Research On Visual Environment Mapping And Localization Algorithm Of Unmanned Vehicle

With the development of society and the advancement of technology,unmanned vehicles have gradually entered people’s life.Simultaneous Localization and Mapping(SLAM)is one of the leading research methods to solve unmanned vehicle localization in the research of unmanned vehicle systems.The camera has a wide range of applications in the study of SLAM technology due to its low cost and rich feature information.However,the camera is easily affected by weather,environment,motion status.This thesis mainly studies the visual environment mapping and localization algorithm based on multi-sensor fusion,and the accuracy and robustness of unmanned vehicle localization are improved.The main contents of this dissertation are as follows:(1)The theoretical basis and camera model of SLAM are introduced.Based on the framework of visual SLAM,this thesis describes the methods and strategies from four aspects: visual front-end odometry,back-end optimization,loop correction,and sparse map.(2)A wheel speed pulse visual-inertial tightly coupled SLAM is proposed.First,visual SLAM is easy to lose in scenes with camera shake and texture features that are not rich.There is good complementarity between the camera and the inertial measurement unit(IMU),this thesis designs the visual-inertial tightly coupled SLAM algorithm.Then,the unmanned vehicle cannot provide sufficient acceleration excitation for the IMU when it moves at a constant speed,leading to scale uncertainty when the monocular visual-inertial SLAM is initialized,the wheel speed pulse visual-inertial tightly coupled SLAM is proposed in this dissertation.The uncertainty of system scale initialization is effectively improved,but also the localization accuracy and robustness of the system are improved.(3)An algorithm for map storage and reuse is proposed.Aiming at the situation that unmanned vehicles repeatedly run a complete SLAM in the mapped environment will cause a waste of running memory,and the map storage and reuse algorithm is proposed in this dissertation.If the map is saved after SLAM is run for the first time in an unmodeled environment,the map can be directly loaded and localized in the next startup.The cost of running memory for unmanned vehicle localization is greatly reduced,and the accuracy of real-time localization is improved by adopting this method.Based on real-time high-precision localization,an offline point cloud reconstruction algorithm is designed to reconstruct the environment,the visualization of the map is improved.(4)An unmanned vehicle trajectory tracking control algorithm based on visual localization is designed.First,because the wheel speed pulse visual-inertial tightly coupled SLAM algorithm can provide real-time high-precision localization,SLAM localization information in the visual coordinate system is transformed into odometry in navigation coordinates for unmanned vehicle navigation.Then,combined with the four-wheel differential unmanned vehicle platform,a visual-based trajectory tracking control algorithm is designed to achieve the unmanned vehicle tracks the desired trajectory smoothly.(5)The unmanned vehicle platform is built.The proposed algorithm is written and implemented on the unmanned vehicle,and then the proposed algorithm is analyzed,evaluated,and compared by experiments.This thesis conducts experiments in multiple scenarios for the proposed wheel speed pulse visual-inertial tightly coupled SLAM algorithm.By comparing with wheel odometry,monocular visual SLAM,and ORBSLAM3 monocular visual-inertial SLAM proposed in 2020,the effectiveness,localization accuracy,and robustness of the algorithm in this thesis have been verified.The proposed map storage and reuse algorithm has also been tested in multiple scenarios.Compared with the VINS-Mono algorithm,which also has map storage and reuse functions from multiple dimensions such as the size of the map storage,storage time,and loading time,the advantages of our algorithm are reflected.The vision-based unmanned vehicle trajectory tracking control algorithm and point cloud reconstruction algorithm have been programmed on the unmanned vehicle platform and have been experimentally verified in the actual environment.