Research On Vehicle Location Technology Based On Visual Odometer

GNSS positioning technology,inertial navigation positioning technology and visual navigation positioning technology are the mainstream of current positioning technology,and the single positioning technology has the problems of error accumulation and special environment limitation.Therefore,the single positioning technology has certain limitations.The emergence of multi-sensor fusion technology has solved the above problems by using the complementary and superposition of multiple sensors to ensure the positioning accuracy,so multi-sensor combined navigation technology has been favored by the academic community.The mainstream and widely used multi-sensor fusion technology in civil life is the combined GNSS/INS navigation and positioning technology.Among them,GNSS positioning technology provides high-precision navigation information,and inertial navigation and positioning technology provides short-term high-precision navigation information.combined navigation and positioning technology of GNSS/INS uses inertial navigation and positioning technology to maintain reliable positioning accuracy in the short term when GNSS signals are blocked,but with the extension of time,the errors of inertial navigation and positioning technology keep accumulating,which will lead to the whole The accuracy of the positioning system decreases.With the continuous progress of image processing technology,the change of object movement can be accurately judged by the change of image position taken by moving objects,which provides a new combined navigation mode to the original multi-sensor positioning,and the combined GNSS/INS/vision-based multi-sensor navigation and positioning technology comes into being.And low-cost vision sensor assistance can be more conducive to meet the positioning needs of applications in civil positioning environment.At present,this technology has been commonly used in the fields of autonomous driving,AR,UAV,robotics,etc.In this paper,we propose to introduce the aid of visual odometry to improve the positioning accuracy for the problems of combined GNSS/INS navigation.The main research of this paper is as follows:For the problems of high mis-matching rate and long matching time of image feature points while the vehicle is driving,this paper proposes the GMS+RANSAC algorithm based on Kdtree improvement.Firstly,the feature points of the image are extracted using the ORB algorithm,and then the Kd-tree is introduced to perform coarse matching on these feature points.Next,the coarse matched feature point pairs in the image are meshed using the GMS algorithm to find the correct matching points in each mesh and the feature points with high eigenvalue support around them,so as to perform the feature point selection.Finally,the RANSAC algorithm is used to further filter and reject the matching results to achieve accurate matching of the image feature points.This paper uses motion images from the KITTI dataset of vehicle driving experiments as samples,and proves through experiments that the algorithm designed in this paper has significantly improved the average matching time and matching accuracy,indicating the effectiveness of the improved algorithm proposed in this paper in vehicle motion feature recognition,which is useful for solving the feature point mis-matching problem.For the problem that vehicles are affected by tall buildings,tunnels and other environments leading to large observation and measurement errors,this paper proposes an adaptive extended Kalman filter algorithm based on Huber improvement.Firstly,it combines Huber’s method to reconstruct the measurement noise covariance matrix to reduce the influence of abnormal measurement values and improve the update accuracy of the measurement equation.Then the adaptive extended Kalman filter is used to better adapt to different data noise situations,improve the robustness of the system,and achieve optimal estimation.This paper verifies in Chapter 4 that the improved adaptive extended Kalman filter based on Huber’s method leads to improved combined navigation positioning accuracy and improved system robustness through real-vehicle testing.In this paper,we design the simulation experiments for constructing the combined GNSS/INS/vision navigation by introducing vision sensors,and verify that the combined GNSS/INS/vision navigation can improve the positioning accuracy through simulation experiments.The improved GMS+RANSAC algorithm based on Kd-tree is used to improve the feature point matching accuracy in the vision navigation system,and the data fusion of the three is achieved by reducing the measurement error by means of adaptive extended Kalman filtering based on Huber’s method.In this paper,simulation experiments are conducted to verify the effectiveness of the fusion algorithm by using the German KITTI dataset,and the effectiveness of the fusion algorithm is demonstrated by comparison experiments.