Study Of 3D SLAM Based On A Multi-line LiDAR And An IMU

In recent years,in order to save manpower and improve productivity,the demand for unmanned logistics vehicles and automated guided vehicles(AGVs)in the logistics industry and factory workshops has increased,and autonomous vehicle products have emerged and are being developed rapidly.In these applications,mobile carriers need determine their location in the environment in real time and understand what the surrounding scene looks like.Simultaneous Localization and Mapping(SLAM)technology is the key to solving the task.This paper studies the SLAM technology that is based on a multi-line LiDAR and applied to three-dimensional space,and uses IMU(Inertial Measurement Unit)sensor data to provide initial values of pose(position and attitude)transformation for the registration method,and builds a 3D SLAM program based on a multi-line LiDAR and an IMU.This paper proposes a fused point cloud feature used to registered with the occupancy grid submaps.The feature consists of edge feature points and plane feature points.The method and the adaptive filtering method both aim at reducing the number of three-dimensional points.Experiments show that compared with the adaptive filter downsampling method,the feature extraction method has the similar speed and meets the real-time requirement.Both methods can control the number of processed 3D points to some extent.The difference is that this feature method makes the odometry registration error less and repetition between the point cloud and the submap more.Thus,it is less dependent on the pose of the LiDAR.This paper proposes an odometry method based on registering multi-resolution occupancy grid submaps with feature points.The feature points are first registered with the low-resolution grid submap,and the obtained pose transformation is used as the initial value of the pose transformation for the feature point and the high-resolution submap registration.Compared with the method for registering feature points with the single-resolution submap,the method effectively improves the initial value of the registration process,alleviates the registration result to fall into the local optimum earlier.Aiming at the interference of cumulative error to the loop-closure detection method,this paper proposes a loop-closure detection method combining the branch and boundary and the random sampling method.The method combines advantages that the branch and bound method can detect the close loop-closure constraints effectively and the random sampling method can detect the long-distance loop-closure constraints with a certain probability.The method can overcome the cumulative error's interference to the loop-closure detection method,and thus ensure the accuracy of the map.In summary,this paper builds the odometry method by registering point cloud edge and plane feature points with the multi-resolution occupancy grid submaps,Using the loop-closure detection method which combines the branch and boundary and the random sampling method,and the pose graph optimation method to reduce the cumulative error of the odometry results.At last,this paper finishs building a 3D SLAM program.Using the public dataset and the real-world scene data to construct the point cloud maps and the two-dimensional occupancy grid map respectively,which intuitively shows effectiveness of this SLAM program.Using the ground truth of a public dataset to calculate errors of the trajectory estimated by this 3D SLAM algorithm.Experiments show that this threedimensional SLAM program has good accuracy under the low speed motion condition(motion speed is from 1m/s to 2m/s).