LIVRe: A Filter-Based LiDAR-Inertial-Visual 3D Reconstruction

Multi-sensor fusion is widely used in 3D reconstruction to achieve both robustness and accuracy. This paper proposes a filter-based, tightly-coupled LiDAR-Inertial-Visual fusion framework termed LIVRe. Data from different sensors are organized in time order and update the system state independently, thereby enhancing the feasibility, efficiency and robustness of 3D reconstruction. To evaluate the quality of 3D pointcloud reconstruction, a set of evaluation criteria has been proposed. Experiments conducted on open datasets and with our handheld device demonstrate that LIVRe provides a real-time, accurate and dense pointcloud map, particularly in complex scenarios such as large-scale environments and strenuous movements. The framework could be widely applied to lightweight platforms with limited computing power, such as robots and drones.