Research On Fisheye Vision Imaging And Localization Based On Division Model

In machine vision,pinhole cameras cannot meet the application requirements of many scenes due to their small visible range.The fish-eye camera has a wide field of view through deformation and therefore can meet application requirements.How to propose a general fisheye camera imaging model and achieve high-precision localization technology is the difficulty and keypoint in this field.This thesis focuses on specific issues,fully combines the prior knowledge of the fish-eye camera imaging model,and proposes algorithms and frameworks based on division model for fish-eye correction,calibration,and localization,which can effectively guarantee the reversibility of the fish-eye imaging system and improve the accuracy of the localization algorithm.The main content is given as follows:(1)A fisheye correction method based on the division model is proposed,The method is mainly divided into the following steps: first,an epipolar geometric model based on radial distortion is established to solve the distortion center;then,the basic equations are used to calculate the parameters of the distortion model;finally,the fish-eye image is corrected according to the distortion model.The algorithm proposed in this theis uses the division model as a distortion model,and uses the epipolar geometry principle to construct an epipolar geometry model based on radial distortion,and then use the least square method to solve the distortion center.For the problem of solving the parameters of the univariate division model and the bivariate division model,this theis constructs the basic equations,combines the kernel voting mechanism,and obtains the parameters of the distortion model.The experimental results show that the algorithm has higher robustness and good correction effect.(2)A fisheye calibration method based on the division model is proposed,the method is as follows: first,the fisheye correction method based on the division model is used to estimate the parameters of the distortion model and the corner pixel coordinates of the corrected checkerboard are obtained;then,the internal and external parameters of the fish-eye camera are calculated by the zhang's calibration method;finally,the optimization of reprojection error is guaranteed by analyzing the reversibility of the imaging model,and then the optimal estimation of the internal and external parameters(including distortion parameters)of the fish-eye camera is obtained.By analyzing the monotonicity of the division model in the interval determined by the image resolution,the algorithm proposed in this thesis solves the inverse problem of the distortion model when the forward projection is calculated in the optimization process,and the efficiency of the optimization is accelerated by providing an analytic expression of the jacobian matrix.Therefore,the optimal values of the internal and external parameters(including distortion parameters)of the fish-eye camera can be obtained by minimizing the reprojection error.The experimental results show that the reprojection error converges faster,the algorithm is more robust,and the calibration parameters are more accurate.(3)A binocular fisheye localization method based on the division model is proposed,once the parameters of the fisheye camera are obtained,the localization can be performed according to the binocular localization system.The method is mainly divided into the following steps: first,the fisheye calibration method based on the division model is used to obtain the internal and external parameters(including the distortion parameters)of the fisheye camera;then,the fisheye correction method based on the division model is used to correct the pixel coordinates of the target in the two fisheye images;finally,the triangulation is used to locate the target according to the fisheye camera's external parameters of the two fisheye images.The algorithm of this thesis uses the division model to solve the problem of calculating the root of the polynomial in the forward projection of the conventional algorithm.The results of the experimental prove that the algorithm has low complexity and higher accuracy to meet the localization requirements.