Urban Scene Reconstruction

Urban scene reconstruction aims to create detailed 3D models of cities from various data sources, primarily images and LiDAR scans, enabling realistic simulations and novel view synthesis. Current research focuses on improving efficiency and scalability using neural radiance fields (NeRFs) and Gaussian splatting, often incorporating techniques like incremental view selection and federated learning to handle large datasets. These advancements are significant for applications in autonomous driving, urban planning, and virtual/augmented reality, providing accurate and photorealistic representations of complex urban environments.