Research On The Application Of SLAM Algorithm For Intelligent AGV Based On The Combination Of Laser And Vision

In recent years,with the rise of smart manufacturing and the expansion of logistics warehousing systems,there has also been more demand and higher technical requirements for AGVs based on unscheduled route navigation.Simultaneous positioning and mapping(SLAM)technology is currently the most important solution to the problem of unscheduled route navigation.Research on SLAM algorithm optimization and application has great research value and practical engineering significance.According to the type of sensor used for scene scanning perception,SLAM technology is roughly divided into two common methods: laser and vision.This paper analyzes the two SLAM technologies based on a single sensor.According to their respective performance differences in different scenarios,a SLAM solution combining laser and vision with high cost performance,wide application range and good mapping effect is proposed.And improve the Bi-RRT algorithm to plan a simple and reliable global path for AGV navigation.First,the overall system architecture of AGV that satisfies the unscheduled path navigation is studied,the relevant mathematical model is established,and a joint calibration method that can spatio-temporally synchronize different sensor information is found.Then,the nature of the SLAM problem is described using a mathematical probability model,the SLAM problem is divided into different modules,the visual odometer based on ORB feature matching is analyzed,and the accelerated correlation scanning matching method is used as the laser front end.It also compares different map models such as vision-based sparse feature maps and laser-based raster maps,selects two-dimensional raster maps that are good for navigation,and gives the specific calculation methods.A SLAM scheme combining laser and vision is designed and implemented.Aiming at the shortcoming that single-line lidar can only obtain one plane environment information,a joint mapping method of laser and vision is proposed: the high-precision laser point cloud is used as the reference value to calibrate the depth point cloud obtained by the RGB-D camera The depth point cloud in the set height space is flattened into two dimensions and fused with the laser point cloud.Then,the map is constructed based on the fused two-dimensional point cloud,which can contain environmental information of different vertical heights.Aiming at the shortcomings of the laser two-dimensional point cloud information feature and the difficulty of eliminating the cumulative error,the closed loop was detected by selecting the key frames of the visual image and constructing a visual bag model.Adding the constraints provided by the closed-loop information to the SLAM back-end threads for graph optimization can solve the problem of misalignment of two-dimensional laser SLAM construction.An improved Bi-RRT algorithm is proposed to plan the global path.The A* algorithm evaluation idea is introduced to the RRT algorithm.The continuous biased target strategy,direct connection judgment mechanism and path post-processing are added to improve the efficiency and effectiveness of path planning.Finally,the physical environment of simulation experiment is built,and different algorithms are tested in the same scene,which verifies the feasibility and superiority of the algorithm,and the map is more accurate and perfect.The improved path planning algorithm is more efficient and the generated executable path is more concise.