The Research And Implementation Of 3D Reconstruction From The Large Scene Based On Binocular Vision

With the improvement of people’s cognitive level and the pursuit of material things and culture,more and more 3D technology-related products began to spread in the crowd.In order to obtain a better using experience,the virtual reality 3D reconstruction has finally returned back to the title aspirations again,and then got more and more attention in recent years.This paper studied and completely realized a system for the three-dimensional reconstruction from the large scenes,including stereo calibration,stereo rectification,stereo matching and 3D point cloud reconstruction of the scene.In the stage of stereo matching stage,the original NLCA algorithm is further improved with its single considering factors so that it can be applied to the reconstruction of the large scenes,which is always with multi-scale,rich objects in it and even the frequent changes in depth.In the first,we improved the initial matching cost function on the basis of the original algorithm,we combined the cost computed by the ADCensus and the gradient of the vertex to strengthen the edge feature of the pixels.We also defined a new weight function that contained both the weight components of the edges and the colors in the reference images to assign the value for the edges between the adjacent pixels.In addition,according to the reasons that causes the edges in an image,we used the SLIC superpixel segmentation to give the edge function a twice constraint while detecting the edges with the sobel operator.In this way,the influence of the edge pixels that are not caused by the changes of depth to the construction process of the MST has been greatly reduced.In the stage of disparity refinement,we proposed a disparity optimization method that is based on the super block and in which we used the high similarity of the pixels in a same superpixel block to repair the disparity data so as to increase percentage of the stable points,after combining the same refinement method that based on the MST aggregation as the original algorithm,we got a higher quality of the optimization.In the last,for improving the integrity of the 3D reconstruction,we applied a series of disparity optimization processes,including the superpixels based method introduced before,an re-populating method for the regions that cannot be optimized by the method and an improved median filter algorithm for those unstable pixels,this part of work effectively reduced the discontinuous blocks and holes on the disparity image,and solved the problem that the disparity image is inconsistent with the actual information which the reference image had displayed of the scene,and increased the accuracy and completeness of the 3D reconstruction process.The experimental data shows that the algorithm in this article has obvious advantages compared with the original algorithm,not only performed a better rusult in the low-resolution image but also for the high-resolution large scene image materials used in this research,proving that the system we constructed for the large-scenes reconstruction is reliability,accuracy and stability.