Research On Fracture Surface Matching Method For Terracotta Warriors Based On Multi-Feature

When the Terracotta Warriors in Qinshihuang Mausoleum was unearthed,most of them were broken into blocks.The number of block is large,the structure and shape are different,and most of them have defects,so it needs a long and difficult work to restore them by artificial method.Therefore,it is a research hotspot to obtain 3D data models of blocks through 3D scanning equipment and assist the virtual splicing of Terracotta Warriors blocks with the help of computer technology.Because the existing blocks matching methods require high integrity of geometric features,and have the disadvantages of single feature,large error and low efficiency,the matching effect of the Terracotta Warriors blocks is not good.In view of the three-dimensional data model of Terracotta Warriors block,block classification and fracture surface extraction are completed firstly,then the blocks matching based on multi-feature on the fracture surface is researched.The main works and contributes of the dissertation are concluded as follows:(1)A block classification and fracture surface extraction method is proposedBlock classification is implemented by a classification algorithm based on salient region and two-dimensional shape feature.Firstly,the salient region feature of block model is extracted and matched by earth mover’s distance(EMD)algorithm to achieve rough classification of blocks.Secondly,the two-dimensional image information of blocks whose surface feature is not obvious is obtained and their shape feature is extracted,and then support vector machine(SVM)is used to further classify them,thus the final accurate classification of blocks is achieved.Fracture surface extraction is realized by a surface segmentation algorithm based on the segmentation line.The algorithm divides the external surface of a block into several surfaces by extracting the ridge feature segmentation lines,and identifies the fracture surfaces according to the roughness of the surface.The experimental results show that the classification method can accurately classify the blocks of Terracotta Warriors.The fracture surface extraction method can effectively extract the fracture surfaces under the condition that the fracture surface is rougher than the original surface,which provides guidance and constraints for the subsequent blocks matching and splicing work.(2)A fracture surface matching method based on ISS feature point is proposedFirstly,the intrinsic shape signature(ISS)feature points with neighborhood radius constraints on fracture surface are extracted,the feature sequences of feature points are calculated,and the fracture surfaces are roughly matched through the matching of feature sequences.Then a scale iterative closest point(ICP)algorithm is used to further match the feature point sets,and the accurate matching of fracture surfaces is completed.The experimental results show that the ISS feature points algorithm with neighborhood radius constraints can achieve much higher accuracy in the feature points extraction than traditional ISS algorithm,solve the problems of long time-consuming of feature extraction and slow matching speed in global matching,and match the fracture surfaces of blocks more effectively.(3)A fracture surface matching method based on feature region and multi-parameter ICP is proposedThe concave and convex feature regions on fracture surface are calculated firstly,and the coarse matching of blocks is achieved through calculating the similarity of the feature regions.Then the parameters of Gaussian probability model,angle constraint and dynamic iteration coefficient are added into ICP algorithm to improve the performance of it,and the multi-parameter ICP algorithm is used to further match the fracture surfaces.The experimental results show that the proposed matching method can effectively restrain the influence of noise points on matching results,solve the failure matching caused by the variety of rotation angles,and have better convergence and higher matching speed than some existing matching methods.(4)A fracture surface matching method based on 2D image feature is proposedFirstly,a point cloud registration algorithm based on region level is used to roughly match the fractured surfaces by the steps of region partition and region registration.Then the 3D point cloud of fracture surface is converted into a 2D image,the speeded up robust features(SURF)algorithm is used to extract the 2D image features,and the pixel pairs matching is solved.Finally,3D corresponding points are gotten according to the 2D matching pixels,and the rigid transformation is solved,thus the fracture surfaces are matched accurately.The experimental results show that the proposed fracture surface matching method can decompose global matching of large scale point cloud into a certain number of small-scale regions matching problem through regional division and image conversion,and improve the role of overlapping areas in fracture surface matching to complete the matching of low overlapping fracture surfaces.(5)A fracture surface matching method based on multi-feature fusion is proposedThe contour curves of fracture surface are extracted,and the coarse matching is completed through the matching of contour curves firstly.Then a variety of geometric features such as curvature,normal line and point cloud density are extracted,and the matching method based on multi-feature fusion is used to achieve accurate matching of fracture surfaces.The experimental results show that the matching method can fuse multiple features to complete much more accurate and rapid matching of fracture surfaces by analyzing the effect of curvature,normal line and point cloud density on matching results,and overcome the problem of low matching accuracy of matching methods based on single feature.The research is supported by NSFC(61731015,61672013).The main methods have been applied to the virtual restoration of Terracotta Warriors in Qinshihuang Mausoleum,and good results have been achieved.