Multiple View 3D Reconstruction Algorithm And Application

3D reconstruction is one of the most important research sub-domains of computer vision,and it's also widely used in a variety of fields,such as film animation,robot navigation,medical imaging and cultural heritage protect.The objective of multiple view 3D reconstruction is to estimate 3D structure of the scene through several images taken from different views.Combining algorithm with application,this thesis studies the key technology of 3D reconstruction,then designs a scheme to reconstruct 3D point cloud from images,and applies the implemented algorithm to the problem of antenna angles measurement.The main contents of the thesis are as follows:This thesis first studies the key technique steps and algorithms in 3D scene reconstruction.Combined with the projective geometry,the detail of camera model and parameters are given,and a method to estimate these parameters is introduced.To detect the robust and accurate overlapping regions,we use image feature detecting techniques to match images in the sequence.By using epipolar geometry methods,false correspondences which don't satisfy the geometric constraints are filtered out.Then the camera poses are estimated from valid correspondences information.This thesis designs and implements a pipeline for multiple view 3D scene reconstruction.By matching images in sub-sequences and chaining the results on the whole sequence,efficiency of image matching is improved.Then a good image pair is chosen to get an initial structure according to the rule.A new camera pose is estimated from 3D-2D correspondences,and new 3D points are added to the point cloud by triangulation.After an iteration on adding new image,estimating pose,triangulation and bundle adjustment,finally all available images are added to the result and a 3D point cloud which represents the scene can be get.Besides incremental method,the thesis also studies the method about global reconstruction which computes the global rotation of each camera first.Because of the high cost and risks of antenna measurement in communication equipment maintenance,we apply the implemented reconstruction algorithm to the measurement of antenna angles.By taking several photos of the antenna scene,camera poses can be recorded while the cloud point structure is computed,then the entity which can represent the antenna is extracted and triangulated.Using the pose information recorded by the device,the scene and 3D entity representing the antenna are rotated to the world coordinate system,and the estimation of the antenna angles can be performed after that.The experiments show,the multiple view 3D reconstruction algorithm in this thesis can get an available point cloud about the object scene.The result is accurate according to reprojection error.From the perspective of visual effects,the point cloud structure is correct and can be recognized easily.The scheme which uses the result of 3D reconstruction to measure antenna angles,also gets a good experiment result and solves the problem of antenna angles measurement effectively.