Research On AGV SLAM Algorithm Based On Lidar And Vision Fusion And Path Planing

With the continuous improvement of national science and technology and economic level,it is urgent for China to transform from a manufacturing country to a manufacturing power.Automated Guided Vehicle(AGV),as an important tool for intelligent manufacturing,is widely used in manufacturing,logistics handling and home services.It is conducive to reducing production costs,improving production efficiency,enhancing industrial automation and it can help reduce production costs,improve production efficiency,and enhance industrial automation and intelligent level.The key technologies to achieve autonomous guidance of AGV are SLAM and path planning.There are limitations of SLAM with a single sensor.Multi-line lidar has a high price and single-line lidar can only detect two-dimensional plane information.Binocular camera has a low accuracy in map construction and it is sensitive to light.Therefore,this paper adopts single-line lidar and binocular camera,aiming to study the laser and vision fusion AGV SLAM system,improve the construction of the environment map,and perform path planning.The main research contents of this paper are as follows:Firstly,the model of AGV system is established.The Ackermann model and odometer model of AGV are constructed,and the observation models of single-line lidar and binocular camera are established.The commonly used grid map,feature map and topology map in the SLAM process are explained.The advantages and disadvantages of these three map expression methods are compared,and the grid map is selected as the map model in combination with the requirements of this paper.The structure and characteristics of Robot Operating System(ROS)are introduced.Secondly,the principles of three laser SLAM algorithms: Gmapping,Hector and Cartographer,are analyzed,and the comparison experiments of the three laser SLAM algorithms are designed.The mathematical models of the SLAM process are derived,including the SLAM process based on the filtering method and the SLAM process based on the optimization method.Detailed analysis and derivation of Gmapping,Hector and Cartographer algorithms are performed.The simulation environment is built based on ROS and Gazebo platform,and the three laser SLAM algorithms are used to build a map for the simulation environment.Through the analysis of simulation experiments,the Cartographer algorithm has a smaller cumulative error and better map building effect.Thirdly,the dense map building method of ORBSLAM2 based on ELAS is designed,and the laser and vision fusion SLAM method based on Bayesian method is proposed,and the fusion map building in simulation experiment is designed.Three main threads of ORBSLAM2 are analyzed: camera motion tracking,local map building and loopback detection.To solve the problem that the sparse map built by ORBSLAM2 is difficult to be used for subsequent navigation,ELAS algorithm is introduced to construct dense map,which is further transformed into octree map and 2D grid map.In order to realize the complementary advantages of lidar and camera,the fusion map building method of laser and vision based on Bayesian method is proposed.And fusion map building simulation experiment is conducted on ROS platform.Finally,a hybrid A* and TEB path planning algorithm is designed by combining global path planning algorithm and local path planning algorithm.The Dijkstra and A*global path planning algorithms are analyzed and compared by Matlab simulation experiments,and the A* algorithm with shorter running time and higher computational efficiency is selected.The DWA and traditional TEB local path planning algorithms are analyzed,and the topological map-based TEB parallel algorithm is introduced and simulated for the problem that the traditional TEB algorithm easily falls into local extremes.The hybrid A* and TEB path planning method is designed and simulation experiments are conducted in ROS to verify the effectiveness of the hybrid path planning algorithm in static environments and environments containing unknown obstacles.