Research On 3D SLAM Technology Based On Multi-source Information Fusion

Simultaneous Localization and Mapping(SLAM)technology is widely used in the field of autonomous navigation for unmanned vehicles,which is used to solve the problem of constructing a scene map in a new environment and localizing the position of the robot on the map.In recent years,many studies have begun to consider fusing information from multiple sensors to overcome the shortcomings of using a single sensor,to improve the robustness and accuracy of SLAM algorithms in a complex environment.The main purpose of this paper is to improve the practicability of the 3D SLAM algorithm for unmanned vehicles,from the perspective of static map construction and multi-source information fusion.There are two problems in the existing 3D SLAM algorithm: one problem is that dynamic objects will leave traces on the map during the process of mapping in dynamic scenes,which not only produce wrong data associations to the positioning process but also affect the obstacle recognition of the navigation algorithm.Another problem is that the existing vision and lidar fusion SLAM optimizes the visual feature and the Lidar feature separately,and then fuses the camera pose and the lidar pose in the pose graph,but camera pose and the lidar pose are related when sensors are connected rigidly.Based on this background,this article accomplishes the following work:1.Aiming at the problem of constructing static maps in dynamic scenes,a new Lidar SLAM algorithm for dynamic 3D scenes is proposed.This algorithm combines the mobile filtering module with Lidar SLAM,by combining object tracking and scene flow methods to filter out moving objects in the map,and finally builds a static map of the environment.2.Aiming at the optimization problem of visual features and point cloud features,an interframe odometer based on stereo camera and Lidar fusion is proposed.The constraint functions united in the same robot coordinate system are designed for the two different features so that the algorithm can optimize the two kinds of features at the same time.And uses the parameter automatic adjustment method to adjust the number of point cloud feature points and the optimization time,and finally estimates the pose changes of the robots between two frames.Experiments show that the above two methods improve the robustness of the SLAM system,meet real-time requirements,and have better practical value.