Research On Mapping And Localization Algorithm Based On 3D Lidar And IMU Tightly Coupled Filter Fusion

Autonomous navigation is the key technology of intelligent mobile robots.The basis of autonomous navigation is that the robot can obtain the environment map and its position information in real time and accurately.The mapping and localization technology based on multi-sensor fusion has the advantage of high accuracy and good robustness.3D lidar and inertial measurement unit are highly complementary and reliable,and they are the focus of multi-sensor fusion research.At present,the mapping and localization algorithm based on 3D lidar and IMU fusion still has the following problems: 1)the information fusion of the two sensors is not tight and the information utilization is not sufficient;2)the accuracy and efficiency of mapping and localization algorithm need to be improved.In view of the problems existing in the algorithm of 3D lidar and IMU fusion,this paper aims to realize the dense map construction and real-time localization of robots in outdoor scenes.The main contents of this paper are as follows:(1)The odometry based on the 3D lidar and IMU fusion is studied,and a lidar inertial odometry based on tightly filter fusion is proposed.The iterative unscented Kalman filter based on modified Rodrigues parameters is used to fuse the data of 3D lidar and IMU;the GICPSO3 algorithm based on lie algebra optimization is used to match the point cloud and get the pose of the current frame of point cloud;according to the optimized pose,the current frame of point cloud is added to the octree point cloud map to complete the construction of the point cloud map.The experiment results show that the average absolute error of our algorithm is 1.88 m,which is28.4% of LOAM algorithm.(2)Loop closure detection and closed-loop optimization in mapping and localization technology are studied.A loop detection and closed-loop optimization algorithm based on distance and intensity fusion point cloud descriptor is proposed.This algorithm uses a point cloud descriptor which fuses the distance information and reflection intensity information of point cloud to detect the loop closure efficiently and accurately;the coarse matching algorithm of point cloud based on the four points congruent set which does not depend on the initial value is improved,which provides the initial value for the subsequent fine matching of point cloud and improves the accuracy of solving the relative pose of the loop frames;deduces the optimization algorithm of pose map,and optimizes the trajectory obtained from the odometry algorithm.The experimental results show that the algorithm proposed in this paper can accurately detect the loop and optimize the trajectory.The absolute trajectory error after optimization is 35.7% of that before optimization.(3)The global and local localization algorithm based on 3D lidar and IMU fusion are studied.A global localization algorithm based on multihypothesis particle filter is proposed.The algorithm makes full use of the information of point cloud descriptors to determine the possible initial poses of particles,which reduces the calculation amount and improves the localization accuracy in repeated scenes.Taking advantage of the lidar inertial odometry algorithm in Chapter 2,a new algorithm which is suitable for the local localization of 3D point cloud map is proposed.The experiments show that the global localization algorithm has a success rate of more than 90%.The average error of local localization algorithm is 2.4cm,and the maximum localization error is less than 10 cm,which meets the demand of decimeter localization.