Research Of Image3D Reconstruction Based On Monocular Vision And Binocular Vision

Image3D reconstruction is as an important branch of computer vision technology, and it has been widely applied in medical image processing, aerospace, military reconnaissance and virtual reality and other fields. Although the existing3D model reconstruct instrument is gradually complete, how to reconstruct more accurate or more complex3D model is still a relatively tedious work.3D model reconstruction based on stereo vision method is a hot research pot in the present, and it also attracted vast people’s attention.The main tasks as follows:1) Monocular vision method:In view of the traditional SFS minimization method use the fixed smoothing factor in the process of solving the energy equation, which makes the reconstruction of3d model is inaccurate, thus, the inconsistent smoothing factor is used to construct a new energy equation to obtain more accurate3D space’s normal vector information. On the basis, work out a corresponding Poisson Equation to get the adjacent pixel’s relative height information. Simulation experiment results show that the application of improved SFS minimization method is able to reconstruct a more complete and more accurate3D model.2) Binocular vision method:carry on research about the most important part of binocular vision method:scale invariance transform (SIFT) algorithm that used in feature points matching process. This algorithm can detect the feature points of image when the image rotates and its scale changes. At the same time, the algorithm possesses good robustness when image illumination changes and noise pollution exists and so on. The dimension of characteristic describe operator constructed in the traditional SIFT algorithm is too high, which will affect the matching efficiency of the algorithm. Walsh-Hadamard kernel projection technology is introduced to construct characteristic describe operator. The improved SIFT algorithm can effectively reduce dimension when construct characteristic describe operator. The simulation experimental results show that the improved SIFT algorithm can improve the efficiency of feature points’matching a certain extent. Then apply the improved SIFT algorithm to image3D reconstruction process, we can obtain a more real and ideal model.