| As an important part of intelligent travel,autonomous positioning and navigation have become more mature with the development of sensors and information fusion technology,and the development prospects of integrated positioning and navigation systems designed with the advantages of multiple sensors are becoming broader.At present,self-driving technology is no longer limited to specific simple environments such as parks.In some actual complex scenes such as high-rise buildings,garages,tunnels,etc.,GPS(Global Position System)signal coverage is weak or even lock-lose.However,it is inevitable that various forms of odometers are used for navigation and positioning,and cumulative errors will inevitably occur.At this time,it is difficult to simultaneously meet multiple practical application requirements by relying on a single integrated positioning and navigation method.According to the specific status of GPS,this paper designs a multi-sensor positioning and navigation system combining three methods based on binocular cameras,IMU(Inertial Measurement Unit)and GPS to adapt to a variety of application scenarios.Aiming at the driving environment where GPS may completely lose lock,such as tunnels and underground garages,a positioning and navigation system based on bundle adjustment of fusion of binocular vision and IMU data is designed.Firstly,the vision and IMU data are respectively preprocessed and the information of the two is synchronized in time.Based on the mainstream average parallax method,an optimization that effectively improves the quality of key frame screening is proposed on the premise of ensuring the overall real-time performance of the algorithm.The method is based on Kalman filtering and uses the short-term integration of IMU measurement values to compensate for the rotational movement between image frames,ensuring that the calculation of the average disparity between the target frame and the previous key frame can reflect the true value.After that,the visual-inertial state estimator is initialized,and the pose trajectories estimated by IMU data and vision is aligned.The measurement errors of vision and IMU data are calculated separately,the optimization objective function is constructed,the least squares problem is solved and finally the optimized pose is obtained.Aiming at the situation that the GPS signal is unstable but not completely in the unlocked state,a positioning and navigation system integrating GPS and visual-inertial information is designed.,GPS data is added as a new constraint to the bundle adjustment process.The limited GPS information is used to eliminate the cumulative error of the visual-inertial odometer and correct the historical trajectory.At the same time,the visual-inertial information is used to complement and optimize the missing GPS positioning data,so that the advantages of the two are complemented.If the GPS is in a normal state,the extended Kalman filtering method is used to fuse GPS and IMU information for navigation and positioning.The IMU kinematics equation and GPS measurement equation are established respectively,and the position,velocity,attitude and other information are solved,and the state variables are updated as the input of the IMU kinematics equation,and the optimized pose is obtained by iterative solution.Simulation experiments and real vehicle experiments verify the effectiveness of the integrated positioning and navigation system designed in this paper.Through the simulation experiment of the public data set,the algorithm of this paper is compared with the excellent mainstream algorithms under the same conditions,and the real-time performance of the visual-inertial fusion algorithm is effectively improved on the basis of good performance.A real vehicle platform is built for experimentation.The experiment shows that the vision/IMU/GPS integrated positioning and navigation system in this paper limits the positioning error to the decimeter level,which effectively eliminates the cumulative error of the visual-inertial state estimator and significantly improves the positioning accuracy. |
