Research On Monocular Visual Odometry And GPS Integrated Navigation System

Navigation and positioning technology is one of the key technologies of driverless vehicle,and it is the basis of vehicle's movement.The satellite navigation system and the visual odometry system have good complementary characteristics,using integrated navigation technology to combine them together can improve the overall performance of the navigation system.Based on the low cost sensor,this paper mainly studies the feature detection and corresponding algorithm of monocular visual odometry system and the information fusion algorithm of the GPS/VO integrated navigation system.Firstly,aiming at the poor real-time performance of visual odometer in the vehicle navigation system,this paper proposes a feature point detection and real-time tracking algorithm based on FAST and pyramid KLT.The algorithm uses FAST feature detector to detect the feature points of the image,and then uses the KLT algorithm to obtain the feature corresponding set between frames by the method of feature tracking.The algorithm is experimentally analyzed from two aspects: the algorithm run time and the number of the feature points involved.The results show that the proposed algorithm improves the efficiency of the algorithm under the premise of getting enough number of feature corresponding set for pose estimation.Secondly,the structure and error model of GPS/VO integrated system are studied.A loosely coupled navigation system model based on indirect method Kalman filter is established,the positioning and attitude error model of the GPS/VO integrated system are deduced.The model was verified by simulation using standard KF algorithm.Finally,aiming at the problem of redundancy calculation in the application of the standard UKF in the GPS/VO integrated system,an improved UKF algorithm is proposed.The algorithm optimizes the time updating process of standard UKF,and takes the same time process as KF to avoid the extra computation caused by UT transformation of standard UKF.The proposed algorithm is applied to the GPS/VO integrated system for experimental verification and compared with standard UKF.The results show that the improved UKF algorithm decreases the computation while the accuracy is basically equivalent to the standard UKF.On the basis of low cost sensors,The GPS/VO integrated system improves the positioning accuracy of monocular visual odometry system and reduces the computational complexity of the algorithm,which has reference value for improving the positioning and navigation performance of driverless vehicles.