Research On Visual-inertial Positioning Method For Driverless Vehicle

The positioning and navigation technology of unmanned vehicles is mainly used to determine the location and speed of vehicles,which is an important prerequisite for ensuring that unmanned vehicles complete their driving tasks.With the expansion of driverless vehicle application scenarios,self-localization methods that do not rely on GPS and high-precision maps have attracted more and more scholars’ attention.The visual odometry system,which is located through the image information collected by the camera,has been widely used.However,in actual application scenarios,due to the complexity and change of the actual road environment,extreme lighting conditions,lack of image texture structure,and short-term rapid movement of the camera,etc.,are not conducive to self-positioning of the visual odometer.The positioning error is large.The addition of inertial information measurement unit can obtain the angular velocity and linear acceleration data of the object,provide better pose estimation under the fast movement of the camera,and can complement the visual information.Therefore,the visual-inertial fusion positioning navigation system has broad application prospects in driverless vehicles.This article takes the vision-inertial combination positioning method of driverless vehicles as the research object,discusses the construction of the visioninertial positioning system and verifies the effectiveness of the system.In order to facilitate the research,this article divides the system into three parts: visual odometer system,visual-inertial odometer system,loopc detection and global optimization system.For the visual odometer part,mathematical modeling and analysis of pure visual positioning and mapping problems are carried out,the projection model of the camera is elaborated,the selection of visual feature points,the correlation and the solution of camera motion and pose are analyzed,and finally completed the establishment of the visual odometer part.After establishing the visual odometer,in order to integrate the inertial information,the mathematical model and theoretical analysis of the positioning and mapping of the visual-inertial fusion are studied.In order to ensure the real-time performance of the system,the sliding window method is used to limit the complexity,and the non-linear optimization algorithm is used to optimize the visual and inertial measurement information to obtain the visual-inertial odometer system.Through the similarity of the images,the system’s loop detection and relocation functions are realized.The visual-inertial odometer system is used to obtain the pose constraints between the camera pose constraints and the system loopback frames.The global pose is optimized to reduce the cumulative error.Finally,the combined vision-inertial positioning system is verified,and the public data set is tested to verify the performance of the system.At the same time,the simulation experiment results are discussed and analyzed,and some suggestions for improving the system pose estimation accuracy are put forward.