The Research On Camera Calibration And Matching Method In Binocular 3D Measurement System

Binocular vision 3D measurement technology is the closest way to the biological vision system, so this method is a hot topic of research all over the world. The technology projects images to the surface of the measured projects with a projector, and then the deformed images are captured by two cameras. At last, the 3D data of the object can be calculated by image processing algorithms. In this thesis, a study on issues related to camera calibrations and stereo matching algorithm is carried out and new algorithms are proposed. The research is applied in binocular vision measurement system in the lab. The main work in this dissertation is outlined as follows:1. An improved camera calibration optimization algorithm in binocular vision measurement is proposed. The existing optimization methods focus on minimizing the sum of reprojection errors for single camera in 2D image pixels coordinate. But the measurement process is carried out in 3D coordinate system by both two cameras. The measurement coordinate system is not the same with the optimization coordinate system. To deal with the problem, a centroid-based optimization method is used to optimize the initial values of parameters firstly. This method reduces the coupling degree of the calibration parameters in the optimization process. And then the distance error between the ground truth in 3D measurement coordinate system and the reconstructed point based on triangulation is minimized. The results of experiments show that campared with the conventional methods, the novel optimization algorithm can improve the accuracy effectively.2. A three-dimensional shape measurement method based on the projection of one digital speckle pattern and phase-shifted fringe patterns is proposed. The preliminary corresponding point of each pixel can be obtained by the projection of a digital speckle pattern onto the surface of the project, and the correspondence refinement using wrapped phase map is performed, which is calculated by projecting phase-shifted fringe patterns. Unlike the conventional methods, the proposed method can determine the location of the correspondence points in different views by the digital speckle pattern, and the phase unwrapping algorithm is not required to obtain the absolute phase, so each pixel can explore its corresponding point directly from the wrapped phase map. Experiments demonstrate that the proposed method can achieve high-quality three-dimensional measurement and shorten the time as compared with the traditional method. Therefore, the method enhances the practicability of the system.