Research On AGV Autonomous Positioning And Navigation System Based On Visual SLAM

In recent years,with the gradual popularization and wide application of mobile robots,the Simultaneous Localization and Mapping(SLAM),as its core technology,has become a research focus.It solves the problem that robots use their own sensors to construct maps and achieve self-localization in the absence of a priori environmental information.Based on the in-depth study of the SLAM technology theory,this article focuses on the shortcomings of the traditional navigation system of Automated Guided Vehicle(AGV),such as poor flexibility and strong environmental dependence,and combines with the mainstream algorithms in the current SLAM research to achieve the autonomous positioning and navigation of the AGV.The main research work and achievements of this paper are as follows:(1)The working principle of the traditional RGB-D SLAM system is analyzed,and an improved RGB-D SLAM system framework is implemented by combining the application scenarios of the AGV with the problems of real-time performance,poor robustness,and serious cumulative error.In order to solve the problem of poor real-time performance of the system,firstly,the time-consuming feature detection part of the original system is analyzed,and the ORB algorithm based on FAST key points and BRIEF descriptor is studied and improved,the SIFT、SURF,and improved ORB algorithms are compared with the experiment.In the overall performance of matching,ORB features with both performance and speed are selected instead of the original SIFT features.The system solves poses by Random Sample Consensus(RANSAC)and Iterative Closest Points(ICP)algorithms.For solving the ICP algorithm,there are many problems such as large amount of computation and easy to fall into the local optimal problem.The direction vector is introduced in the process of point cloud registration to get more accurate position and pose estimation quickly.Finally,the back-end optimization and loop closing detection module was added to achieve an improved RGB-D SLAM system framework based on the Kinect somatosensory camera.(2)To solve the problem of large memory overhead and long computation time for A* path finding algorithm in large scenes,this paper studies the implementation principle of A* algorithm,and proposes an improved A* algorithm combined with jump point search algorithm.The algorithm selects jump points and uses the points instead of the neighbors in the A* algorithm to perform a large number of unnecessary operations to extend until the final path is generated.In order to verify the feasibility of the improved A* algorithm,simulations were performed in different grid maps.The simulation results verify the effectiveness of the improved A* algorithm.(3)The validity of the improved RGB-D SLAM system and the improved A* algorithm are verified through standard data set testing and field experiments.Firstly,the improved SLAM system is tested with standard data sets.The motion trajectory error and data processing speed of each frame are compared with different data sets.The real-time and robustness of the system are analyzed.Then,an experiment platform based on Robot Operating System(ROS)was built,and the field experiments of the improved RGB-D SLAM system framework and the improved A* path finding algorithm were performed respectively,thus verifying the effectiveness of the proposed improved algorithm and framework.