Automatic quantitative recognition method for vertical concealed cracks in asphalt pavement based on feature pixel points and 3D reconstructions

Non-destructive detection and characterisation of concealed cracks in asphalt pavements is a key aspect of ground-penetrating radar (GPR) road maintenance. This paper proposes a method based on feature pixel points to quantify and calculate the vertical height of cracks. Firstly, the team used gprMax to conduct numerical simulations to study the GPR image characteristics of concealed cracks in asphalt pavement with varying lengths and widths. Subsequently, the relationship between the pixel value of the crack area and the two-way travel time was established to obtain the relationship between the vertical height of the crack and the pixel. This study combined with the deep learning model (YOLOv5) allows for the calculation of the vertical height of a crack while simultaneously recognizing it, with the minimum error being only 1.3 %. Finally, concealed cracks were visualized in three-dimensional(3D) by slicing the results of vertical crack simulations of asphalt pavement and observing cloud images, and the principle of computed tomography(CT) was employed to reconstruct 3D models of cracked asphalt pavement and estimate the vertical height of the cracks. This method achieved a minimum error of only 2.9 %. The research presents a theoretical framework for recognizing and more intuitive characterizing concealed cracks in asphalt pavement accurately, enabling it to be used in practical engineering applications.