Research On Visual SLAM Algorithm Based On Multi-sensor Fusion

When the robot enters an unknown environment without any prior information,the robot needs to calculate its own position in real time based on the information feed back by the sensors,and draw a map of the environment.The process by which a robot perceives its own position and draws a map is called simultaneous localization and mapping(SLAM).In the application of mobile robots,SLAM is a core technology and a hotspot in current academia.SLAM can be implemented with a variety of sensors.Using only the camera as a sensor is called visual SLAM,using only lidar is called laser SLAM,and using multiple different types of sensors is called multi-sensor fusion SLAM.The multi-sensor fusion SLAM algorithm has greatly improved both in accuracy and robustness,and compared with the single sensor,the calculation amount has not increased significantly.It is the main research direction of the current mobile robot SLAM technology.Visual inertial odometry(VIO)is the main part of the visual inertial multi-sensor fusion SLAM algorithm,based on nonlinear optimization and the latest research results,this paper designs a multi-sensor fusion VIO algorithm that integrates inertial measurement unit(IMU),magnetometer,global positioning system(GPS),and camera.The research contents and innovation of this article include the following aspects:(1)This paper has completed the construction of an outdoor mobile robot experimental platform based on ROS(Robot Operating System)system.This article builds a software and hardware experimental platform based on the ROS system,and completes the compilation of ROS nodes such as IMU/magnetometer node,GPS node,mobile robot remote control node,upper and lower computer communication node.And this article gives the block diagram and critical code of the above program.(2)This paper completes the tightly coupled algorithm of VIO and magnetometer,shows the program flow of the entire fusion algorithm,and gives a program architecture diagram.This article starts with the classic kalman filter framework,and derives the fusion method of magnetometer and bundle adjustment(BA)optimization from the perspective of maximum likelihood estimation.First,the magnetometer was calibrated based on ellipsoid fitting,and then the attitude estimation of the magnetometer was carried out by using an improved double vector attitude determination algorithm.This paper calculates the coordinate transformation between sensors,and finally constructs the magnetometer residual through quaternion multiplication.The nonlinear algorithm is used to realize the tightly coupled fusion algorithm of magnetometer and VIO.This paper also analyzes the calculation of the fusion algorithm from the perspective of solving the incremental equation,and proves that the original sparse structure of the hessian matrix is not destroyed after adding the magnetometer.(3)This paper completes the loosely coupled fusion algorithm of VIO and GPS.The fusion methods based on extended kalman filter(EKF)and nonlinear optimization are briefly introduced.This paper uses a double ended queue to align the timestamps of the GPS message and the VIO output message,then calculates the transformation of the GPS coordinate system and the VIO coordinate system,and finally uses the pose graph model to complete the fusion algorithm.The pose graph fusion algorithm used in this paper adds GPS constraints to the original pose graph,and the inconsistencies in the timestamps of multiple sensors are compensated.The nonlinear optimization is used to realize the fusion algorithm of GPS and VIO.(4)This paper uses the outdoor robot mobile platform to test the effect of the proposed multi-sensor fusion algorithm.Experimental results prove the reliability and correctness of the proposed algorithm.This article also illustrates some of the shortcomings and improvements that were found during the completion of this thesis.