Design And Implementation Of Visual SLAM Positioning And Mapping Method In AGV Navigation System

At present,it is a critical period for China to transform from a manufacturing country to a manufacturing power.It is necessary to continuously improve the intelligence level of the manufacturing industry.Automated guided vehicle is an important tool for intelligent manufacturing,which can greatly improve production and living efficiency and reduce industrial costs.Therefore,it has very important research value.In the industrial production process of many products,frequent material transportation is required.Compared with manual transportation,using AGV can improve the efficiency of material transportation and reduce costs.After years of development,automated guided vehicles have also been widely used in various industries,and their navigation systems have also evolved from the most traditional fixed-orbit navigation to electromagnetic navigation and laser navigation.However,considering factors such as flexibility and cost,electromagnetic navigation and Laser navigation is still not the best way to navigate.In recent years,with the emergence of deep learning,the research on computer vision technology has become more popular in academia,and at the same time,it has driven the rapid development of visual SLAM technology,and many classic visual SLAM solutions have also appeared.In this paper,based on the classic SLAM technical scheme,the design and implementation of the visual SLAM positioning and mapping method for AGV navigation system are carried out,and a positioning scheme based on the fusion of feature point method and optical flow method is proposed.This scheme can effectively avoid the mismatch of feature points,and lay a good data foundation for subsequent pose estimation,back-end optimization and dense mapping.In view of the demand of AGV navigation for dense maps,this paper proposes a dense mapping scheme based on epipolar matching.First,the binocular camera is used to perform epipolar matching to construct an initial local map,and then multiple frames before and after are used for depth filtering,so that the map the position of the point converges to an acceptable state value.Finally,point cloud filtering is used to downsample the point cloud and remove outliers,and the fusion of the local dense map to the global map is realized through the relative pose transformation relationship.This paper implements it by a modular method.Through experiments on the KITTI dataset and actual scenes,the feasibility and effectiveness of the visual SLAM positioning and mapping method proposed in this paper for AGV navigation system are proved.