Research On The Camera Calibration Method For Three-dimensional Detection

With the vision theory and electronic technology increasingly growing and maturing, machine vision technology has been more widely used in medical equipment, industrial detection, artificial intelligence and other fields, especially in three-dimensional detection, where its application is becoming increasingly mature. Because the stability and accuracy of the camera calibration algorithm will have a great impact on the entire visual system, so study camera calibration technology is of great significance. Camera calibration is a procedure to determine the internal and external parameters of camera based on the corresponding relationship between world coordinates and image coordinates of feature points on the object's surface, in which the difficulty is how to effectively extract features and rectify distortion, the above two aspects are focused on in this thesis.Firstly, the classic feature extraction algorithms are enumerated in this thesis, aimed at the calibration template used in this thesis, harris corner detection algorithm is selected to extract features, and on this basis, two corresponding improvement are made, the one is circular fitting algorithm, the another one is combining the Forstner operator and Harris algorithm, through experiment and contrast, it can be found that both of two algorithms could reach sub-pixel precision level, and the accuracy of first algorithm is relatively higher than the second algorithm.Secondly, the method of camera calibration was studied. The advantages and disadvantages of three calibration methods are separately introduced in this thesis, the calibration method which is suitable for three-dimensional detection of industrial is selected after comprehensive analysis and comparison, and as reference, a repeated calibration method based on black and white checkerboard is proposed, which calculated the linear solution for a single camera at first, and then refined the solution by nonlinear optimization fine. After considering the nonlinear distortion of camera and optimizing the solution using nonlinear method again, the accuracy and stabilization of calibration are improved, the binocular camera can be calibrated after the parameters of monocular camera are determined.Finally, the whole process of camera calibration are completed on the matlab experimental platform. The experiments include image pre-processing?sub-pixel feature point extraction?the calibration of internal and external parameters in single camera system?the calibration of parameters in stereo vision system.