Research On Spatial Three-dimensional Reconstruction Method Based On Monocular Vision

As a significant branch of AI(artificial intelligence)technology,machine vision has been widely used in industrial production.Three-dimensional reconstruction is a key technology in machine vision,and plays a significant part in intelligent manufacturing,satellite navigation,archaeological research,virtual reality,armarium,remote education,military technology and other fields.The optical imaging model of surface of object was established based on the monocular vision system,and the computational method of the height information of the object surface based on image gray value was proposed in this dissertation.The main contents are as described below:1.Analysis of image characteristics of monocular vision system.Firstly,the characteristics of the monocular vision system and its effect on imaging quality were analyzed to establish the essential foundation of the proposed three-dimensional reconstruction method.Secondly,the imaging model of monocular vision system was established,and then the transformation model was established for the surface point coordinates,which is from the world coordinate system to the image coordinate system.Finally,based on Zhang's calibration method,camera calibration experiments was conducted by using autonomous checkerboard calibration board to determine the camera internal parameters and distortion parameters,which provided a prior knowledge and the theoretical basis for subsequent three-dimensional reconstruction experiments.2.Spatial three-dimensional reconstruction method based on monocular vision was proposed.Firstly,color image was preprocessed.The image filtering was used to remove the Gaussian noise and salt-and-pepper noise in the image,meanwhile,the image segmentation based on color threshold was used for surface region extraction of the object,background redundant information removing and three-dimensional reconstruction acceleration.Secondly,the optical imaging model was established for object surface based on monocular vision system,and then the monochromatic light height information mapping model was established by analyzing the influence factors of the image gray value in the optical imaging model.Since the surface height information could not been calculated by monochromatic light height information mapping model,the trichromatic light height information mapping model was established and the solution of calculating surface height information was given in the dissertation.Thirdly,there were unknown parameters which could not been determined by calculation or direct measurement in the height information mapping model,the diffuse reflection coefficient calibration experiment was designed in this dissertation.Finally,due to the huge computational complexity of solving the equation,the height information mapping model was simplified to accelerate spatial three-dimensional reconstruction by simplifying the equations and cutting down number of unknowns.3.The experimental platform based on the proposed spatial three-dimensional reconstruction method was established.Firstly,the diffuse reflection coefficient calibration experiment was carried out to obtain the unknown parameter.Secondly,the spatial three-dimensional reconstruction experiment of the pins and solder joints of integrated circuit chip and the glue on the camera module chip were carried out.The experimental results showed that the method propose in the dissertation is suitable for the spatial three-dimensional reconstruction of tiny object.By comparing other three-dimensional reconstruction method,the validated method proposed in this dissertation were proved.Finally,the experiments on speed improvement of three-dimensional reconstruction was conducted by taking the glue reconstruction as the example,and its effectiveness was proved.