Research On Calibration Method Of Monocular Camera Based On Circular Array Target

With the continuous development and progress of science and technology,visual measurement,as a new technology that uses machines instead of human eyes to locate and measure spatial geometric objects,is attracting more and more attention.Camera calibration is an important basis of vision measurement,and the precision of camera calibration directly affects the performance of vision measurement system.The circular array target is one of the commonly used planar calibration targets.Due to the deviation of lens perspective projection,when the optical axis of the camera is not perpendicular to the plane of the target,the target appears as an ellipse on the imaging plane.The center of gravity of the ellipse is obtained by the traditional center of the circle positioning method,rather than the actual imaging position of the center of the circle.In addition,the traditional L-M(Levenberg-Marquardt)optimization algorithm has some problems such as insufficient global search ability due to the large number of parameters to be optimized when optimizing the initial value of camera parameters obtained from the camera imaging model.These factors restrict the improvement of camera calibration accuracy.Therefore,this paper focuses on solving the problem of high-precision location of the center of the circular array target imaging and solving the initial value of camera parameters and its optimization.1.An error modeling method is proposed for the center positioning error of the circular target and an error compensation strategy is combined to improve the center positioning accuracy.Firstly,the simulation image set of the circular target image is established,the image is preprocessed and the center coordinates of the image are located by elliptic fitting method.Then,GA-BP(Genetic Algorithm-Back Propagation)neural network was constructed and trained to establish the relationship model between center location error and lens pose.Finally,the error model is used to predict the center positioning error value and the error compensation strategy is combined with the center positioning error compensation.The experimental results show that the prediction performance of the constructed GA-BP error model is obviously better than that of the comparison model,and the accuracy of the center of the circle is higher after the error compensation,which indicates that the error modeling and error compensation methods can effectively improve the accuracy of the center of the circle.2.A method of camera parameter optimization based on multi-population genetic algorithm is proposed.First,the monocular camera calibration system was built,the circular array target image was collected,and the target image was processed to obtain the data set for training the GA-BP circular center positioning error model.The constructed error model was trained and tested.The experimental results showed that the mean value of the absolute value of the prediction error of the model was lower than 0.0060 pixels.The prediction accuracy of center positioning error and the fitting degree of error model are consistent with the simulation results.Then,the elliptic fitting center of the calibration image was compensated with the error compensation strategy,and the center of the circular array target was numbered to obtain the 3-2D coordinate point set of the center of the circle.The singular value decomposition method was used to solve the initial value of camera parameters,and the 5-parameter model of lens distortion was constructed.The initial value of the distortion coefficient was solved by the least square method.Finally,a multi-population genetic algorithm based on adaptive computation of crossover probabilities and mutation probabilities is constructed to optimize the initial values of camera parameters.Experimental results show that the average value of camera calibration reprojection error after center positioning error modeling and compensation is 0.1351 pixels,and the camera calibration reprojection error optimized by the improved multi-population genetic algorithm is further reduced to only 0.0775 pixels.At the same time,when the number of calibration images is reduced,The fluctuation amplitude of camera calibration reprojection error remains stable.