Research On VIO System Based On Multi-sensor Data Fusion

Visual inertial odometry(VIO)is a key technology for autonomous navigation and positioning of mobile robots.In recent years,with the widespread application of unmanned aerial vehicles and unmanned driving technology,high-precision VIO systems have gradually become a hotspot in robot navigation and positioning research.The VIO system uses a vision camera and an inertial measurement unit(IMU)as sensing sensors,and realizes complementary sensing of robot position and posture by fusing sensor data,thereby improving the accuracy and robustness of robot navigation.This thesis first studies the algorithm of the binocular VIO system.The ORB algorithm is used to extract and match the feature points in the camera data,the IMU pre-integration algorithm is used to solve the robot pose estimation problem between adjacent image frames,and the nonlinear optimization algorithm is used to fuse the sensor data to reduce the pose estimation deviation.To ensure the real-time performance of the system,the algorithm processes key frames in the local map to reduce the amount of calculation.In dynamic navigation,a sliding window is used to marginalize key frames,and the residual items of prior information,IMU measurement information,and signpost observation information are used to construct an overall objective function to achieve the optimal nonlinear pose estimation.Aiming at the trajectory deviation existing in the binocular VIO system,a study on optimization and correction algorithm was launched.When the robot passes the same waypoint,the loop detection and optimization algorithm is used to eliminate the cumulative error of the pose estimation generated by the binocular VIO system for a long time.When the mobile robot is outdoors for a long time,long distance movement,and the scene of severe light changes is navigation and positioning,high-precision GNSS data and binocular VIO estimation data are fused to improve the robustness of pose estimation.In this thesis,a high-performance embedded platform is used to implement a real-time binocular VIO system,which completes the calibration of the camera and IMU,as well as the joint calibration of the two sensors.The binocular VIO system has been tested experimentally,the experiment shows that the improved binocular VIO system has higher accuracy and robustness.