Research On Calibration And Monocular Measurement Technology Of Omnidirectional Camera

Omnidirectional cameras are widely used in computer vision,photogrammetry and optoelectronic imaging due to their advantages of large field of view,simple structure and low cost.However,it causes severe image distortion due to nonlinear imaging.The monocular measurement algorithm based on omnidirectional camera has limitations such as poor model versatility and low accuracy of calculation results.Aiming at these problems,this paper takes omnidirectional camera calibration and monocular measurement as the research background.The thesis has studied the calibration of the omnidirectional camera,the pose estimation of the point feature omnidirectional camera,the error analysis of the pose estimation of the spherical model,the extraction of the line feature and the pose estimation.Main work completed:Firstly,in the traditional calibration algorithm,the rectangular calibration plate isan irregular fan shape in the omnidirectional image,which causes the low detection accuracy of corner coordinates and low calibration accuracy.An iterative omnidirectional camera calibration algorithm is proposed.First,the traditional calibration algorithm is used to obtain the initial parameter values,which are used to construct a projection model.The distorted checkerboard is projected to the undistorted plane,Then the corner coordinates are extracted from the undistorted image and back-projected to obtain new corner coordinates.New parameter values are recalculated from the new corner coordinates.The iterative process improves the accuracy of camera calibration.Experiments show that the reprojection error of corners is reduced by 39% compared with the traditional method.Secondly,due to the nonlinear imaging of the scene and the large image distortion,the pose estimation of the omnidirectional camera requires complex mathematical calculations.Aiming at this problem,a point feature pose estimation method of omnidirectional camera is proposed.By establishing the projection relationship between the omnidirectional camera and the ordinary perspective camera,the equivalent linear imaging equation of the omnidirectional camera is constructed.It realizes the EPnP(Effective Pespective-n-Poin)and ASPnP(Accurate Scalable Perspective-n-Point)pose solving algorithms for omnidirectional cameras.Through a series of experiments,it is proved that the rotation matrix errors of the two algorithms are reduced by 29%和 21%,respectively,compared with the existing omnidirectional camera point feature pose estimation algorithms.Thirdly,the PnP algorithm based on the spherical model is deduced,and the error analysis model is constructed according to the error propagation theory.In order to improve the accuracy of pose estimation from the view of error source,the influence of principal point error,focal length error and mirror parameter error on the pose estimation accuracy is analyzed.In addition,the influence of the distribution of feature points and the number of feature points on the accuracy of pose estimation is analyzed from environmental factors.The simulation results are consistent with the theoretical calculation values of the pose estimation model,which verifies the correctness of the error analysis model.Finally,a line feature pose estimation method is proposed for the problem of omnidirectional camera localization in less texture scenes.The projection of the line feature of the omnidirectional camera is explored,and a line feature extraction method is proposed based on the projection of the line feature through the spatial distance constraint.Then,the constraint equation of the rotation matrix is given by aligning the two axes of the MF(Manhattan Frame)with two perpendicular lines in space.The pose estimation is performed by calculating the third straight line in the space,and then the rotation matrix is optimized by the robust parameter estimation method.The effectiveness and reliability of the algorithm are verified by simulating a structured environment simulation system and building an UGV(Unmanned Ground Vehicles)system in a typical looped corridor for experiments.This thesis starts from four parts: high-precision calibration of omnidirectional camera,point feature pose estimation algorithm,pose estimation error analysis and line feature pose estimation algorithm for the key technologies in the calibration of omnidirectional camera and monocular measurement technology.Relevant algorithms are proposed for each link,and experimental verifications are carried out from different aspects.It provides a new idea for the next step of visual navigation,target detection,robot positioning and other applications based on omnidirectional cameras.