3D Reconstruction Under Weak Illumination Using Visibility-Enhanced LDR Imagery

Abstract

Images of objects captured under poor illumination conditions inevitably contain noise and under-exposed regions where important geometric features may be hidden. Using these images for 3D reconstruction may impair the quality of the generated models. To improve 3D reconstruction under poor illumination, this paper proposes a simple solution for reviving buried features in dark images before feeding them into 3D reconstruction pipelines. Nowadays, many approaches for improving the visibility of details in dark images exist. However, according to our knowledge, none of them fulfills the requirements for a successful 3D reconstruction. Proposed approach in this paper aims not only to enhance the visibility but also contrast of features in dark images. Experiments conducted using challenging datasets of dark images demonstrate a significant improvement of generated 3D models in terms of visibility, completeness, and accuracy. It also shows that the proposed methodology outperforms state-of-the-art approaches that tackle the same problem.