Using full-lane 3D pavement data, this paper proposes an automated and systematic approach to efficient measurement of pavement faulting occurring at transverse joints. First, a Fully Convolutional Network (FCN) is proposed in the paper to detect pavement joints. The experimental results demonstrate that the proposed FCN outperforms FCN-VGG16 and U-Net in terms of detection accuracy, but results in a slower processing speed due to the increased number of hidden layers. Compared with the 3D Shadow Modeling, the proposed FCN shows significant improvements in terms of both accuracy and time efficiency. The precision, recall and F-measure achieved by the proposed FCN on 1130 testing images are 92.64%, 97.14% and 94.83% respectively. Based on detection outputs of the proposed FCN, this paper applies Hough Transform only at orientations feasible for transverse joints. Such a strategy helps retain transverse joints for analysis, and eliminate unneeded longitudinal joints. Finally, this paper proposes a practical method to conduct comprehensive faulting measurements along the entire detected transverse joint, while field measurements today are generally on limited spots on the joint. Meaningful statistical indicators are also recommended in the paper to describe the general faulting condition along a transverse joint. The field test on a jointed concrete pavement reveals that the proposed automated approach is comparable in accuracy with manual investigation. Measurement errors at most testing locations are within ± 2 mm, and the average absolute error with respect to all testing locations is 1.2 mm. Compared with manual means or automated methods using only a few longitudinal profiles, the proposed automation of faulting survey achieves the following improvements: detecting transverse joints to the full-lane width through full-lane 3D pavement data as well as deep learning technology, and measuring faulting values along the entire transverse joint in a more efficient and more complete manner.

利用全车道3D路面数据，本文提出了一种自动，系统的方法来有效测量横缝处发生的路面断层。首先，本文提出了一种全卷积网络（FCN）来检测人行道缝。实验结果表明，所提出的FCN在检测精度方面优于FCN-VGG16和U-Net，但由于隐藏层数的增加，导致处理速度较慢。与3D阴影建模相比，拟议的FCN在准确性和时间效率方面均显示出显着改进。所提出的FCN在1130张测试图像上实现的精度，召回率和F量度分别为92.64％，97.14％和94.83％。根据拟议FCN的检测结果，本文仅在横向接头可行的方向上应用霍夫变换。这种策略有助于保留横向关节以进行分析，并消除不必要的纵向关节。最后，本文提出了一种实用的方法，可以对整个检测到的横向关节进行全面的断层测量，而如今的现场测量通常仅在关节的有限位置进行。本文还建议使用有意义的统计指标来描述横缝的一般断层情况。在节理的混凝土路面上进行的现场测试表明，所提出的自动化方法的准确性与手动调查相当。大多数测试位置的测量误差在±2 mm以内，所有测试位置的平均绝对误差为1.2 mm。