Research On SLAM Method Based On Lidar And Vision Fusion

Simultaneous Localization and Mapping(SLAM)has always been a research hotspot in robotics,of which laser and visual SLAM are the two most commonly used methods.However,with the rapid development of the robotics industry,only using a single sensor has great limitations.Although the single-line laser SLAM commonly used in indoors has high precision,it can only collect environmental information on a certain height plane.The vision based on RGB-D SLAM can obtain rich environmental information,but it is easily disturbed by light.In view of the above problems,this paper combines the advantages of the two sensors and uses the indoor mobile robot as a platform to fuse laser and visual information.The two-dimensional grid map is obtained by lidar,and the threedimensional information in the environmental space is obtained by the depth camera,and then the obtained three-dimensional environmental map is fused with the twodimensional laser grid map by the Bayesian method.The main research contents include the following aspects:Firstly,the ROS operating system of the mobile robot platform is introduced,the kinematics model of the robot under the ROS system is mathematically analyzed,the establishment process of the single-line lidar model and the camera model is introduced in detail,and completed the camera distortion removal and calibration.Secondly,in the process of laser SLAM,the point cloud information scanned by the single-line lidar is relatively discrete.Based on this,the commonly used point cloud filtering methods are analyzed.Aiming at the problem of poor accuracy of the filtering methods,an improved method is proposed.Gaussian filter method.The simulation results show that the method can effectively smooth the discrete noise in the point cloud.When the lidar mapping experiment is carried out in the actual environment,it also shows that the two-dimensional grid map constructed by the improved method is more accurate.Then,the depth camera is used to collect environmental information affecting the range of the robot,and a 3D dense point cloud map is constructed on the basis of the ORB-SLAM2 algorithm,using octomap method to convert 3D dense point cloud map to 2D projection into raster map.The mobile robot can build a two-dimensional grid map with rich environmental information in an indoor environment through experiments.Finally,a fusion SLAM method is implemented using lidar and depth cameras.The multi-sensor data fusion uses Bayesian estimation to fuse the raster map constructed based on the ORB-SLAM2 method and the improved Gaussian filtering method.On the constructed robot experiment platform,taking the actual indoor environment as the application scenario,a comparison experiment between laser,vision and fusion SLAM is carried out.The experimental results show that the raster map obtained by the fusion SLAM method has higher accuracy and contains more environmental information.