Reconstruction Of 3D Points Based On Single Camera And The Application In Large Structure Defomation Measurement

The measurement of target points three-dimensional coordinates based on single camera is one of digital close-range photogrammetry measurements. It uses a single camera take multiple pictures that contain target points at different place and from different angle. By using digital image processing and photography measurement algorithm to reconstruct the three-dimensional coordinates of the target points. This kind of measurement method for its high degree of flexibility and adaptability has become a hot direction of the optical non-contact measurement. In this paper, the research which is based on photogrammetry and computer vision theory include imaging model of camera, camera calibration, the design of the target points, recognition of target points and the sub-pixel localization of circle’s center, photo orientation method, non-coding point matching method and the self-calibration bundle adjustment algorithm. The digital close-range measurement software CMM2014 is written alone by using these algorithms. The measuring accuracy of the system is analyzed through the experiment. And the system is applied to the measurement of yoga ball and deformation experiment of civil steel structure. The researchs and achievements are as following:（1） The classification of the camera is introduced. Some basic knowledge that such as lens imaging, dispersion circle, depth of field and the field of view were analyzed. The mathematical model of camera imaging is proposed by the formula derivation. The calibration method of Zhang Zhengyou and self-calibration method based on bundle adjustment were introduced.（2） The coding and non-coding target points were introduced. The capacity and coding process of 15 bite coding target points were analyzed.（3） According to the coding target points-the recognition steps of target points which include image preprocessing, edge extraction, feature selection and code point decoding were presented. The traditional center sub-pixel positioning method which include gray centroid localization algorithm and centroid method based on sub-pixel edge ellipse fitting were introduced. At last three improved multi-threshold centroid localization algorithms were proposed.（4） The experiment of target point recognition proved the recognition algorithm in this paper has higher stability and accuracy. Through the center sub-pixel positioning simulation and real images of all sub-pixel positioning algorithm mentioned were analyzed. The multi-threshold centroid localization algorithms have higher accuracy which is better than 0.02pixel.（5） Aiming at the auto orientation of images, a preliminary orientation method that the autobar is not necessary is present. The match method of noncoding points in multiple images is present by the analysis of epipolar line matching method. Simplified solution of normal equation was deduced for the self-calibrated bundle adjustment, and this method is faster.（6） The Composition of CMM₂014 digital close-range photogrammetry system was introduced. The measurement accuracy of the system which is better than 0.2mm/m was analyzed by experiments. The system is also used to measure the volume of yoga ball and civil steel deformation successfully.