ABSTRACT
Cracking is one of the main defects of pavement deterioration, which is caused by the environment and pressure. Automated pavement crack detection is being developed to obtain the number, classification and location of pavement cracks in images through object detection, which is a challenging task because of the irregular pattern and the noises on pavement. However, most of the previous methods are more related to object segmentation, which segments crack pixels from pavement background. As a result, an automated pavement crack detection method was proposed based on region-based convolutional neural network (R-CNN) like you only look once version three (YOLOv3) in this paper, whose network architecture was modified to fit the special features of pavement crack detection. A dataset of pavement crack images was built, which was used to train and test the proposed method. Results show that precision, recall and F1 score of proposed method are 91.95%, 89.31% and 90.61% respectively, which are higher than other state-of-the-art pavement crack detection methods.
