|
Automatic Crack Detection Algorithm from 3D Pavement Images
Based on Shape Analysis at Pixel-Subpixel Level
PENG Bo1, HUANG Darong1, GUO Li2, CAI Xiaoyu1, LI Shaobo1
2018, 37(09):
34-42.
DOI: 10.3969/j.issn.1674-0696.2018.09.06
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%), GAVILN 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, GAVILN 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.
References |
Related Articles |
Metrics
|