Tightly-Coupled SLAM Algorithm Based on LiDAR-Inertial-Visual Fusion

This paper proposes a tightly-coupled SLAM algorithm based on LiDAR-inertial-visual fusion for the problem of reduced localization accuracy or even failure of single-sensor SLAM algorithms in complex environments. The algorithm splits the processing unit based on the point cloud, and performs state prediction and motion compensation through IMU integration. The image photometric error and the LiDAR point-to-surfel distance residual are fused to update the state, and the map is constructed using both dense point cloud and octree. Experiments on the public NTU dataset and self-built simulation environment show that our proposed algorithm outperforms other similar algorithms with improved localization accuracy and well map building capability.