Under rapid motion conditions, such as high-speed and vigorous angular velocity, motion distortion in Light Detection and Ranging (LiDAR) point cloud will be severe and reduce the accuracy of LiDAR Inertial odometry (LIO). To address this issue, a novel algorithm based on Multi Output Gaussian Process Regression (MOGPR) is proposed in this work to eliminate motion distortion. Firstly, the Single Sample plus Previous Sample (SSPS) algorithm is used to obtain an accurate vehicle state at the same rate as the raw Inertial Measurement Unit (IMU) data. Next, MOGPR is applied to obtain interpolated states that correspond to the moment of each point in the LiDAR point cloud. The motion distortion can be eliminated by aligning all points in the same LiDAR point cloud to a common reference coordinate frame, using the interpolated states as a guide. Finally, the corrected LiDAR point cloud and IMU data are fused using a factor graph, and the IMU bias is estimated to compensate the raw IMU data, further improving the accuracy of motion distortion elimination. Experiments are conducted on sumulation data, the public KITTI dataset and our platform, demonstrating the effectiveness of the proposed method in motion distortion elimination compared to existing methods. Additionally, the localization accuracy of the LIO algorithm integrating the proposed method is improved by approximately 30%.

For most existing simultaneous localization and mapping (SLAM) systems, the surroundings of autonomous vehicle are assumed to be static, which is impractical in urban environments and compromises the effectiveness of existing SLAM algorithms. In this work, a dynamic-object-aware light detection and ranging (LiDAR) visual inertial odometry (LVIO), termed Dynam-LVIO, is introduced to enhance the accuracy of SLAM in dynamic environments by constructing both environment map and object map. Initially, reprojection error and iterative closest point (ICP) error are calculated with object map, serving as observations for the error state iterated Kalman filter (ESIKF) to accurately estimate the object state on manifold. Subsequently, the 2-D bounding boxes derived through YOLO-V5 are tracked with the proposed multiobject tracking (MOT) algorithm, LVI-SORT, to achieve stable MOT in complex scenes. Specifically, to improve the accuracy of MOT in fast moving scenes, the 2-D object flow, calculated with object state, vehicle state and object map, is used to predict object state in the prediction process of LVI-SORT. Furthermore, to mitigate MOT failures caused by temporary object occlusion, a hybrid object association is proposed within the association process of LVI-SORT, which is accomplished by incorporating object map points and intersection-over-union (IoU) of bounding boxes to form the cost matrix in the Hungarian algorithm. Finally, the tracked 2-D bounding boxes are leveraged to segment the recent global map into environment map and object map, thereby reducing the impact of dynamic objects on LiDAR visual inertial SLAM. Compared with other benchmark algorithms, the results indicate that the proposed algorithm can enhance localization accuracy by 5%–10% in dynamic scenarios. Concurrently, the proposed algorithm also enhances object tracking accuracy by nearly 3%.

To enhance localization accuracy in urban environments, an innovative LiDAR-Visual-Inertial odometry, named HDA-LVIO, is proposed by employing hybrid data association. The proposed HDA_LVIO system can be divided into two subsystems: the LiDAR-Inertial subsystem (LIS) and the Visual-Inertial subsystem (VIS). In the LIS, the LiDAR pointcloud is utilized to calculate the Iterative Closest Point (ICP) error, serving as the measurement value of Error State Iterated Kalman Filter (ESIKF) to construct the global map. In the VIS, an incremental method is firstly employed to adaptively extract planes from the global map. And the centroids of these planes are projected onto the image to obtain projection points. Then, feature points are extracted from the image and tracked along with projection points using Lucas-Kanade (LK) optical flow. Next, leveraging the vehicle states from previous intervals, sliding window optimization is performed to estimate the depth of feature points. Concurrently, a method based on epipolar geometric constraints is proposed to address tracking failures for feature points, which can improve the accuracy of depth estimation for feature points by ensuring sufficient parallax within the sliding window. Subsequently, the feature points and projection points are hybridly associated to construct reprojection error, serving as the measurement value of ESIKF to estimate vehicle states. Finally, the localization accuracy of the proposed HDA-LVIO is validated using public datasets and data from our equipment. The results demonstrate that the proposed algorithm achieves obviously improvement in localization accuracy compared to various existing algorithms.