Reconstruction of sub‐mm 3D pavement images using recursive generative adversarial network for faster texture measurement

Abstract

It is challenging to collect 3D pavement images with desired resolution for accurate texture measurement at driving speeds with current devices, particularly in the longitudinal direction. This paper presents a novel superresolution technique with recursive generative adversarial network, called Pavement Texture Super Resolution Generative Adversarial Network (PT‐SRGAN), to reconstruct 0.1‐mm pavement 3D image from low‐resolution data for faster texture measurement. With the proposed pseudo‐Laplacian pyramid and an array of learning strategies, the developed PT‐SRGAN reconstructs 0.1‐mm 3D texture images with multiple upscaling factors in longitudinal direction. Combined with the evaluation mask, the proposed method is substantially superior to other methods in terms of three metrics when comparing the image quality of reconstructed 0.1‐mm 3D images against ground truth. The preliminary results indicate that the proposed method enables data collection at driving speeds up to 24 mph to collect 3D pavement images at sub‐mm resolution for faster texture measurement.

An integration–competition network for bridge crack segmentation under complex scenes

“…With the development of computer vision technology, image-based non-destructive detection technology has © 2023 Computer-Aided Civil and Infrastructure Engineering. become a research hotspot of bridge crack detection (Jang et al, 2021;Meng et al, 2023;Siriborvornratanakul, 2023;G. Wang et al, 2023).…”

“…With the development of computer vision technology, image‐based non‐destructive detection technology has become a research hotspot of bridge crack detection (Jang et al., 2021; Meng et al., 2023; Siriborvornratanakul, 2023; G. Wang et al., 2023). The common digital image methods are the minimum cost path search method (Amhaz et al., 2016), 3‐dimensional multi‐feature detection method (Qiu et al., 2020), K‐means cluster segmentation method, and so forth (P. Li et al., 2020).…”