Research On Nonlinear Filtering Method In Vision Positioning Of Unmanned Vehicle

Positioning is the key for a vehicle to accurately determine its own position in unmanned driving.Visual positioning is a method of obtaining vehicle position by extracting and matching feature points from the collected images,and performing motion estimation and filtering optimization.Compared with traditional positioning,it has the characteristics of low cost and small size.However,in the process of long-distance and long-term driving of unmanned vehicles,due to the influence of vehicle motion and camera internal parameters,the system model is nonlinear,and it is difficult to obtain accurate vehicle position information.This dissertation focuses on the positioning estimation problem of this nonlinear system.The specific research content is as follows:This article first introduces the basic theory of visual odometer,establishes a mathematical model related to visual odometer,and then designs a Harris-SIFT feature point extraction algorithm for the problem of long extraction time and low accuracy of feature points in visual odometer.After extracting Harris corner points,the algorithm extracts SIFT feature descriptors in the neighborhood of Harris corner points,which greatly reduces the calculation time of the algorithm while ensuring accuracy.Finally,simulation experiments are performed on various occasions to lay a foundation for the research of filtering algorithms in the subsequent chapters.In the traditional visual positioning system,due to the large truncation error of the extended Kalman filter,a stereo vision positioning algorithm based on the strong-tracking cubature Kalman filter is proposed.The algorithm uses the Spherical-Radial Cubature criterion to sample the system state,and then approximates the target state by a non-linear function weighted,and at the same time,multiple sub-optimal fading factors are introduced in the iterative process,and the gain matrix is adjusted in real time according to the state changes to improve the robustness of the system in the case of sudden changes in the state.Finally,the KITTI data set is used to verify the algorithm by simulation experiments,and the results show that the algorithm can effectively reduce the errors in visual positioning.Aiming at the inaccurate position estimation problem caused by the lack of particle diversity in the particle filter,a stereo vision positioning algorithm based on particle flow filter is proposed.This algorithm solves the differential equation of the homotopy logarithm instead of the particle filter resampling.The particle flow smoothly moves the state particles to the target area,and then obtains accurate position information from the target area.Simulation experiments show that the particle flow filter algorithm can effectively improve the accuracy of visual positioning.In order to further verify the effectiveness of the algorithm in this paper,this paper builds a hardware experiment platform,collects the image data of the campus through a stereo camera,builds a data set,and verifies the two algorithms mentioned in this paper in real vehicles.The results show that the visual positioning algorithm based on the strong-tracking cubature Kalman filter and the particle flow filter can effectively reduce the positioning error.