Dynamic Voxel-Based Lidar Inertial Fusion SLAM Algorithm for Denial Scenarios

To solve the problem of vehicle location mapping limitation in urban canyons, tunnels, and other low-texture, low-light, and high-dynamic denial scenes, a dynamic voxel-based lidar inertial fusion SLAM algorithm DSC-Algo for denial scenes is proposed. The algorithm introduces a dynamic voxel data structure to maintain the local map and nearest neighbor search, which can efficiently insert and delete new lidar points, to improve the computational efficiency of the algorithm. At the same time, the pose optimization method of factor map was used to introduce the loop detection factor of scan-context descriptors that could reflect the three-dimensional structure characteristics of the point cloud, and the objective function including odometer factor, IMU pre-integration factor and the error term of scan-context environment loop detection factor was constructed. The optimal pose state was solved by a nonlinear optimization algorithm, and the positioning accuracy was improved. A complete SLAM algorithm framework is constructed by odometry algorithm, factor graph optimization, and loop closure detection. Compared with the mainstream lidar inertial SLAM algorithm FAST-LIO2, the computational efficiency of the proposed DSC-Algo algorithm in the iterative Kalman filtering stage and point cloud distortion removal stage is improved by nearly 40%. The test results of real vehicles in the reality denial scenario of the underground parking lot and urban tunnel show that DSC-Algo algorithm has higher positioning accuracy and more accurate mapping effect than the FAST-LIO2 and LeGO-LOAM algorithms.