Visual-inertial Odometry Software Design Based On IMU And Monocular Visual Fusion Algorithm

SLAM technology is widely used in unmanned aerial vehicles,unmanned vehicles,AR,VR and other industries.The visual inertia odometer with camera and IMU as the sensor is the odometer front end in the SLAM system.It is mainly responsible for the positioning function,that is,measuring the trajectory of the carrier in three-dimensional space.The main content of this thesis is to design a visual inertial odometer software with high precision,good robustness and good real-time performance,which can realize autonomous positioning.In the paper,the visual inertial odometer software framework for monocular camera and IMU data is given.The whole software is divided into three parts: image and IMU data preprocessing,visual inertia odometer initialization,and visual inertia odometer sliding window optimization.Image and IMU preprocessing refers to processing the original image and IMU data.In this paper,the image is tracked using the optical flow method,and the observation map model of the feature points and image frames is constructed.For the IMU data,the IMU pre-integration formula between two frames of image frames is derived,and the preintegration term is obtained by pre-integrating the data of each adjacent two image frames by numerical integration method.The preprocessed image and IMU data will be used for visual inertia odometer initialization and optimization calculations.The initialization of the visual inertia odometer is a very important part of the visual inertial odometer system.This paper analyzes the importance of initialization in visual inertia odometer and gives a joint calibration method based on the fusion of vision and IMU measurement.Firstly,the initial motion structure is restored according to the pure visual data,and the least squares term is constructed according to the rotation difference between the rotation measured by the IMU and the visual rotation,and the gyro drift is initialized by the least square method.Then the least squares term is created according to the IMU measurement model,the gravity vector,the visual scale factor,the initial velocity and the initial pose are initialized,and the global reference coordinate system is determined according to the direction of gravity.After initialization,a small trajectory of the carrier at the initial moment can be obtained.The optimization of the sliding window is the last part of the visual inertia odometer,which is used to solve the pose of the carrier at the latest moment and optimize the last segment of the trajectory in real time to reduce the cumulative error.A sliding window containing a fixed number of image frames is arranged at the end of the trajectory,and the visual error term,the IMU observation error term and the a priori error term in the sliding window are extracted,and the state of each image frame in the sliding window is performed in real time through the Ceres nonlinear optimization library.Optimize the calculation.Finally,different data sets were used to test the performance of the visual inertial odometer software in this paper.At the same time,the sensor equipment was used to collect data to test the robustness and real-time performance of the odometer software under different motion environments.