The Research Of 3D Model Reconstruction Based On Video Sequence

The products of Virtual Reality(VR)and Augmented Reality(AR)appear constantly in the current time.It has brought people many entertainment,and big electronics factories also announced their plans to explorer into this field.It has improved the enthusiasm of the research of 3D model reconstructions since the research of VR & AR depends on this basis research.The research has also improved the related subjects such as computer stereo vision and computer graphics.Essentially,the research of 3D model reconstruction based on video sequences is still in the same condition of the research of 3D model reconstruction based on images.Its core problems are feature extraction and feature matching and camera calibration.Different feature descriptors describe different properties,and it is the same to the feature matching methods.It has made the camera calibration a big challenge due to the moving and the change of focal length when recording videos around an object.In this article we give a set of feasible scheme for 3D model reconstruction.Firstly,we analyse the current common used feature descriptors and different feature matching methods.Aiming at the existing problems of SIFT feature points,we put forward an improved SIFT feature extraction algorithm.The experiments prove that this algorithm can reduce the computation time effectively,and can also remove the parts of the low quality points effectively.An improved k-d tree methods has been used for feature matching,then we get a set of point pairs after robust estimation and removing error matching pairs.Since every image has its own electronic information,we use this information to estimate camera parameters.Then we get the other camera parameters according to the dual-view geometry.Finally we can get the three-dimensional coordinates of the point by using the method of triangulation,the sparse point cloud model can also get after the iterative method of bundle adjustment.We also research the reconstruction of dense point cloud models by the PMVS method on the basis of our previous work.Finally,we show the results of our work.The results of 3D model reconstruction prove the feasibility of the scheme we proposed in this work.