Seeing in the Dark: Low-Light Image Enhancement for Modular Construction

Abstract: Modular construction uses off-site modules for on-site assembly; computer vision supports progress monitoring, inspection, and assembly. These tasks may occur in poorly lit spaces, where low light degrades image quality, while prior work has not addressed low-light enhancement for this setting. We propose RetinexRefiner, a two-stage framework that first performs illumination-based enhancement and then refines the result to suppress residual artefacts such as colour cast and noise. Trained end-to-end, it achieves improved visibility on night construction images, reaching 28.07 dB PSNR and 0.83 SSIM, and is applicable to construction sites and similar dim environments.