Research On 3D Point Cloud Registration And Virtual Splicing Of Fragments

Terracotta warriors and horses,cultural relics blue and white porcelain have a long history.The excavated fragments of terracotta warriors and horses are numerous and complex in shape,while the number of blue and white porcelain fragments are small and the structure is simple.According to the different characteristics of these two research objects,the thesis conducts research on them from three-dimensional and two-dimensional perspectives respectively.Aiming at the problem of insufficient extraction of key feature information and loss of effective information in the three-dimensional registration stage,based on the existing deep network registration bottleneck,the research aims to achieve accurate registration between fragments.Aiming at the phenomenon of artifacts in the overlapping area after splicing of blue and white porcelain fragments,and content distortion in the non-overlapping area,innovation is made on the basis of the existing deep network stitching methods to achieve accurate image splicing.The major works are as illustrated:(1)In order to solve the problem of insufficient core information extraction in the stage of3 D correction feature extraction,a deep Gaussian mixture drop correction method based on dualism mechanism was proposed.Through the dualism mechanism,namely the spatialism mechanism and the channelism mechanism,the weight value of the input point cloud is automatically allocated in the feature extraction stage.The features are extracted.According to the implicit probability relationship between the point and the Gaussian mixture model,the corresponding relationship matrix is designed.In order to achieve the final accurate registration,the matching parameters are estimated through the two parameter unit calculation blocks to restore the optimal transformation.Experimental results show that the proposed method in this chapter has high registration accuracy and good robustness.(2)Aiming at the problem of insufficient key features extraction in the main distribution direction of data and time-consuming due to redundant and useless features in the 3D point cloud registration,a method based on PCA minimization of feature metric mapping was proposed.In the feature extraction stage of the main network Point Net,PCA is used to reduce the dimensional data space to low dimensional data space,the first K important features are selected,and then estimate the transformation matrix by minimizing the feature difference to obtain the final registration result.Experimental results show that the proposed method in this chapter has high registration efficiency and good robustness.And the method in this chapter shows good performance.(3)Aiming at the problem of artifacts in overlapping areas and content distortion in nonoverlapping areas after mosaic of blue and white porcelain fragments,a method of blue-andwhite porcelain fragments based on UNet++ to eliminate ghosting was proposed.The overall splicing process is divided into three parts.The first stage is used for global homography matrix estimation.The second stage obtains a coarse splicing result.According to the contour information of the coarse splicing result is a priori condition,the UNet++ network is used in the third stage.The content is corrected for the coarsely aligned images,and the final accurate stitching result is obtained.The experimental results show that the method proposed in this thesis not only eliminates the ghosting in the overlapping area of the splicing,but also improves the quality of image splicing to a certain extent on the basis of reducing the content distortion through evaluation indicators.(4)This thesis designs a splicing system for the design and realization of blue and white porcelain fragments.The system is mainly divided into five modules,namely data loading,image grayscale,rough image splicing,splicing outline display and image precise splicing modules.Each function module follows the norm of low coupling.By calling relevant methods,the image running results are displayed in the form of interface,and the accurate Mosaic of blue and white porcelain fragments from coarse to fine can be realized.It not only verifies the validity of the method in the previous chapter,but also promotes the development of virtual cultural relic splicing technology.