Research On SLAM Fusion Of Vision And Lidar In Indoor Dynamic Environment

With the continuous progress of science and technology,intelligent mobile robots have gradually entered production and life,and Simultaneous Localization and Mapping(SLAM)is one of the core technologies of mobile robots.With the increasingly complex usage environment,the existing 2D laser SLAM has poor map quality when facing dynamic environments due to the interference of dynamic objects.At the same time,laser SLAM has the problem of decreased positioning accuracy in structurally simple environments.In addition,the environment map constructed by 2D laser radar has a certain degree of structural deficiency,while vision-based SALM can effectively use environmental texture information despite problems such as susceptibility to lighting changes and poor mapping accuracy.Based on the Cartographer algorithm and combined with visual SLAM,this paper studies a visual and laser fusion SLAM method in indoor dynamic environments.The main research contents include dynamic target filtering based on laser radar,fusion positioning,fusion map construction,and navigation.A dynamic target detection and filtering method based on clustering and motion relationship are investigated to reduce the impact of dynamic targets on map construction.The study of dynamic target detection and filtering is carried out for the existing laser SLAM that does not take into account the influence of dynamic objects on map construction-,and constructs maps that contain dynamic objects.The dynamic targets in the environment are pre-detected by the Euclidean clustering method,and the correspondence between the clustered targets in the adjacent point cloud frames is established using the point cloud shape descriptors and the position relationships.Finally,the clustered targets are combined with motion relations to constrain the removal of dynamic targets.Experiments were conducted in an indoor environment,and the results showed that the improved algorithm effectively removed dynamic objects from the environment in a single dynamic target environment,and was able to remove dynamic targets from the environment most of the time in a dual dynamic target environment,but a small number of dynamic targets were not removed due to occlusion.Aiming at the problem that single-sensor LiDAR SLAM cannot locate effectively in a single environment with a single structure,a method of vision and LiDAR fusion localization is studied to improve the localization accuracy of the SLAM system.The constructed synchronization and communication method is used to realize the communication synchronization between the visual odometer and the radar odometer calculated in parallel,to detect the validity of the odometer based on the environmental information and the odometer speed difference,and finally to realize the fusion of visual and LiDAR localization information based on the extended Kalman filtering method.The experimental results show that the improved algorithm effectively improves the localization accuracy of the SLAM system in the environment of long straight path with a single structure,and in the environment of combined building and long straight path,the improved algorithm can still maintain good localization throughout the whole process,although the improvement of localization accuracy decreases.The map structure constructed by single-line laser mine is not complete enough to provide correct information for robot navigation.A map construction method based on vision and laser radar fusion was studied.By screening the key frames of the mapping,the 3D point cloud map of the space is constructed by using visual information.The raster maps are generated by vision and lidar information respectively,and the two raster maps are fused according to the occupancy state of the raster.Through the Adaptive Monte Carlo Localization algorithm and A*path planning algorithm,an autonomous navigation method for planar mobile robots is built on the ROS system.Experiments in different environments show that the constructed fusion grid map can describe the environment structure more completely,and the robot autonomous navigation method can complete path planning and autonomous movement on the constructed grid map.