region Growing method respectively,thus the fused segmentation image was obtained.Secondly,image skeletons at pixel level and subpixel level were extracted and reconstructed.Finally,image skeletons at pixel level and subpixel level were fused,and cracks targets were extracted by comprehensively applying morphological operators and shape features of connected components like contour length,roundness,flat-ratio,et al.Tests of the proposed algorithm and other 5 existed algorithms were conducted based on 150 3D pavement images(992 pixels×992 pixels).The results show that the proposed algorithm achieves relatively high precision(averaging 90.45%)and recall rate(averaging 96.49%),and average F-measure values rank as follow:the proposed algorithm(90.72%),the parallel seed growing method(39.65%),GAVIL N method(33.46%),the anisotropy method(30.32%),Canny detection(25.85%),OTSU segmentation(5.85%).Algorithm applicability analysis indicates that the proposed algorithm is suitable for recognizing tiny or thin cracks,the parallel seed growing method,GAVIL N method and anisotropy method have advantages in detecting wide and obvious cracks,and Canny and OTSU can be commonly utilized as an image processing unit of crack detection algorithms.