A three-dimensional fragment splicing and reconstruction method based on fracture surface shape characteristics

Abstract

The present invention provides a three-dimensional fragment splicing and reassembling method based on fracture surface shape features, comprising the following steps: extracting the three-dimensional contour of the fragment fracture surface according to the feature combination of the fragment edge; performing fracture surface matching according to the multi-scale shape features of the three-dimensional contour of the fragment fracture surface to obtain the pairwise matching results between the fragments, screening the unmatched fragments, obtaining four different combination features of the fragment edges of the unmatched fragments, performing abnormal processing on the fragments that have not completed the matching: optimizing the matching matrix to obtain the reassembled fragments. This method performs feature matching according to the fracture surface contour of the fragment, can directly reassemble the three-dimensional fragments, does not need to pre-process the fragment data, is not only applicable to a single object (plate), but also can splice and reassemble mixed fragments of multiple objects (plates, bowls, etc.), achieving the technical effect of high reassembly accuracy and short running time.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional fragment splicing, and in particular to a three-dimensional fragment splicing and reassembly method based on fracture surface shape features. Background Art

[0002] As the use of 3D scanning equipment in field archaeology becomes more and more widespread, the problem of fully automatic splicing and reassembling of broken 3D objects has received more and more attention. When the number of tool fragments is large and multiple tool fragments are mixed, manual splicing and reassembling is a heavy and monotonous task. Similarly, for a large number of extremely precious cultural relics, the lack of manual participation in restoration and restoration personnel is also a very serious problem. Therefore, computer-assisted splicing and reassembling of tool fragments is very important for the restoration and reassembly of cultural relics. Most of the existing technologies only splice and reassemble fragments of the same tool. Summary of the invention

[0003] In view of the problems existing in the prior art, the present invention discloses a three-dimensional fragment splicing and reassembly method based on fracture surface shape characteristics, which specifically adopts the following steps:

[0004] Extract the three-dimensional contour of the fragment fracture surface according to the characteristic combination of the fragment edge;

[0005] According to the multi-scale shape characteristics of the three-dimensional contours of the fracture surfaces of the fragments, the fracture surfaces are matched to obtain the pairwise matching results between the fragments;

[0006] Screen the unmatched fragments and obtain four different combination features of the fragment edges of the unmatched fragments, namely LLE four-dimensional feature, PCA two-dimensional feature, mixed feature and LLE two-dimensional feature. Use the four different combination features to repeat the above two steps to perform fracture surface matching to obtain the initial fragment matching result;

[0007] Exception processing is performed on the fragments that have not been matched: for the fragments that have not been matched yet, the number of clusters of the fragment edges is increased to obtain a larger number of fracture surface contours, and the multiple fracture surface contours are matched with the unmarked fracture surfaces of other fragments to output a matching matrix;

[0008] Optimize the matching matrix: determine whether the number of each group of fragments is less than the set threshold. If so, perform contour matching on this group of fragments and other groups of fragments, and merge the two groups of fragments that meet the criteria. For the same group of fragments, adjust the position of the current fragment according to other fragments around the fragment until the distance between the fragments is less than the set threshold, thereby obtaining the optimized matching matrix. Splice and reassemble the fragments according to the optimized matching matrix.

[0009] The present invention discloses a three-dimensional fragment splicing and reassembling method based on fracture surface shape features, which can directly reassemble three-dimensional fragments without pre-processing the fragment data. It is not only applicable to a single object (plate), but also can splice and reassemble mixed fragments of multiple objects (plates, bowls, etc.). The method has accurate reassembly and short running time. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 A simple flow chart of the method of the present invention;

[0012] Figure 2 A schematic diagram of the extraction of the fracture surface of the fragments in the present invention;

[0013] Figure 3 It is a schematic diagram of the splicing and reassembly of fragments in the present invention;

[0014] Figure 4 Schematic diagram of the linear geometric features of the midpoint of the present invention. DETAILED DESCRIPTION

[0015] In order to make the technical solutions and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention:

[0016] like Figure 1 A three-dimensional fragment splicing and reassembly method based on fracture surface shape features is shown, which specifically adopts the following steps:

[0017] S1: Extracting the three-dimensional contour of the fracture surface of the fragment according to the feature combination of the fragment edge. The specific method is as follows: obtain the edge point set of the fragment according to the curvature value of the midpoint of the three-dimensional fragment, use the Kmeans method to pre-cluster the edge points of the fragment to obtain a suitable number of clusters, extract the features of the edge points of the fragment, and perform Bayesian Gaussian mixed clustering according to the edge features of the fragment and the number of clusters to obtain the three-dimensional contour of the fracture surface of the fragment. This step uses the clustering method to extract the three-dimensional contour (such as Figure 2 The specific steps are as follows:

[0018] S11: Curvature extraction. For each fragment, the maximum curvature of each point in the fragment model is calculated, and points with maximum curvature values ​​greater than a threshold C are extracted to form a point set S, most of which are edge points of the fragment.

[0019] S12: Clustering to obtain the outline of the fracture surface. Clustering is performed on the extracted fragment point set S to separate the cross-sectional outline of the fragments. The specific steps are:

[0020] S121: Determine the number of clusters. First, use Kmeans clustering to pre-cluster the point set S according to the three-dimensional coordinates of the points. Based on our observation of the fragments, we concluded that the number of clusters of the point set S is generally 3-7. Therefore, we perform Kmeans clustering on each number of clusters, and accumulate the squared distance between each point and its cluster center to obtain the sum of squared differences [d1, d2, d3, d4, d5], and further obtain the difference between adjacent sums of squares [cd1, cd2, cd3, cd4] (where cd1 = d1-d2, and so on). Compare the sum of squared differences under each number of clusters and find the sum of squared differences d i Below the threshold d and cd i / cd i+1 The number of clusters <= 2.5 is taken as the final number of clusters.

[0021] S122: Use one of the four feature combinations of fragment edge points.

[0022] S123: Bayesian Gaussian mixture clustering. Perform Bayesian Gaussian mixture clustering on the point set S according to the number of clusters and the features of the points obtained in the previous step. After obtaining the clustering results, count the number of points in each class. If a class includes most of the points, it means that there are clustered abnormal point groups in the point set, and the clustering result is not what we expect. At this time, we use the point features in the abnormal situation to re-cluster; if no abnormal situation occurs, we retain the original clustering result.

[0023] S2: Match the fracture surfaces according to the multi-scale shape features of the fracture surface contours of the fragments to obtain the pairwise matching results between the fragments. The specific method is as follows: pre-process the data of the three-dimensional contour points of the fracture surface to obtain a single-point, equidistant contour point cloud, select the pre-matched fracture surface set of the three-dimensional contour, extract the multi-scale linear features of the points in the contour point cloud, and match the fracture surfaces of different scale features to obtain different matching results. According to the method of the first matching taking an absolute advantage, the matching results that meet the standards are recorded, and the corresponding fracture surfaces and fragments are marked. This step uses the linear geometric features of the points to match the fracture surfaces of the fragments (such as Figure 3 The specific steps are as follows:

[0024] S21: Data preprocessing. Sample the contour point cloud obtained from the segmentation of the fragment cross section to obtain a single-point, equidistant (approximate) contour similar to a pearl necklace. For each point, the density of the point is calculated with the point as the center and 10.0 as the radius, and the density values ​​of all points are counted. Generally, the point with the highest frequency of density value in the point set is selected as the starting point, and the next point to be added is searched along the left or right direction, with a spacing greater than 0.5, and in the direction with the smallest angle with the previous added point (the points that have passed the path will be marked), until the distance between the last added point and the starting point is less than or equal to 1.5.

[0025] There are two cases of fracture surface matching: one fracture surface completely matches another fracture surface (referred to as "one-to-one matching method") and one fracture surface matches multiple fracture surfaces (referred to as "one-to-many matching method").

[0026] S22: Extract multi-scale features of the points, and perform a one-to-one matching method on the fracture surface for the features of each scale. The specific method of the one-to-one matching method is as follows:

[0027] S221: Use the bounding box method to find the contour f i Pre-matching set.

[0028] S222: Find the cross-sectional profile f i The characteristic corner point set in f i Each corner point h in the feature corner point set of f q Find the best corresponding point in the feature corner point set and get f i With f q The method for calculating the best corresponding point is to calculate the large (medium, small) scale features of the corner point h and f q The Langmuir distance of the large (medium and small) scale features of all corner points in the image is calculated, and the point with the smallest distance is selected as the best corresponding point.

[0029] S223: f i With f q Perform RANSAC on the matching point pairs to eliminate mismatches, perform ICP precise matching based on the rotation and translation matrix obtained by RANSAC to obtain the rotation and translation matrix bestT and record it, and record the f after bestT i With f q The number num of contour points whose corresponding point distance is less than 0.5.

[0030] S224: Cross-section profile f i Each section in the pre-matching set is calculated by comparing the section with the section profile f iThe rotation matrix bestT and the number of matching point pairs num are sorted from large to small. If the first number of matching point pairs num meets the following setting requirements: the first number of matching point pairs num>A and the first number of matching point pairs num>the second number of matching point pairs num*B, where B can be 1.5, then the corresponding section and fragment are recorded. (The size of M is related to the number of section points).

[0031] S23: Repeat step S22 using features of three scales.

[0032] S24: "One-to-many" matching method in fracture surface, the specific steps are as follows:

[0033] S241: Select a part of the cross sections as the "many" set SmallF, and the rest of the cross sections are classified as the "one" cross section set BigF. The selection rules are:

[0034] or

[0035] Among them, l, w, h are the length, width and height of the cross-section bounding box, and m is the number of cross-sections.

[0036] S242: For the section sf in SmallF i , calculate the pre-matching set according to the characteristic method.

[0037] S243: Find the cross-sectional profile f i The feature corner point set in , and calculate the 1-neighbor feature set and the 2-neighbor feature set.

[0038] S244: Use the medium (small) scale features of the characteristic corner points of the cross section to calculate sf i A set of matching point pairs of characteristic corner points between each section in its pre-matching set is obtained, and further matching pairs that meet the standards are obtained. The calculation method is consistent with the previous "one-to-one matching method" (i.e., sub-steps S223-S234 in the one-to-one case under S22).

[0039] S25: Repeat step S24 using the mesoscale and microscale features.

[0040] S3: Use different feature combinations of the fragment edge (four in total) in turn, repeat the above two steps, and obtain the initial fragment matching result. The specific method is as follows: the four feature combinations of the fragment edge are LLE four-dimensional feature, PCA two-dimensional feature, mixed feature, and LLE two-dimensional feature. Use the new fragment edge feature to extract the new fracture surface contour of the fragments that were not marked in the previous iteration, and perform multi-scale matching of the fracture surface contour on the new contour and the old contour that was not marked in the previous iteration. The initial fragment matching result is obtained by combining the four matches, and the fragments that have a matching relationship with each other are put into the same group to obtain multiple groups of fragments that do not intersect with each other. The specific method is as follows:

[0041] S31: Select a pair of fragments with the most matching points from the matching of the same group of fragments, where the fixed fragment is regarded as the first fragment added to the group, with level 0, and add the fragments that have a direct matching relationship with the fragment, with level 1, and the fragments that have a direct matching relationship with level 1 are recorded as level 2, and so on, until all the fragments in the group are added.

[0042] S4: Exception processing is performed on the fragments that have not been matched yet, and the final matching results are obtained in the following specific manner: (1) For the fragments that have not been matched yet, the number of clusters of the fragment edges is increased to obtain a smaller fracture surface contour, which is matched with the unmarked fracture surfaces of other fragments; (2) The contour of the entire fragment is matched with the existing multiple groups of fragment contours. The matching relationship that meets the standard is recorded, and the corresponding cross section and fragment are marked. The specific method is:

[0043] S41: In step S121, we assume that the number of clusters of the point set S is generally 3-7, which limits the fragments to be divided into at least 3 contours and at most 7 contours. For unmatched fragments, we can increase the number of contours, subdivide the contours (for example, divide them into 5-10 categories, etc.), and perform multi-scale matching between the subdivided contours and the unmarked cross-section contours in the above steps. Record the matching relationships that meet the standards and mark the corresponding cross-sections and fragments.

[0044] S42: The multiple groups of fragments obtained in step S3 that do not overlap with each other are recorded as g1, g2...g i Wait, g i The fragment contour point sets in the are spliced ​​together to form a new contour point set according to the obtained matching relationship. For each point in the new point set, if there are no points from other fragments around it (that is, there is no matching fragment in the area), then the point is saved, and the unmatched contour area obtained by calculating all points is recorded as the fragment contour of the group of fragments.

[0045] S43: Compare the overall edge contour of the fragments that have not been matched in S41 with g iThe fragment contours are matched at multiple scales, the matching relationships that meet the criteria are recorded, and the corresponding fragments and fragment groups are marked.

[0046] S44: Fragments that do not meet the criteria are recorded in unmatched fragments.

[0047] S5: Optimizing the matching matrix between fragments is specifically carried out in the following manner: if the number of fragments in a group is less than 4, perform contour matching of this group with other groups, and merge the two groups that meet the criteria; for fragments in the same group, adjust the position of the current fragment according to other fragments around the fragment, iterate multiple times, and obtain an optimized matching matrix. The specific method is:

[0048] S51: For the same group of fragments, calculate the size C of the largest gap between the fragments, and then start from the last added fragment, adjust the position of the fragment according to the point set whose distance from the fragment contour does not exceed C, replace the original fragment position with the adjusted fragment, and adjust the previous added fragment in turn until the first added fragment is reached, completing one iteration.

[0049] S52: Iterate K times to obtain the optimized result (K is between 5 and 10).

[0050] The present invention provides a three-dimensional fragment splicing and reassembling method based on fracture surface shape characteristics, which can realize the splicing and reassembly of three-dimensional fragments belonging to multiple tools; compared with the prior art, the accuracy of the method is greatly improved.

[0051] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A three-dimensional fragment splicing and reassembly method based on fracture surface shape characteristics, characterized in that The following steps are involved: The three-dimensional contour of the fragment fracture surface is extracted according to the characteristic combination of the fragment edge. The specific method is as follows: the edge point set of the fragment is obtained according to the curvature value of the midpoint of the three-dimensional fragment, the edge points of the fragment are pre-clustered by the Kmeans method to obtain the required number of clusters, the features of the edge points of the fragment are extracted, and the three-dimensional contour of the fragment fracture surface is obtained by performing Bayesian-Gaussian mixed clustering according to the edge features of the fragment and the number of clusters. According to the multi-scale shape characteristics of the three-dimensional contours of the fracture surfaces of the fragments, the fracture surfaces are matched to obtain the pairwise matching results between the fragments; Screen the unmatched fragments and obtain four different combination features of the fragment edges of the unmatched fragments, which are LLE four-dimensional features, PCA two-dimensional features, mixed features and LLE two-dimensional features. Use the four different combination features to repeat the above two steps to perform fracture surface matching to obtain the initial fragment matching results; Exception processing is performed on the fragments that have not been matched: for the fragments that have not been matched yet, the number of clusters of the fragment edges is increased to obtain a larger number of fracture surface contours, and the multiple fracture surface contours are matched with the fracture surfaces of other fragments to output a matching matrix; Optimize the matching matrix: In the initial fragment matching results, determine whether the number of matching fragments in each group is less than the set threshold. If so, perform contour matching of this group of fragments with other groups of fragments, and merge the two groups of fragments that meet the criteria. For the matched fragments, adjust the position of the current fragment according to other fragments around the fragment until the distance between the fragments in this group is less than the set threshold, thereby obtaining the optimized matching matrix. Splice and reassemble the fragments according to the optimized matching matrix. The LLE four-dimensional feature is: the three-dimensional coordinates of the point are subjected to LLE dimensionality reduction to obtain the first two-dimensional features (X lle ,Y lle ), the original three-dimensional coordinates are reduced to two-dimensional coordinates by PCA, and for each point, the R points closest to the point are found according to the two-dimensional coordinates to form a point set for least squares fitting to obtain the estimated slope m and intercept b of the point. lle ,Y lle ) is concatenated with (m, b) to obtain the four-dimensional features of the point; The PCA two-dimensional feature is: the three-dimensional coordinates of the point are subjected to PCA dimensionality reduction to obtain the two-dimensional feature of the point; The hybrid feature is a combination of PCA two-dimensional feature and LLE four-dimensional feature; LLE two-dimensional features: The three-dimensional coordinates of the point are reduced by LLE to obtain the two-dimensional features of the point.

2. The method according to claim 1, characterized in that: The three-dimensional contour of the fracture surface is preprocessed to obtain a single-point, equidistant contour point cloud. The pre-matched section set of the three-dimensional contour is selected, and the multi-scale linear features of the points in the contour point cloud are extracted. Different matching results are obtained by matching fracture surfaces with different scale features. The erroneous matching results are deleted according to the method that the first match occupies an absolute advantage, and the matching results that meet the standards are recorded. The corresponding fracture surfaces and fragments are marked to obtain the pairwise matching results between the fragments.

3. The method according to claim 2, characterized in that: The first matching method that occupies an absolute advantage is: i Each section in the pre-matching set is calculated by comparing the section with the section profile f i The rotation matrix bestT and the number of matching point pairs num are sorted from large to small. If the first number of matching point pairs num meets the following setting requirements: the first number of matching point pairs num>A and the first number of matching point pairs num>the second number of matching point pairs num*B, the corresponding sections and fragments are recorded.

4. The method according to claim 1, characterized in that: When four different combinations of features are used for fracture surface matching: the new fracture surface contour is obtained by using the new fragment edge feature extraction for the fragments that were not marked in the previous iteration, the new fracture surface contour and the old fracture surface contour that were not marked in the previous iteration are subjected to multi-scale matching of the fracture surface contour, and the initial fragment matching result is obtained by combining the four matchings. The fragments that have a matching relationship with each other are placed in the same group to obtain multiple groups of fragments that do not overlap with each other.

5. The method according to claim 4, characterized in that: The abnormal processing of the fragments that have not been matched also includes: matching the contour of the entire fragment with the existing multiple groups of fragment contours, recording the matching relationship that meets the standard, and marking the corresponding cross-section and fragment.

6. The method according to any one of claims 1 to 5, further characterized in that: Characteristics of the fracture surface profile include: For a point p in the contour, take k neighboring points (l1,l2,l3,.......,l k-1 ,l k ) and (r1,r2,r3,........,r k-1 ,r k ) The distance between each point is set to M, and the relationship between the 2k neighboring points and the point p is constructed as follows: For the k points on the left, calculate the Euclidean distance between these k points and point p to obtain the k-dimensional features |l1p|,|l2p|...|l k p|; Calculate the cosine of vectors pl1 and l1l2, the cosine of l1l2 and l2l3, ..., l k-2 l k-1 and l k-1 l k The cosine value of the angle between three adjacent points is obtained by the cosine of the angle between the three adjacent points to obtain the k-1 dimensional features cos(∠pl1l2), cos(∠l1l2l3)...cos(∠l k-2 l k- 1l k ); Similarly, the k points on the right are calculated to obtain another 2k-1 dimensional features; Calculate l1 and r1, l2 and r2, l k and r k The lateral distance between k-dimensional features |l1r1|,|l2r2|...|l k r k |; Finally, add cos(∠l1pr1) and cos(∠l k pr k ) That is, a total of 5k dimensional features are obtained.

7. The method according to claim 2, further characterized in that: The pre-matching section set is selected in the following manner: when one section matches with another section, a bounding box method is used; when one section matches with multiple sections, a feature method is used; The bounding box method is as follows: calculate each cross-sectional contour f i The bounding box of the object is obtained by obtaining the length, width and height (l, w, h) and calculating the cross-sectional contour f i The L2 distance between the (l,w,h) vector of the cross-section contour and the (l,w,h) vector of other cross-section contours, and sort them in ascending order, starting from the distance f i The smallest cross section f q Start, if f q Two or more of the length, width, and height of i The difference is less than or equal to P, then f i With f q Similar in shape, f q Add to f i The number of sections added to the pre-matching set is N; The characteristic method is as follows: the Hausdorff distance between the first adjacent feature set of the section and the feature corner point set of other sections is calculated, and they are sorted in order from small to large, and the N sections with the smallest distance are selected; similarly, the Hausdorff distance between the second adjacent feature sets is calculated, and the N sections with the smallest distance are selected, and these 2N sections form a pre-matching set of the section; The characteristic corner point is defined as: for a point p in the contour, select the point where ∠l1pl2 is between 20° and 160° as the characteristic corner point, and all the characteristic corner points constitute the cross-sectional contour f i The feature corner point set of The first neighboring feature is (∠l1pl2,|pl1|+|pr1|); The second adjacent feature is (∠l1pl2,|l1r1|,|pl2|+|pr2|,-cos(∠l1l2l3)-cos(∠r1r2r3)); The first or second neighboring features of all feature corner points constitute the first or second neighboring feature set of the cross section.

Images