Monocular Camera Calibration And Object Localization Based On Modified Genetic Algorithms

Machine vision has been developed greatly since its first appearance in the mid-1960s,and its theory and application researches have also gained rich achievements.Machine vision has extensive and mature application in many fields such as scientific research,social life,industrial production and so on.In the robot visual servo,photogrammetry,and other vision application fields,the camera calibration is the fundamental premise to extract precise 3D information from the image,and the calibration result can directly influence the precision of the visual measuring results and the stability of the visual processing algorithm.In this thesis,we have systematically studied the camera calibration theory and improved the calibration techniques based on the existed researches at home and abroad.These improvements include improved image feature extraction in the process of calibration optimization of camera parameters and monocular camera positioning technology,and experiments were designed to validate the effectiveness and accuracy of the proposed algorithm.Main work and innovations in this thesis are as follows.1.Based on the analysis on common features used in the camera calibration process,an improved feature extraction algorithm was proposed for circular array template,which utilized the characteristics of circular array template,and quickly found the projection of the template on the image plane through 3 steps of filtering.Finally,the accurate positions of the ellipse-centers were calculated with the improved Hu method and sub-pixel method.2.A new camera calibration algorithm was proposed,which optimized the solutions for the camera intrinsic parameters with a modified genetic algorithm.The algorithm firstly obtained the initial values of the camera intrinsic parameters through Zhang calibration method,and then a modified genetic algorithm was used to iteratively optimize the initial values to get more precise camera intrinsic parameters.This method was applied to two different types of camera calibration templates,i.e.,checkerboard and circular array template.The proposed methods through experiments are demonstrated that it can achieved more accurate values of the camera intrinsic parameters than Zhang's camera calibration method.3.An improved localization method with monocular vision was proposed,which used two target images for localization.Artificial marks were used to avoid mismatches caused by using RANSAC method in the matching process,and a modified genetic algorithm was used to solve the fundamental matrix.The proposed algorithm could achieve more precise localization of the objects.