Research On Visual Navigation Of Lunar Roving Vehicle

With the continuous development of Chinese lunar exploration project,the patrol and detection of the lunar surface has gradually become the focus,and the precise navigation and location function of the lunar rover is the premise guarantee for the effective detection of the lunar surface.Visual navigation is a high-precision and autonomous navigation method,but it is limited by the unknown environment on the surface of the lunar,so it still needs to be combined with other navigation methods.This paper studies the binocular visual navigation and filter fusion algorithm of visual-inertial integrated navigation for lunar rover,which is based on the extended Kalman filter and particle filter respectively to improve the navigation accuracy.Simulation results verify the effectiveness of the algorithm.Firstly,the mathematical model of the visual-inertial integrated navigation of the lunar rover is presented.The theoretical basis of visual navigation and the system model of inertial navigation are given,which lays a foundation for the design of visual odometer and filter algorithm.Then we introduce the overall flow and specific method of binocular vision odometer implementation and use a specific feature point detection and tracking method for soft and uneven surface of the lunar.In order to eliminate the false matching,we combine the random sample consensus algorithm with the least squares method to estimate the motion parameters.The simulation results verify the effectiveness of the algorithm.Next,the extended Kalman filter is adopted to fuse the visual-inertial navigation of the rover to improve the accuracy of positioning.We use the error state model with the inertial navigation acting as time update and visual navigation acting as measurement update,the simulation tests of the two data sets prove that the extended Kalman filter can achieve the desired effect.Finally,for the limitation of the extended Kalman filter that it can only solve the weak nonlinearity and Gaussian problem,while the systems of the actual projects are nonlinear nor non-Gaussian,the particle filter is introduced to take the fusion task.Firstly,the basic theory,existing problems and improvement methods of particle filter are introduced in detail.Then a particle filter is designed according to the application background of this paper.Finally,the feasibility of the algorithm is verified by simulation test and comparison of results.