Study On Calibration Technology Of The Microscopic System Used In Micro-assembly

Camera calibration is the process to determine the relationship between the two-dimensional (2D) geometrical information from images taken by camera and the corresponding spatial three-dimensional (3D) geometrical information of an object, and can be beneficial to realize target identification, location, and reconstruction accurately. In this case, camera calibration is one of key techniques of machine vision. In recent years, with the rapid development of micro-electro mechanical systems (MEMS) technology, the micro-vision technique has got great attention by domestic and international. However, the traditional camera calibration techniques cannot be directly adapted to calibrating micro-vision systems due to the narrow depth of field and small field of view of microscopes. So it is necessary to study the improved calibration technology that can be used in micro-vision systems.In this thesis, the calibration techniques of zoom micro-vision systems have been researched. The imaging model of a micro-vision system is established. Based on the multi-view calibration approach of Zhang, an improved calibration algorithm based on single homography matrix decomposition has been proposed. The experimental setup has been established, and the calibration accuracy has been tested on microscopes with fixed parameters and varying parameters, respectively.The major research works completed in this thesis include:1. The structure of the micro-vision system for the micro-assembly system and the camera imaging model are studied, the pinhole imaging model is used to simulate the imaging process of microscopic vision system with fixed parameters. Based on the features of zoom microscope, a simple online camera calibration method is established, which can simplify the complex process of recalibration after the focal length is changed.2. Aiming at the short depth of field and small field of view of microscopes, and installation of the calibration plate in parallel with the CCD image plane, an improved camera calibration method based on the single view homography matrix decomposition is proposed. In order to test the stability and applicability of the proposed algorithm, the experimental setup is established. Microscopes with fixed parameters and varying parameters are calibrated by this algorithm, and the results show that the mean reprojection error is less than 0.5 um. 3. In order to obtain the basic data by ellipse detection from the calibrated images, the random sample consensus (RANSAC) estimation algorithm is employed. The RANSAC estimation algorithm uses the Euclidean distance as the threshold to select the inner points, and uses the least-squares algorithm to estimate the ellipse parameters. Experiments show that the RANSAC estimation algorithm is good at detecting ellipse when there are more serious noises or occlusion in microscopic images.4. The variation of the principal point along with varying focal length is analyzed. The principal point is calibrated by the method of varying focal length. The results show that the principal point does not change with adjusting the focal length, and the displacement error of the main point is±0.5 pixel.The research works in this thesis lay the theoretical and technical foundation for establishing micro-assembly systems.