Research On High Resolution Binocular Stereo Vision Technology Based On Multi-vision Fusion

In recent years,the 3D reconstruction method based on binocular stereo vision has been more and more widely used due to its advantages of fast and simple.While using the 3D model geometric information of the object,the requirements for the texture details and realism of the 3D model are becoming higher and higher.Therefore,how to obtain a 3D model with high-resolution texture has become a research hotspot of scholars.In this paper,on the basis of the binocular stereo vision3 D reconstruction,a texture camera dedicated to collect high-resolution images is added.The mapping method of high-resolution texture images and binocular point cloud model is studied based on feature matching and multi-view fusion technology to realize the reconstruction of high-resolution texture 3D model.The main work and innovations are as follows:1.This paper combines 2D image feature matching with binocular stereo vision3 D reconstruction,and proposes a new high-resolution texture mapping method by using a high-resolution 3D reconstruction system combined with binocular stereo vision system and long focal length texture camera.Using the binocular stereo vision left-view image as a bridge,matching the high-resolution texture image with the 2D feature of the binocular left image,combining with the match relationship between the binocular left image and the 3D point cloud to establish the mapping registration relationship between the texture image and the point cloud model.The mapping of high-resolution texture images to point cloud model is realized.The experiments on different 3D objects verify the feasibility and effectiveness of the method.2.The high-resolution 3D reconstruction of the complete surface is achieved through the mapping and fusion of multiple texture images with different perspectives and 3D model.The guide line point cloud data partition method is adopted to solve the problem of data redundancy of multi-view texture images overlapping regions through the partition mapping of point cloud data,and the effective fusion of multi-view texture images is realized.Four texture images experiment was used as an example.The experimental results show that the texture images are well aligned on the reconstructed model.This method effectively improves the texture resolution of the 3D model.3.Aiming at the problem of high-resolution 3D reconstruction of large scene targets,the mapping method of multi-site multi-view texture fusion is studied.First,the point cloud data of multiple sites are unified in the same coordinate system through point cloud stitching,and the relationship between the sites is obtained.Then,the high-resolution texture images of each site are matched with the binocular left image of each site,and the relationship between the sites is combined to obtain the mapping registration relationship between the texture image of each site and the stitched point cloud model,the high-resolution 3D reconstruction of the large scene target is realized.The effectiveness of the 3D reconstruction method of multi-site multi-view texture fusion was verified by the two-site outdoor bronze drum stitching and mapping experiment of 18 texture images.