Search

The traditional LiDAR simultaneous localization and mapping (SLAM) algorithm is subject to large outlier interference and inaccurate keyframe matching. To address this a LiDAR SLAM algorithm using normal distributions transform (NDT) matching combined with two-step screening of key frames is proposed. The algorithm is built on light weight and groundoptimized LiDAR odometry and mapping (LeGO-LOAM). First, the point cloud is pre-processed using statistical outlier removal (SOR). The current frame is matched with its corresponding key frame for NDT and the matching result is added to the loop closure detection. Feature extraction is performed on the keyframe and the feature points are added to the loop closure detection and pose map. A two-step keyframe screening method is used in the loop closure detection module to effectively reduce map building errors. The proposed method was evaluated on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) dataset. Additionally, the experimental results show that the proposed algorithm can improve the mapping accuracy as measured by the value of the root mean square error (RMSE). [ABSTRACT FROM AUTHOR]

Current LiDAR systems are limited to vertical field of view (FoV) have a blind scanning region which leads to uncompleted three-dimensional (3-D) spatial reconstruction. To solve this problem this paper proposes a device mounted with a multi-layer LiDAR to collect data, which spins clockwise around the LiDAR's y-axis. Statistical outlier removal (SOR) and an optimized voxel-grid filter are employed to do the point cloud registration. The coarse registration and the fine registration are performed by 3-D normal distributions transform (NDT) and point-to-plane iterative closest point (ICP), respectively. The completed 3-D reconstruction is unrestricted by the LiDAR's vertical FoV. Additionally, the experimental results show that the proposed algorithm can improve the registration accuracy as measured by the value of the root mean square error (RMSE). [ABSTRACT FROM AUTHOR]