Research Of 3D Modeling Based On Sequence Images

Reverse engineering is an important means of product innovation. The 3D data of the parts' surface is the basis for reverse engineering. Recently the demand of high-accuracy 3D measurement is very big in a variety of computer vision applications, for instance, plane, car, ship, medical imaging modality, robot vision, human-computer interface, cultural relic preservation, 3D face biometrics recognition, etc. Based on the binocular stereovision theory, structured light 3D vision theory and the technology of camera's calibration, a system of 3D modeling is proposed and implemented in this paper, which using sequence images. The system has the merits of non-contact survey, good precision, fast speed, simple instrumentation, etc.To reconstruct the object in three-dimensional space with only two different view points is research focus in computer stereo vision. The fundamental principle of computer vision with dual-cameras imitates the visual perception procedure of mankind. We can get images in different angles from two view points and then calculate the 3D information of images pixels based on triangular measure. In the computer vision,the input image is the 2D image which is a matrix included many factors such as the geometry character of 3D object, the color of objects, the surface character of material and the parameters of camera. The process of computing these parameters by image is an inverse problem which is not linear. The solution is not exclusive and very sensitive to noise and discretization error.To construct the 3D model, this paper will mainly carry out the following work: calibration of camera, extraction of active clue feature in images, stereo matching and 3D reconstruction. Calibration of camera's interior and external parameters, active feature selection and stereo matching are key techniques of 3D reconstruction of the computer vision.The first step is the calibration of the camera. The camera image model determines the relation between the position of a 3D point and its 2D image. We compute the camera parameters by using Zhang Zhengyou camera calibration algorithm in this paper. The main idea of the algorithm is as follows: Print a pattern picture and attach it onto a planar surface; Use one camera to get several images from different angles; Detect the corners based on image processing knowledge; Compute the camera parameters according the physical and image coordinates of the same feature point.To get the physical characteristic of 3D object, we use projector to project structured light on 3D object. Because of the figure characteristic on object's surface, the structured light has the deformation. And the deformation is different in observer's eyes from different angle. Such distortion stripe carry outline characteristic of the object. Then we get a series of 2D images from one camera which is moved regularly in horizontal direction on a sliding table. The camera is moved in same distance in horizontal direction and same focus every time. After image pretreatment, we obtain thin and robust edges by using improved Canny edge detection algorithm in which I notice the effect from the figure of light projection. Considering the influence of illumination and the noise, we get better edges by using thin algorithm and stripe amendment algorithm which is proposed in this paper. Because of the camera moving in horizontal direction, the outermost physical edges of object in different images are different. So we should throw off these wrong edges which would affect stereo matching. Canny edge detection algorithm is often used in image processing. It can extract edges of image in a good way compared with other algorithm. Compared with ordinary Canny algorithm, improved Canny algorithm reduce computation time and ensure detection accuracy. Stripe edges are more exact by stripe amendment processing. After these processing introduced above, we get good active clue feature in images. They are the foundation for stereo matching and 3D reconstruction.Stereo matching from different images is the most difficult and important step in 3D reconstruction. Because the camera is moved regular in horizontal direction, we can use the character in some place for stereo matching. After summarizing many exiting image matching algorithm, an image matching algorithm is proposed in this paper which suits the actual need. In the experiment,we get a series of 2D images from one camera which is moved regularly in horizontal direction on a sliding table. The matching points have the same pixel coordinates in vertical direction from different images. So the problem of matching is mainly in vertical direction. According to experiment characteristic, matching have two steps in this paper: coarse matching and accurate matching. In coarse matching step, partial points are in early matching based on Least Square Method. We do not compute Least Square distance on all points in matrix. So it raises the computation speed. Accurate matching step has two parts: stripe marking and corresponding points matching. Because stripes do not carry enough matching information, we mark stripes to reduce wrong matching. We choose several rows in the image matrix as"seed rows"and mark points of these rows in two kind of alternate colors. Then we mark other points with the same color as points of"seed rows"in the same stripe. Stereo matching of corresponding points mainly use the unique constraint, the compatible constraint and the continuous constraint.Finally, we obtain 3D points from 2D matched points in different images, based on the theory of binocular stereo vision and Least Square Method.Based on sequence images, a system of 3D modeling is proposed and implemented in this paper. The system includes calibration of camera, extraction of active clue feature in images, stereo matching and 3D reconstruction. The program has low coupling, so the system is convenient for improvement and update. We obtain satisfactory 3D modeling result after the experiment. This system has the value of further research and the good application prospect.