Pixel-Level Assessment of Pavement Cracks With Light-Weight Convolution Network and Metric-Skeleton Measurement Using High-Resolution Images From Low-Cost Camera

This study introduces a visual severity grading method for pavement cracks by using high-resolution images as inputs. The method incorporates a light-weight, fast crack segmentation network (CS-Net) to accurately extract cracks from images, and measure their lengths and widths with an image metric-skeleton model. Additionally, a five-level severity grading system is proposed to delineate the severity of crack lengths and widths. The CS-Net performs better than existing state-of-the-art networks on experiments, with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$F_{1-score}$ </tex-math></inline-formula>, intersection over union (IoU), frames per second (FPS), and model sizes of 78.66%, 65.25%, 55.00fps and 1.66MB, respectively. The metric-skeleton model precisely measures the crack with an overall mean error of approximately 0.30 pixels. According to maintenance guidelines, the distress grading level of pavement cracks is correspondingly assigned based on the length, width, and area values. The study findings can guide in allocating the maintenance budget for fixing highly defective road sections.