Research On Visual And Inertial SLAM

Simultaneous Location and Mapping(Simultaneous Location and Mapping,SLAM)refers to using built-in sensors to estimate the pose of the mobile robot,while building and updating the map of the unknown environment.It is the premise and foundation of mobile robot control and planning.The pure visual SLAM system is prone to be affected by the environment and has the problem of scale ambiguity,while the inertial navigation unit(Inertial Measurement Unit,IMU)has the characteristics of not being affected by the environment.It’s obvious that IMU has a complementary relationship with camera.Based on this,we proposes a SLAM system combining monocular vision and IMU information on the basis of the research on vision and inertial navigation data processing methods.The main research contents of this article are as follows:1)Research and analysis on the processing methods of visual data and IMU data.First,the projection equation and distortion model under the pinhole camera model are introduced,and the three representation methods of rotation and their mutual conversion equations are derived;then the pure visual pose estimation method(Sf M)is derived based on the projection equation and the geometric invariant of stereo vision,including epipolar geometry,triangulation,Pn P and bundle adjustment,etc.,besides these the limitations and applicable scenarios of various pose estimation methods are analyzed;finally we deduced the IMU pre-integrated observations and its uncertainty on the manifold from the dynamic equations.2)We proposed a robust high-precision real-time visual inertial navigation SLAM system.The system first initializes the system parameters and initial state through loosely coupled visual and inertial observations.The initialization method does not need to assume that the initial state of the system is stationary or the initial direction is aligned with gravity,which enhanced the robustness of the system.We also estimate the acceleration bias,for the fast convergence of low cost IMU device.The initializing method provides accurate initial values for the subsequent nonlinear optimization model;then we simultaneously perform IMU pre-integration and visual feature tracking,and tightly couple IMU pre-integration observations and visual feature point observations in order to estimate the real-time state of the system;for the problem of system drift caused by accumulated errors,we use the loop detection algorithmreduce the impact of drift and increase the stability and robustness of the system.3)The proposed system was evaluated by specific methods.We use the open source dataset Eu Ro C to evaluate the system from the perspectives of accuracy,computational efficiency,and robustness.The evaluation includes three parts: system initialization convergence,parameter evaluation,and system accuracy evaluation.The results show that the proposed system can still be initialized quickly in fast motion and weak texture scenes.The gyro bias is only related to the attitude so the convergence time is shorter than the acceleration bias;There is no direct dependency between the system accuracy and the number of feature points,we can improve the computational efficiency of the system without losing the accuracy of the system by reducing the number of feature points;The accuracy of the system is comparable to the state-of-art visual inertial navigation SLAM framework VINS,and it performs better than VINS on some sequences of the dataset;There is a significant improvement of positioning accuracy of loop version over non-loop version,which indicates that loop detection algorithm can effectively correct the system drift caused by the cumulative error.