Lighting Estimation in Outdoor Scenes

This dissertation, titled "Lighting Estimation in Outdoor Scenes," addresses the complexities of outdoor lighting in computer vision, crucial for augmented reality, 3D reconstruction, and autonomous driving. Traditional single-image lighting estimation methods often struggle with noise and natural element interplay. This research proposes new methodologies incorporating spatial and temporal lighting analyses for more accurate and realistic interpretations of outdoor scenes. The thesis makes two significant contributions: first, it explores intrinsic image decomposition and lighting estimation through a U-Net architecture, providing valuable insights for future research. Second, it introduces advanced spatio-temporal lighting estimation methodologies, including a four-stage method and a Transformer-based model for robust global sun direction estimation, advancing lighting estimation for real-world applications.