Dense Simultaneous Localization

Dense simultaneous localization and mapping (SLAM) aims to create detailed 3D models of environments while simultaneously tracking a camera's position within them, using visual input (RGB, RGB-D). Recent research focuses on improving accuracy and efficiency through novel scene representations, such as neural implicit fields (including NeRFs), Gaussian splatting, and point clouds, often incorporating techniques like global bundle adjustment and uncertainty learning to enhance robustness. These advancements are driving progress in robotics, augmented reality, and other applications requiring accurate and real-time 3D scene understanding.