Research On Algorithm Of Stereo Visual Inertial Odometry

In this paper,we focus on achieving the stereo visual inertial odometry algorithm,the main work and conclusions of this paper are as follows:(1)Summarize the mainstream VIO algorithm,such as MSCKF,MSCKF2.0,VINSmono and so on,classifying them via the state estimation algorithm they use and the way they combine the measurements of camera and imu measurements and so on.We shortly introduce the important elements which will deeply influence the performance of VIO system,such as marginalization strategy used in sliding window optimization,methods used to restrict observability,which will leads to inconsistency of the state estimation system.(2)Introduce the basic knowledge about visual observation and imu propagation,such as JPL quaternion,rotation dynamic,imu measurement model,optical flow tracking,ORB key point detection and description and so on.Next,we explain some detail about the state estimation problem,talk about the filter based and optimization based estimation framework.(3)Bring up the hybrid stereo visual odometry algorithm combing optical flow and keypoint detection,which is the foundation of the further VIO algorithm,(4)Give out the stereo VIO algorithm,which is based on the MSCKF algorithm.We discuss all the detail of the algorithm,including initializing,imu state transmission,visual measurements updating and so on.We testify dual algorithms with public datasets.Results show that the hybrid VO algorithm has lower localization accuracy but higher calculation efficiency compared to current open source ORB-SLAM2 algorithm.Meanwhile,the stereo VIO can be improved.