Abstract:
Visual SLAM (simultaneous localization and mapping) for scenes with dynamic objects is surveyed as a current research hotspot in robot navigation, automatic driving and other fields. Three research directions of dynamic SLAM are classified according to the different ways of handling the dynamic objects in localization and mapping, including dynamic robust SLAM and static background reconstruction, non-rigid dynamic object tracking and reconstruction, and moving object tracking and reconstruction. The three research directions are reviewed respectively, and the dynamic SLAM approaches combined with deep learning are highlighted. Finally, future development directions of dynamic SLAM are envisioned.