| With the rapid development of intelligent manufacturing and robotics technology,mobile robots are widely used in inspections,unmanned cleaning,logistics,agriculture and so on.Positioning technology is the basis and core of intelligent mobile robots,and it has received extensive attention from researchers.Due to the complexity of actual application scenarios,only a single sensor cannot meet the high-precision and robust positioning requirements of mobile robots.The camera has the advantages of low cost,small size and rich information of collected images,but it is easy to fail in tracking in the case of obstacle occlusion,image blur caused by fast motion,illumination change and so on.The Inertial Measurement Unit(IMU)has a fast response,and it is not affected by the imaging quality,it can provide scale constraints for monocular camera,but there is a problem of cumulative error.This dissertation uses the good complementarity of the two sensors,the sensor data are fused in a tight coupling way to achieve high-precision and strong robust positioning of mobile robots.Specific research contents are as follows:(1)The mathematical theory and derivation of related formulas used in the mobile robot positioning system based on visual inertial navigation are analyzed,and the coordinate system transformation in the positioning system is defined.Meanwhile,the camera model and the IMU mathematical model are discussed.(2)Research on the vision-based positioning technology of mobile robots.Aiming at the problem that orb algorithm is difficult to effectively extract feature points in scale transformation,illumination change,blur change and other scenes,an improved orb algorithm is proposed,and the feature matching accuracy is improved by eliminating false matching.The effectiveness of the improved algorithm is verified by comparative experiments on four datasets.The pose estimation algorithm of mobile robot based on multi view geometry is studied and deduced,and the problem of optimizing the pose estimation by using bundle adjustment is analyzed.Through the research of loop detection algorithm,the goal of reducing the cumulative error and obtaining the global consistent localization trajectory of mobile robot is achieved,and the reverse retrieval is used to speed up the algorithm.(3)On the basis of vision positioning,a positioning system of mobile robot based on vision inertial navigation is proposed.The vision and IMU are loosely coupled to initialize the positioning system,and the least squares problem is constructed by combining the vision residuals and IMU residuals,and Nonlinear optimization is used to iteratively solve the optimal pose estimation.In order to solve the problem of large amount of calculation in the optimization process,sliding window algorithm and key frame mechanism are used to limit the amount of computation,and marginalization is used to eliminate the old variables and corresponding observations.Finally,the positioning system outputs the pose of the mobile robot and sparse map.(4)The positioning system is tested under the public dataset,and the positioning system is compared with the other two mainstream positioning systems from qualitative and quantitative aspects.The experimental platform of mobile robot is built,the camera and IMU carried on the mobile robot are calibrated,and the three positioning systems are tested and compared in the real scene.The experimental results of dataset and real scene verify that the proposed positioning system of mobile robot with vision inertial navigation has high positioning accuracy and robustness.In summary,this dissertation studies the positioning technology of mobile robot based on visual inertial navigation and deeply discusses the front-end data association,nonlinear optimization and sensor data fusion of the positioning system,a strong and high-precision positioning method of mobile robot is proposed and the positioning system of mobile robot is established,which improves the intelligent level of mobile robot. |
PDF Full Text Request