The Design And Implementation Of SLAM System Based On Fusion Of Vision And Laser

Mobile robots need to maintain the basic ability of environmental awareness and self-positioning,which can be achieved through simultaneous positioning and mapping(SLAM)technology.At present,lidar and camera are the most commonly used environment-aware sensors for robots,respectively forming two technical routes: lidar SLAM and visual SLAM.Lidar SLAM relies on high-speed,high-resolution lidar ranging to obtain sufficient accuracy,and visual SLAM has a larger optimization space based on rich image features.In order to make full use of the advantages of lidar and vision,this thesis proposes a SLAM system based on fusion of vision and lidar.The system is based on Racecar's unmanned vehicle hardware platform and uses a low-cost sensor solution,including a monocular global shutter camera,2D triangular lidar,and inertial measurement unit(IMU).The software design is based on the ROS robot operating system and follows the SLAM positioning and mapping requirements.The camera and IMU are tightly coupled to form a high-precision visual inerometer(VIO)to achieve indoor positioning.Match to generate a flat map.According to the overall framework of SLAM system positioning and mapping,the system is divided into VIO initialization module,pose estimation module,map generation module and global optimization module.Vision and lidar fusion are mainly used in VIO initialization and map generation modules.At the initialization stage of visual inertial odometer,lidar matching data is introduced to align the visual image SFM pose estimation,IMU pose estimation and lidar scan matching pose estimation,and the visual inertial odometer initialization and camera-lidar external parameter calibration are performed respectively.The map generation module uses real-time visual inertial odometer data as the lidar's estimated pose for lidar point cloud scanning matching.The experimental results of the system in offline data sets and Racecar unmanned vehicles show that the SLAM system based on the fusion of vision and lidar has higher accuracy advantages than the 2D lidar SLAM using only a single lidar sensor or using the code wheel odometer.,Under the condition of low-cost sensors,the SLAM task can be completed well,and a high-precision planar grid map can be output.