IBLI-SLAM: Intensity-Based LiDAR-Inertial SLAM for Dynamic Environments

Conventional LiDAR SLAM algorithms are mostly based on the static environment assumption, and dynamic objects have a significant impact on the positioning and mapping performance. To address the problems of poor removal accuracy and low computation efficiency in current dynamic removal algorithms, a dynamic SLAM framework IBLI-SLAM is proposed. In the front-end, non-ground point clouds are transformed into Birds-Eye-View (BEV) intensity images to accomplish fast and accurate object segmentation. By detecting the motion of the segmented region centers, dynamic objects are subsequently selected. In the mapping session, the occupancy probability is constructed with intensity change rate. Then residual discrete dynamic point clouds on the road region are further detected and removed. The performance of the algorithm is evaluated widely with indoor and outdoor datasets. The experimental results show that the proposed IBLI-SLAM removes dynamic objects online with an average removal rate of ${9 2. 4 1 \%}$, and improves the positioning accuracy by ${4 7. 1 0 \%}$ in dynamic environments compared to the baseline algorithm.