Deep Learning based Pavement Crack Detection System

Abstract

Abstract The pavement crack causes the highway service life to shorten, the safety hidden danger to increase. The low efficiency and high cost of manual inspection makes it difficult to detect pavement cracks. This paper proposes a fast and efficient deep learning pavement crack detection system. CRACK2000, an image segmentation dataset with complex interference background and multiple crack types, is constructed based on perspective transformation and image cropping. The scheme corrects the pavement crack images by perspective transformation. The extraction of pavement crack depth features is completed by applying the U-Net network. Finally, the pavement condition index PCI (pavement condition index) is calculated by quantifying the different types of crack information based on the segmentation results. The experimental results show that the Precision, Recall, F1-score and AUC of the U-Net network are 76.67%, 72.32%, 74.43% and 99.46% respectively. The AUC values reflect that the method is more capable of filtering out complex background interference from cracked images. The automatic pavement crack detection system designed in this paper can accurately locate and classify the location and category of pavement cracks, and perform quantitative pavement evaluation to obtain the pavement deterioration of the road section and the corresponding repair recommendations, enhancing the practicality of pavement crack detection.