Research Of 3D Reconstruction Based On Structured Light 3D Vision

As we know, if we can get the actual size of object, we can easily construct the 3D model of the object, but the measurement precision of the object's size will directly affect the effect of the 3D model, and sometimes it is difficult to get the actual size of some complex object. So human beings start thinking about obtaining the 3D information of object directly without knowing the imprecise measurements of the object, thus avoiding the complicated mesurement stage, and nowadays this technology is also an important area of computer vision. In this paper, a system of 3D modeling, which is based on the theory of structed light and computer vision, is designed. This system can construct the 3D model of object without touching the surface of object, it will not damage the object, and accuracy of the model constructed is ideal, so the system has a certain practicality.To reconstruct the 3D model of object, this paper has mainly done the following work: calibration of camera, imaging preprocessing and feature extraction, stereo matching,3D reconstruction and the display of 3D point cloud data. In these works, the process of feature extraction and 3D reconstruction is most important.Stereo calibration is a necessary step in 3D reconstruction. The main work of the camera calibration is to solve the matrix of the intrinsic parameters and matrix of the external parameters. In this paper, firstly we calibrate the camera using Zhang Zhengyou camera calibration algorithm, and then,we compute the position-relationship using StereoCalibration. We combine MFC with OpenCV and implement the stereo calibration algorithm with intuitive interface.The accuracy of feature extraction directly affects the accuracy of the3D model we reconstructed. In this paper, the surface of object is projected the black and white stripes grating structured light. The strips will have distortion along with the ups and downs of the object's surface, and the distortion is fraught with the depth information of the object. In this paper we slide the digital camera in the horizontal direction and then get a series of 2d figure of objects in different perspectives. In this process, the focus length of the camera and the distance we slide each time is constant. As the preprocess, we remove the background and convert the color image to gray image. After the preprocess, we extract the edges of the object with the improved Canny algorithm. Canny algorithm satisfy optimality criteria, the results of edge extraction using Canny algorithm are always ideal. The improved Canny algorithm put forward in this paper has better effect in noise removing and edge extraction. As the edges we extracted are not single pixel width, we thin the edges we got to make the "thickness" of the edges reaches single pixel width. For the subsequent stereo matching work, we restore the original grayscale and the gradient direction of each edge point in the edge picture. After that. according to the actual situation, some local edge points deviated from practical direction are revised, and finally we get ideal edge profile features.Stereo matching is the most important step in the process of 3D reconstruction. In the process of stereo matching, there are many matching criteria, which add many constraints. Generally, commonly used matching criteria include continuity criterion, compatibility criterion, uniqueness criterion, epipolar constraint criterion and ordering constraint criterion, etc. In this paper, we slide the camera only in the horizontal direction, so the imaging planes of the two images are parallel, and the polar of the two images are in the same horizontal line. That makes the calculation of the polar easier. In this paper, we put forward a matching algorithm based on four images. While matching, firstly we estimate the relative position of the object in adjacent images using the method of calculating the least square distance. Secondly, we find the possible matching points with the similarity of the grayscale. And at last we find the best matching point in the possible points using the gradient angle.Using the matching points we obtained combined with the intrinsic parameters and external parameters we got in the process of calibration, we can easily calculate the spatial coordinates of the points. In order to observe the result of the experiment, we designed a program to show the 3D point cloud model. This program are designed using the 3D software development kit called Direct3D. In order to adapt to different model, the program increases the function of rotation,translation and scaling.Through experiments we can know that the result of the 3D reconstruction is ideal, the 3D point cloud data we obtain has high accuracy. The 3D reconstruction system we designed in this paper doesn't need expensive devices, and we can operate it easily. If we improve the system in some aspect, there will have broad application prospects.