The Study Of Camera Intrinsic Parameters Calibration Using Genetic Algorithms

With the development of modern industry, large-scale and high-precision 3D measurement technology is demanded more and more urgently in the fields of navigation, spaceflight, war industry and Automotive Industry. The traditional coordinates measurement technology can not meet the requirement mentioned above, but the vision coordinates measurement technology, which is a rising 3D spatial coordinate measurement technology based on computer vision technology and coordinates measurement technology could fit to the modern production which needs measurement technology that holds higher measurement precision and larger measurement scale from a bran-new coordinates measurement idea.The basic working principle and main purpose of the vision measurement system is to calculate the geometric information of the three-dimensional object from the images which are obtained from the camera, and reconstruct and identify them. However, the relationship between the 3D points of the three-dimensional object and its 2D image points is determined by the geometric models of the camera which are called camera parameters. And those parameters must be obtained only by the experiment and calculation, whose process is just called camera calibration. That is to say, camera calibration is the primary problem and the premise of the application in computer vision. As the key component of the vision measurement system, the precision of the intrinsic parameters of the camera could have a direct impact on uncertainty of the entire system. So, the intrinsic parameters of the camera must be calibrated accurately.In this paper, on the basis of a wide range of analysis and comparison about the existing classical methods of the conventional calibration, the relevant knowledge of the camera calibration was analyzed in-depth. In allusion to the feature that the intrinsic parameters of the camera must be calibrated accurately in the field of high-precision coordinate measurement and that the conventional optimization algorithms exist the defects, namely, susceptibility to getting trapped in local extremum, a new camera calibration method based on genetic optimization algorithm was proposed. By experiment in the last, this method is more accurate and robust than conventional method and could meet the demand of the practical application.Firstly, after the analysis about the major nonlinear distortion of camera, a very comprehensive model of camera was put forward, which was an effective mathematical model for the camera calibration. Especially, the distortion types of camera could be easily added or reduced in this model. If we don't need high accuracy in some occasion, the number of the parameters needed to be calibrated could be reduced, which will greatly reduce the time of optimization.Secondly, in order to make genetic optimization algorithm faster and ensure the optimal solution algorithm to converge to the global optimal solution, an initial value estimation for the follow-up genetic algorithm optimization is very necessary. The pin-hole model of camera in which the nonlinear distortion was neglected was used in the initial parameters estimation in the paper. The method based on planar pattern for camera calibration only requires the camera to observe a planar pattern shown at a few different orientations, and the images obtained. Corner points are extracted to calibrate the correspondence between the image planar and the planar object. The results showed that the method is very efficient and could get a satisfied estimated initial parameters.Thirdly, in order to extract the corners in the template images effectively, the image preprocessing process was analyzed in detail and a sub-pixel accurate corner extraction algorithm was put forward in the paper. Through the corresponding experiment, median filter, which can suppress the random noise especially pulse noise and pepper and salt noise without blurring the image edge was decided to be used for the image processing. A sub-pixel accurate corner extraction algorithm was obtained from the observation that any vector from true corner location to its neighborhood is orthogonal to the image gradient. So the sub-pixel algorithm was calculated iteratively by minimizing an error function. The result of the camera calibration showed that high precision can be acquired by using this sub-pixel algorithm, which could improve the accuracy of the camera calibration greatly. Based on the ideas above, the program about the estimation of initial parameters of camera was be developed, which was use the Intel's open-source computer vision libraries in VC + + 6.0.Finally and importantly, the scheme of genetic algorithm optimization of the intrinsic parameters of the camera was designed. And the camera intrinsic parameters calibration with high accurately was realized using all the external parameters and some of the intrinsic parameters of the camera from the initial estimation. In allusion to the characteristics of the intrinsic parameters of the camera, the chromosome vector and the definition of fitness function was put forward. As the realization of genetic optimization algorithm in the application of genetic algorithms has a very important position, the individuals coding, selection of initial population, calculation of the fitness, genetic operator and termination of the designed criteria were analyzed in the paper. And we finally got the optimized results of the camera parameters from the program we designed.Error Analysis showed that the average absolute error of the projected image points is 0.2559 pixels and the standard deviation of them is 0.1393 pixels. Compared to the results from the initial estimation, we got a better result and the distortion of the image points in the edge of the template model was restrained well. In order to validate the robustness of the optimization we put forward, the simulation experiment, in which different levels of Gaussian noise was put in the image data, was designed. And compared to the famous Levenberg-Marquardt nonlinear optimization algorithm under the same conditions, the camera calibration based on genetic algorithm had a high robustness and got a better result.