Research On Stereo VIO Algorithm Based On Feature Points

At present,all kinds of mobile robots are widely used in medical treatment,rescue and relief,entertainment,service,agriculture and other fields.It is hoped that robots can ensure the accuracy of completed tasks while completing complex tasks.Therefore,the theoretical techniques related to mobile robots have received enthusiastic attention.The mobile robot can accurately estimate its position and attitude,which is the precondition and the most important part of the upper tasks such as path planning and automatic obstacle avoidance.Therefore,visual navigation and positioning has become one of the research hotspots in the field of robotics.There are two common solutions for visual navigation,visual odometry and visual SLAM(Simultaneous Location and Mapping).Academia and industry have invested a lot of energy in vision-based robotic positioning technology,which plays a vital role in the areas of automatic driving,augmented / virtual reality and so on.In this paper,aiming at the accurate positioning of mobile robot in the scene without label and sensor deployment,research on stereo vision odometry and visual-inertial odometry positioning algorithms,dedicated to improving positioning accuracy.Firstly,a stereo vision odometry system is designed,in which the basic feature point matching algorithm is prone to mismatching.This paper proposes a ring matching algorithm of feature points combined with bidirectional optical flow method.This algorithm can not only remove mismatching feature points,but also retain feature points that are not easily affected by the environment,and can make feature points evenly distributed on the image.The 3D-2D point-based pose is used to solve the pose and the Random Sample Consensus algorithm is used to remove outliers with large errors to obtain the initial pose estimation result.The g2o(General Graph Optimization)library is used to further optimize the pose estimation result.Through the data set test,compared with the basic feature point matching algorithm,the ring feature point matching algorithm combined with bidirectional optical flow method effectively reduces the mismatched feature points and improves the positioning accuracy.Aiming at the problem of inaccurate positioning caused by missing feature points in environments such as few feature points,obvious lighting changes,and fast moving speed,a filter-based visual-inertial odometry system is designed in this paper.In the stereo visual odometry algorithm verification,it has been proved that the feature point ring matching algorithm combined with the bidirectional optical flow method is superior in positioning accuracy,so this algorithm is directly applied to the visual-inertial odometry vision front end.Derived the IMU(Inertial Measurement Unit)kinematics equations,mathematically modeledthe IMU,and combined navigation and positioning by visual information correction of the IMU pose solution.A visual-inertial navigation range based on multi-state constraint kalman filtering tighten the combination system.Using the data set to evaluate the accuracy of the algorithm,it is verified that the visual-inertial navigation odometry has higher positioning accuracy than the visual odometry.In the end of this paper,the algorithm is verified by the experimental data.The image information and IMU information collected by the camera are transmitted to the visual-inertial odometer system through the ROS(Robot Operating System)communication mechanism.The experimental and simulation results based on the measured data show that the visual-inertial odometry system is better than the vision odometry system in positioning The accuracy is higher,and it is proved that the vision odometry and visual-inertial odometry are feasible and the algorithm can run stably.