Research Of3D Reconstruction Based On2D Images

Computer vision is to use computer to perceive and understand images, so that it can dosomthing that is similar to human visual.3D reconstruction play an important role in computervision, and it is widely used in areas such as cultural relics recovery, medical imageprocessing and apparatus manufacturing, and so on. While the traditional3d modelingmethods have many limitations and complexity,this paper puts forward a method that thecamera absorb two images, and we get depth through different perspective of the scene.This paper based on analysis and research, designed a3d reconstruction system based onthe structure light.The main steps of the structure technology: the surface of object isprojected the black and white stripes grating structured light. The strips will have distortionalong with the ups and downs of the object’s surface. The camera shot a set of photos tocomputer, then computer software analysis the images,finally we obtained the depth of theobject information. This system includes a stereo calibration module,image processingmodule,the three-dimensional matching module,3d coordinate calculation and point clouddata display and trianglation module.This system put space geometry, projection geometryand computer image processing together, make full use of the knowledge and has practicalvalue.Stereo calibration module mainly puts forward a stereo calibration method based onbinocular camera. The main steps for calibration: shooting templates, angular point extraction,OpenCV stereo calibration, and image correction.We get the internal and external parametersand the distance of two cameras through calibration, and once we have internal and externalparameters, then we get projection matrix. This paper realized stereo calibration method usingthe open source library OpenCV. Inevitably the camera has distortion, and this paper producesa corrected image.Image processing module do a series of pre-treatment to images, firstly, we convert thecolor image to gray image, so that we remove some useless information.After that we extractthe edges of the object.Through the analysis and comparision, we choose the improved Cannyalgorithm.Due to the influence of the nosie, the edge which we get is not the ideal edge, sonext we optimize the edge. First, in order to remove the edge which is not structure light edge,we choose mathematical morphology method. Medial Axis Transform algorithm which isused for structure light is proposed in this paper. We refine the images with improvedrefinement algorithm first. Then in order to remove the points out of the trend, outlier andminimal discontinuous points to get central axis of the edge, we chose eliminate template.Stereo matching is most important step in the process of3d reconstruction, choose good matching algorithm can make matching accuracy higher. Stereo matching module match twoimages with two step method.The first step, we should estimate parallax and calculate themigration distance about the two images. Create a window, then calculate the minimummigration distance by using the least square method to every point on the window.The secondstep, to increase information, we firstly mark the strips, then choose Rank transform matching.While there are more than one matching point, we chose Census transform for furtherjudgment to find the most accurate corresponding point.The last module,3d coordinate calculation and point cloud data diaply and trianglationget z coordinate by using projection matrix based on binocular vision principle,and then weget the space of point coordinate. We also display all the space of point, and design somefunctions like rotating and rooming, so that we can observe the object from all directions.Finally, in order to better perform the object, we developed triangulation program which candeal with do these points. Considering its advantages and disadvantages and the efficiency, wechose the method based on region-growing surface reconstruction algorithm by comparingand analyzing the existing triangular mesh algorithm. And we chose Delauany criteria toconstruct the triangle.This program are designed using the3d library called OpenCV and OpenGL with MFCframework development. We use Visual++6.0for our development platform. The program haslow coupling with the very different functional module, so the system is convenient forimprovement and update. This system has the value of further research and the goodapplication prospect.