基于深度学习的道路表面裂缝检测方法

道路表面裂缝检测是道路安全检测的一项重要指标,随着计算机视觉技术的不断发展,越来越多的基于深度学习的目标检测方法被应用到裂缝检测中.然而这些检测方法大都是对裂缝位置的粗略的检测或分类,无法定量的衡量裂缝为了定量衡量裂缝,提出了一种基于深度学习的像素级道路表面裂缝检测方法.使用卷积神经网络对裂缝原始图像进行分割得到裂缝的二值化图像,并实现对裂缝的面积、长度、平均宽度的自动计算,提高了测量效率并降低了检测成本.实验表明,本系统对裂缝面积测量准确率达到93％,长度测量准确率达到92％,平均宽度测量准确率达到89％.

Road surface crack detection method based on deep learning

Road surface crack detection is used as important indicator of road safety detection. With the continuous development of computer vision technology, it is showed that a growing number of detection methods based on deep learning have been applied to crack detection. However, most crack locations can be only detected roughly or classified by these detection methods,and cannot be quantitatively measured. As a result of the fact, a pixel-level road surface crack detection method based on deep learning is proposed in this paper so that cracks can be measured quantitively. Convolutional neural network is applied to segment the original crack image thus the binary image of the crack can be obtained. The area, length and average width of the crack can be calculated automatically, which improves efficiency and reduces cost. It is demonstrated in experiments that the accuracy rate of this system for the measurement of crack area reaches 93%, the accuracy rate of length measurement 92%, and the accuracy rate of average width measurement reaches 89%.