Visual-inertial Odometry Based On IMU Pre-integration

The Visual Odometry(VO)has become an important part of autonomous navigation and localization research for mobile robots.Visual information and inertial information complement each other.Visual information can effectively correct the drift of inertial information,while inertial information can be used to locate in the case of rapid movement or missing features.Sensor settings that incorporate cameras and inertial components are ideally suited for navigation,but they present enormous research challenges simultaneously.Therefore,this paper focuses on the application of inertial sensors in the visual odometry,designs a visual inertial odometry system,and validates the effectiveness of the system through public datasets and actual indoor and outdoor environmental experiments.This article centers on the study of the visual odometry system.Firstly,the article expounded the basic camera projection model,and the transformation relationships among various coordinate systems are analyzed as well.Secondly,the pose estimation method is introduced,which is based on the feature method and the optical flow method.For some main algorithms,their efficiencies and properties are analyzed and compared.At the end,three visual odometry systems are designed and constructed,corresponding to two-two frames,local maps and optical flow method respectively.And their real-time properties and positional accuracies are analyzed through experiments on the open dataset.In order to optimize the repeated integration problem caused by the change of initial conditions,the pre-integration algorithm is used to process inertial data.Considering the effects of noises and drifts,the IMU is modeled,the kinematic formula of the object is constructed,and the pre-integral equation is derived.The noise and drift are processed by error propagation and drift update processes,and the IMU data is pre-integrated.To make full use of IMU information between two frames of images,the article designed a visual inertial odometry,in which the inertial pre-integration algorithm was used to integrate inertial measurements into the constraints of relative motion.An optimization algorithm is used to optimize the visual and inertial pre-integration information and update the pose,so as to realize the pose estimation of the camera motion.Through the public dataset experiment,the effectiveness of the system was verified.In order to verify the design of the visual inertial odometry system based on IMU preintegration,experiments were conducted both in the indoor and outdoor real environments.Experimental hardware platforms include TurtleBot mobile robots and binocular cameras with IMU,While the experimental environment includes the lab environment and the campus environment.The camera loading methods include handheld,mobile robot platform and vehicle.Through the indoor and outdoor environment tests,the effectiveness and reliability of the odometry system are verified.