Research Of 3D Reconstruction Based On Multi-View Stereo Vision

Deformity teeth will not only affect the appearance of human,but also do damage to patient’s confidence.Moreover,the oral cavity has more functionalities.Dentists use plaster model to stimulate the movement of teeth and jaw,aiming to help patients recover.Plaster model can also help analyze the shape of teeth.However,traditional plaster model have many problems,one of which is hard to preserve.It is also against the trend that medical data will be digitalized and visualized.What’s more,in some western countries,with the use of cameras putting into patient’s mouth,doctors can implant those crown made by 3D printing.Facing such problems,based on stereo vision,this article propose a method to 3D reconstruct the object.Using calibrated cameras to take photos from different views,I can achieve the success of 3D reconstruction of an object with the help of SIFT and structure from motion.The main tasks of the article come below:Step 1:Based of SIFT,build the Gaussian pyramid and sift descriptors.Match the features using k-d tree.At last,erase the mismatch with the help of RANSAC,Step 2:Using the principal of stereo and the matched features,I can compute the track of cameras while taking photos.Triangulate the features to get the original position.Optimize the result by bundle adjustment.Step 3:Design the environment and experiment using the ideas of step 1 and 2.The experiment uses industrial camera and canon single-lens reflex camera to take photos,aiming to verify the feasibility and reliability.The result of the experiment shows that it is feasible to achieve 3D reconstruction of an object by taking photos.The point cloud data can reflect the local information of its surface.Using the method of this experiment can push the digitalization and visualization of medical data.