Research On Digitization And Reassembly Of 3D Cultural Relics Fragments With Defective Fractures

The virtual assembling technology of cultural relics fragments by digital means mainly involves the acquisition of three-dimensional measurement data,data preprocessing(denoising,simplification),feature point extraction of data and automatic assembly of fragments,etc.The acquisition and preprocessing of data is the basis of debris assembling,and the feature extraction of data is the key technologies of virtual merging.For this reason,this paper focuses on some specific problems involved in the splicing process of cultural relics fragments.The main work is as follows:(1)In order to improve the accuracy of camera calibration in the application of wide field of vision and improve the accuracy of three-dimensional measurement,a camera calibration method integrating multiple distortion factors is proposed.Firstly,By analyzing the influence of lens radial distortion,tangential distortion and thin prism distortion on the imaging system,an imaging model with multi-distortion factors is established.Then,the initial internal and external parameters of camera are calculated under the constraint of plane homography,and the distortion coefficients are calculated based on the assumption that the corrected image points conform to the principle of perspective projection.Finally,the exact values are gradually approached by iterative calculation under multi-distortion imaging model.The experimental results show that the accuracy of the proposed method is not less than that of Tsai's two-step method and Zhang's calibration method,and the coefficients of variation of the calibration results are lower under the interference of different intensity noise,which indicates that the proposed method has strong anti-noise performance.Finally,the experimental results show that the proposed method can effectively improve the measurement accuracy in binocular stereo vision measurement system.(2)Aiming at the problem of point cloud simplification with feature points preserved,a point cloud reduction algorithm based on dynamic partition of spatial grids is proposed.On the basis of the spatial grid octree partition,a dynamic partition method of spatial grid is designed,which can locate the feature and non-feature regions of the model without calculating the differential information of all points.Therefore,it is only necessary to extract and retain the feature points of the point cloud within the feature area,while the points in the non-feature area are reduced to different degrees according to the size of this grids,the model can be simplified in different degrees,so as to ensure that the model can be quickly simplified while retaining its detailed features.The application results show that the proposed method can maintain the fine features of the model and avoid the appearance of holes while simplifying the point cloud.(3)In order to improve the accuracy and stability of feature point extraction,a feature point extraction algorithm based on geometric structure of point space is proposed.Firstly,the proposed algorithm divides the model into feature regions and non-feature regions by the dynamic division method of spatial grid.Then the feature points only need to be extracted from the small grids,which can greatly reduce the amount of calculation.In addition,in order to solve the problem of feature point extraction,a new feature detection operator of point cloud is proposed.Firstly,the relationship between the detection operator and the normal vector is studied.On this basis,the condition function of feature point extraction is constructed.Aiming at the problem that the non-uniformity of point cloud may lead to the misjudgement of feature points,a Gaussian function of point's distance is introduced to correct the intercept ratio of straight lines,which can reduce the contribution weight of remote points to feature recognition.The experimental results show that the proposed method can quickly and accurately select out the feature points,and it has good anti-noise ability and stronger feature recognition ability.For point cloud models with different densities and different structures,the proposed algorithm can better identify the subtle feature points and take less time.(4)In order to solve the splicing of thin-walled fragments with partial edge defects,a segment matching method of contour curve based on derivative dynamic time warping is proposed,this method transforms the matching of spatial curves into the judgment of spatial distribution consistency of matched curve segments.Firstly,based on the theoretical knowledge of curvature and torsion of spatial curves,a discrete curvature and torsion estimation method is proposed,and the shape feature descriptors of 3d curves are constructed accordingly.Then the contour curve is segmented according to the corner points,furthermore,the dynamic time warping algorithm based on derivative is used to extract the matched curve segments.Finally,the function of triangular similarity and the constraint condition of normal consistency are constructed to eliminate the false matching curve segment and select the best matching curve.The experimental results show that the proposed method can solve the splicing problem of defect debris,and the splicing time is less and the error is smaller.(5)Aiming at the piecing of the thick wall fragments with defects in some regions,an automatic splicing algorithm of point cloud fragments based on ICP for complementary region pairs is proposed.Firstly,the flat function of the local region was constructed with integral invariants,and the fault surface was divided into different sub-regions(concave,convex and flat regions).Then,the shape feature descriptors of the sub-regions are constructed,and the similar sub-regions between the fracture surfaces are further searched to form complementary region pairs;Furthermore,the constraints conditions based on distance and principal direction and the consistency of spatial distribution are established,which can be used to eliminate pseudo-complementary region pairs and select the best matching fracture surface.Finally,the rough splicing of fragments is realized by quaternion algorithm,and the precise splicing of fragments is realized by improved ICP algorithm.The application results show that the proposed method can effectively solve the problem of splicing between thick-walled debris with defects in some regions,and the precision of sokucubg is high.