The Research On A Distortion Calibration Method Based On Adaptive Fuzzy Neural Network

Telescope imaging/collimation measuring light tube is used for self-adaptive recognition and interpretation of slit position of collimator,star target position of star simulator and normal position of prism.In the process of telescopic imaging/collimation optical tube working,because of the errors in the design,processing and assembly of telescopic imaging/collimation optical system,distortion will occur,which makes the position of the actual image point deviate from the ideal imaging position.Therefore,the image information can not accurately reflect the three-dimensional spatial position information,which will affect the results of performance testing.In order to obtain the exact relationship between the position of the space target and the corresponding image point in the image,it is necessary to calibrate the image plane of the telescope imaging/collimation optical system.Therefore,in order to guarantee the detection accuracy of telescope imaging/collimation measuring optical tube,this paper realizes the distortion calibration of telescope imaging and collimation measuring optical path in telescope imaging/collimation optical system by studying the image plane calibration method based on neural network.The specific work done in this paper mainly includes the following four aspects:1.The existing image surface calibration methods are analyzed.At present,among all the image surface calibration methods,the linear calibration method is not accurate;the non-linear calibration method adds distortion factors to improve the calibration accuracy,but the camera imaging model used is difficult to describe an imaging system completely.Neural network has the characteristics of infinite approximation to arbitrary nonlinearity.This paper presents a calibration method based on neural network,and illustrates the feasibility of the calibration method in this paper.2.Image surface calibration is the process of obtaining three-dimensional spatial information from two-dimensional image information.Detecting the corner coordinate information in the image is an preliminary step in the calibration process.The accuracy of the detected coordinates will directly affect the accuracy of the calibration results.In this paper,based on Harris corner extraction algorithm,sub-pixel coordinate corners are extracted by utilizing the objective phenomenon that the line between corners and adjacent pixels is perpendicular to the gray gradient of the pixels in the checkerboard.In this paper,the algorithm is realized by MATLAB programming,which verifies that the accuracy of corner coordinates extracted by this algorithm can reach sub-pixels.3.Aiming at the problem that the generalization ability of the neural network has a great influence on the calibration results in the calibration method proposed in this paper,an adaptive fuzzy neural network system based on maximum correlation entropy(MCC)criterion is introduced into the calibration method.The neural network system can solve the highly non-linear problem more effectively,and has good robustness,higher accuracy and faster operation speed,so as to improve the accuracy of system calibration.4.A telescopic imaging/collimating optical system is designed and tested based on the refractive system.The experimental steps are as follows: Firstly,the calibration plate is placed on the focal plane of the reference collimator to simulate the infinite target,and the camera system composed of the telescope imaging/collimation system and the CCD collects the image of the calibration plate of the telescope imaging/collimation system;then,the calibration plate is placed on the focal plane of the telescope imaging/collimation system,and is reflected by the mirror and imated on the plane of the CCD detector to collect the image.Finally,the corner coordinates of the calibration board are extracted,the image coordinates are taken as input,and the corresponding non-distortion coordinates projected on the calibration board are taken as output to train the neural network.The calibration of the optical path of telescope imaging detection and the optical path of collimation detection is realized.