Research On Binocular 3D Reconstruction Based On Markov Random Fields

Three-dimensional reconstruction is a branch of computer vision,collection information of the scene,through two or more cameras.Because of the difference in the spatial position during the scene acquisition process of the camera,it is convenient to generate a parallax with depth information using the difference geometric information combined with the collected two-dimensional image information,and determine the position of the spatial point in the scene,finally carry three-dimensional reconstruction.It has the characteristics of simple equipment,low cost,easy realization and so on.It is widely used in agricultural,military,medical,unmanned Aerial vehicles,archaeology,virtual reality and so on and it has high scientific research value.Stereo matching is establishing the corresponding relationship between the projection points of two or more images in the spatial scene,which is the focus of three-dimensional reconstruction.It is generally divided into local stereo matching and global stereo matching according to the matching way.Local stereo matching only rely on one or more pixels around it,only consider the correlation with the neighborhood pixels,find the corresponding pixel matching,matching difficulty and time consumption are both relatively low.Although the effect of matching accuracy of global matching is good,especially obvious for low texture area,occlusion area and texture-free area,but compared with the local matching algorithm,matching complexity and time costs are greatly increased.The construction of the global energy function is the most important step in the global stereo matching,the calculation of the function involves all the pixels in the image,solving the problem of minimizing the global function need many optimization algorithms to improve the matching precision.In this paper,the binocular image is three-dimensional reconstructed,the global matching is applied,through the theoretical knowledge of markov random field,the corresponding model of two pixel points is established,and a data item of globalstereo matching function is obtained,for the establishment of smoothing terms,this paper uses the four neighborhoods of the disparity graph as constraints,and finally obtains a global matching function.This is a critical step in this article,because the matching function directly affects the final result.After the matching of the energy function is established,the next step is how to optimize the problem.In this paper,it is optimized through the use of simulated annealing algorithm,solution space selection has become a focus of this paper,through continuous iteration,the resulting parallax graph optimized,and ultimately get the final disparity map.Finally,the three-dimensional reconstruction of the obtained parallax is obtained,and a three-dimensional three-dimensional image is obtained.