Research On Self-Adaptive Integrated Navigation Technology Based On Vision/Inertial Fusion

In recent years,modern economic life puts forward higher requirements for the autonomy and anti-interference of navigation.The integrated navigation system based on vision / inertia can maintain high positioning accuracy and continuity in complex environment because it is not affected by wireless communication,so it has become one of the key research directions of integrated navigation technology.Taking the system modeling,feature matching and data fusion involved in the indoor and outdoor continuous autonomous navigation technology of mobile carrier as the coincidence point,this paper deeply carries out the research on the key technologies of vision / inertial adaptive fusion navigation.Based on the analysis of the principle of inertial navigation solution,the error equation of inertial navigation system is established.From the perspective of pinhole model,camera distortion,camera correction and epipolar geometry modeling,the visual navigation error model is established.Complete the effective update of the carrier attitude solution,and provide a theoretical guarantee for the research of vision / inertial integrated navigation method.Comparing the advantages of SIFT,surf and orb feature detection and matching algorithms,an image feature tracking and matching scheme based on orb algorithm and KLT optical flow method is proposed;A visual feature matching method based on random sample consensus(RANSAC)is proposed to reduce the false matching and invalid matching of feature points and solve the problem of feature matching accuracy in visual navigation;The effective estimation of the position of feature points is realized by triangulation method.After obtaining the spatial threedimensional coordinates of feature points,the real motion of visual camera is calculated by trifocal tensor.In order to further improve the stability of integrated navigation system,the principle of covariance adaptive algorithm is revealed by analyzing the noise variability of vision / inertial integrated navigation system in complex scenarios;For the case of serious mismatch of innovation variance,the current measurement information is eliminated and the measurement update is skipped;For the case of light innovation variance mismatch,a new fusion algorithm combining comprehensive innovation and fading adaptation is proposed based on Kalman filter.By introducing two tuning parameters into the system,one performs the adaptive iteration of process noise and the other performs the adaptive iteration of measurement noise,so as to improve the covariance matching speed and suppress the filter divergence of visual / inertial integrated navigation system in complex scenarios.Focusing on the problem of accuracy divergence of integrated navigation system caused by inertial navigation system error drift,a data fusion estimation method based on adaptive multi state constraint Kalman filter(amsckf)is proposed based on state constraints and adaptive ideas.Monocular and binocular measurement models are established respectively,and the state of inertial measurement unit(IMU)is used to expand the dimension of the filtered state.After building the model,amsckf filtering is used to complete the parameter estimation of vision / inertial integrated navigation system,so as to improve the robustness of vision / Inertial Integrated Navigation system.In order to verify the effectiveness of the combined adaptive method proposed in this paper,the visual / inertial adaptive fusion algorithm proposed in this paper is experimentally verified with Kitti data set and euroc data set.Firstly,RANSAC is used to optimize the image feature matching algorithm,and then the photo processing process is improved.The preprocessed visual data and inertial navigation data are fused and estimated by adaptive multi state constraint algorithm,and the experimental results show that the optimization algorithm has better robustness and higher estimation accuracy;Secondly,the estimation results of the optimized monocular combined amsckf algorithm and other typical algorithms are compared and analyzed.It is proved that amsckf algorithm can effectively suppress the error divergence of inertial navigation system,and its estimation accuracy is higher than that of classical Kalman filter algorithm;Finally,in order to verify the performance of binocular integrated amsckf algorithm,ape and RPE are used to compare the positioning results of amsckf algorithm and other classical algorithms in various environments.The experimental results verify that the binocular vision / inertial integrated navigation fusion algorithm based on amsckf has high positioning accuracy.There are 74 figures,24 tables and 116 references in this paper.