ABSTRACT

A new crack detection approach based on local binary patterns (LBP) with support vector machine (SVM) was proposed in this paper. The propsed algorithm can extract the LBP feature from each frame of the video taken from the road. Then, the dimension of the LBP feature spaces can be reduced by Principal Component Analysis(PCA). The simplified samples are trained to be decided the type of crack using Support Vector Machine(SVM). In order to reflect the directional imformation in detail, the LBP processed image is devided into nine sub-blocks. In this paper, driving tests were performed 10 times and 12,000 image data were applied to the proposed algorithm. The average accuracy of the proposed algorithm with sub-blocks is 91.91%, which is about 6.6% higher than the algorithm without sub-blocks. The LBP-PCA with SVM applying sub-blocks reflects the directional information of the crack so that it has high accuracy of 89.41% and 88.24%, especially in transverse and longitudinal cracks. In the performance analysis of different crack classifiers, the F-Measure, which considered balance between the precision and the recall, of alligator cracks classifier was the highest at 0.7601 and hence crack detection performance is higher than others.

摘要

本文提出了一种基于局部二值模式（LBP）和支持向量机（SVM）的裂纹检测方法。所提出的算法可以从道路上拍摄的视频的每一帧中提取 LBP 特征。然后，可以通过主成分分析（PCA）来降低LBP特征空间的维度。使用支持向量机（SVM）对简化样本进行训练以判断裂纹类型。为了更详细地反映方向信息，将LBP处理后的图像分为九个子块。在本文中，进行了 10 次驾驶测试，并将 12,000 个图像数据应用于所提出的算法。所提出的有子块算法的平均准确率为91.91%，比没有子块的算法提高了约6.6%。带有支持向量机应用子块的LBP-PCA反映了裂纹的方向信息，使其具有89.41%和88.24%的高精度，特别是在横向和纵向裂纹中。在对不同裂纹分类器的性能分析中，考虑精度和召回率平衡的鳄鱼裂纹分类器的F-Measure最高，为0.7601，因此裂纹检测性能高于其他分类器。