Research On The Positioning Method Of Visual Inertial Odometery For Internet Of Vehicles

With the rapid development of artificial intelligence,cloud/edge computing and 5G network technology,the Internet of Vehicles(Io V)technology has attracted extensive attention from academia and industry.The Io V is an important infrastructure to support automatic driving,which can complete real-time road information exchange,sharing and vehicle positioning,and vehicle autonomous positioning is the key to improve intelligent decision-making and safe transportation capacity of automatic driving.Global Positioning System(GPS)is widely used in the traditional Io V.Due to the meter error and multi-path effect of GPS,especially in high-rise cities and closed tunnels,it cannot meet the key application requirements of automatic driving.In order to solve the aforementioned problems of vehicle autonomous positioning,based on the Simultaneous Localization and Mapping(SLAM),this paper aims to designing an autonomous positioning method combining visual positioning and Inertial Measurement Unit(IMU).The drift of IMU can be estimated and corrected effectively by using the no drift characteristic of vision equipment.IMU can obtain the body angular velocity and acceleration data to make up for the time when the camera data is invalid,combining the complementarity of the two.In this paper,we design a Visual-Inertial Odometery(VIO)location method,and propose a Semi-Direct Monocular Visual-Inertial Odometery location method based on point line features(SDMPL-VIO).The low-cost IMU and monocular camera obtain the surrounding environment information as the input of SDMPL-VIO model.Semi-direct method is used to deal with the features of points and lines in the image.IMU information is calculated by means of IMU pre integration.Finally,tight coupling framework is used to fuse the two data,making the odometery more stable and robust.It can effectively solve the problem of vehicle autonomous positioning in the situation that the vehicle cannot use GPS in the Io V.The main contents of this paper are as follows:1.The basic principle of Visual-Inertial Odometery is studied.Firstly,this paper systematically analyzes the basic theories of multi vision geometry,three-dimensional space rigid body motion,group and manifold,and deeply studies the basic principles of inertial navigation system,IMU kinematic model and nonlinear optimization theory,which lays the foundation for the following research of visual front-end feature extraction,feature description and rear end optimization.2.A semi-direct visual tracking model is proposed.In view of the fact that the point feature has a great dependence on the environment,and it is easy to lose the feature tracking in the scene such as texture missing.Based on the principle of multi vision geometry,this paper presents a semi-direct vision tracking model for the field of Io V.This model can extract point and line features from the images collected by camera at the same time.In addition,according to the characteristics of the scene,the semi-direct method is used to deal with different scenes.If there are few features in the scene at this time,the direct method is used for pose tracking,otherwise the feature point method is used.This model has a good feature tracking effect in the weak texture scene.It can effectively solve the problem that the point feature has a large dependence on the environment,and it is easy to lose the feature tracking in the scene of texture missing.The proposed semi-direct visual tracking model improves the efficiency of image processing and the accuracy of feature matching.Experimental results show that the model can provide more robust initial results for back-end optimization.3.The back-end optimization model of vision inertial odometery is proposed.Based on the combined positioning method of ORB-SLAM2 vision model and inertial measurement unit data fusion,a back-end optimization model of VIO is proposed.The optimization model of ORB-SLAM2 is improved.In this paper,IMU data is used to build the vehicle motion prediction model,and the output pose of ORB-SLAM2 is used as the vehicle pose update.The point line feature of the visual odometery and the output pose of the IMU are estimated by the Bundle Adjustment(BA)method.The model has high accuracy in dynamic environment and long-time operation.It can effectively solve the interference of the external environment on the camera and the problem that the error accumulation caused by the long-time operation of the IMU device results in the inaccuracy or failure of the vehicle self-positioning.The experimental results on the Eu Ro C and KITTI datasets show that the SDMPL-VIO method proposed in this paper is superior to many odometery methods of the same type in performance,and it verifies the reliability and stability of the semi-direct monocular visual-inertial odometery positioning method based on the point line feature.