Mobile Robot Localization Based On Vision And Inertial Sensor Fusion

In recent years,mobile robots have been successively used in cleaning,intelligent warehousing,and logistics.Visual-inertial odometry(VIO)is widely used to solve the localization problem of mobile robots in unknown environments because of its high accuracy and no need to interact with the environment.However,the robot operating environment is complex and changeable,and robots are usually equipped with low-precision IMUs and cameras to reduce manufacturing costs.The low-precision IMU has disadvantages such as large noise,unstable bias,and time-varying delays from camera measurements.These factors lead to the problem that the positioning accuracy decreases or even the localization fails when the VIO algorithm runs on the robot.Therefore,it is of great significance for mobile robots to develop high-precision and robust positioning algorithms for low-precision IMUs and cameras to improve the positioning accuracy of mobile robots in complex and unknown environments.To this end,this paper takes the low-precision IMU and camera as the research object,takes the filtering theory as the basic framework,and aims to improve the robustness and accuracy of the positioning algorithm.This paper conducts research from four aspects: the fisheye camera model,online estimation of IMU bias,time synchronization,and observability consistency.Additionally,an actual mobile robot platform is built to verify the effectiveness of the algorithm.The main research contents are as follows:(1)Aiming at the problem of the feature degradation caused by the robot moving too fast or the illumination intensity is uneven,which further leads to the failure of localization,a VIO algorithm based on the enhanced unified camera model(EUCM)is proposed.First,to solve the problem of image distortion of the fisheye camera,the assumption of uniform motion of features is used to eliminate the outlier to improve the matching accuracy.Then,the EUCM camera model is used to model the fisheye camera,which reduces the computational complexity of the projection model and increases the computational speed.Then,considering the insufficient sampling of the camera parameters for offline calibration,the fisheye camera parameters are added to the system state variables,and the parameters are calibrated online to improve the accuracy of the camera model.Finally,the first estimation Jacobian(FEJ)is used to fix the linearization points of the state transition matrix and the observation matrix to improve the observability consistency and the positioning accuracy of the VIO algorithm.(2)To solve the problem of unstable bias of a low-cost IMU,IMU parameter calibration is studied,and an online estimation algorithm of IMU bias based on a two-stage Kalman filter is proposed.The acceleration measurements are used to estimate the gyroscope bias online,which improves the stability of the gyroscope bias and improves the robustness of the VIO.When marginalizing the camera state,the camera pose is marginalized based on the Schur complement to avoid losing the relevant information between the marginalized variable and other state variables.The experimental results show that the DS-KF algorithm can accurately estimate the gyroscope bias and improve the robustness of the algorithm when the feature observation is inaccurate.Compared with the classical MSCKF-VIO algorithm,the proposed algorithm can effectively improve the trajectory accuracy.(3)The low-cost IMU and the camera do not support a time synchronization trigger,and there is a time delay between the IMU and camera measurements.To this end,a time delay calibration algorithm based on the photometric error of features is proposed.First,the time delay between the camera and the IMU is added to the system state variables,and then the photometric error of features is used to update the time delay.The experimental results show that the photometric error of the features can accurately estimate the time delay between the images and the IMU measurements,and compared with the classic VIO algorithm,the VIO algorithm considering the time delay has a significant improvement in positioning accuracy.(4)In the classic VIO algorithm,the FEJ is used to improve the observable consistency of the VIO by fixing the linearization point,which causes the linearization point to be far away from the real point,resulting in model error.To this end,the observable consistency of the fusion model of the VIO algorithm is studied,and an invariant state Kalman filter with Lie algebra is proposed to improve the observable consistency and positioning accuracy of the algorithm.The algorithm uses Lie algebra to represent the state variables of the system,the state transition matrix and observation matrix of the EKF are derived based on group theory,and the observability of the proposed algorithm has been proven theoretically.The experimental results show that the invariant state Kalman filter model can improve the observability consistency of the VIO algorithm and improve the positioning accuracy of the algorithm,and the accuracy of the proposed algorithm is better than that of the FEJ-based VIO algorithm under the condition of large noise.Compared with the VIO algorithm without consistency constraints,the accuracy of the algorithm proposed in this paper is improved by 44.74% on average.