A three-dimensional model reconstruction and structure refinement enhancement method for cultural relic display

Abstract

The application provides a three-dimensional model reconstruction and structure refinement enhancement method for cultural relic display, comprising the following steps: in the state of supporting the cultural relic on a display stand, synchronously collecting multi-view images, support reaction force and environmental parameters, and establishing material coordinate anchor point time sequence data; determining a reference environmental state and a reference support state, separating camera motion, cultural relic overall rigid body motion and material deformation, and generating reference state surface data; implementing support load perturbation, forming load perturbation state correlation data set; accordingly, inverting local stiffness, material directionality, support contact range and initial stress distribution, and constructing a corrected display stand constraint mechanical model; and grading removing the display stand support load and the cultural relic dead load. The application improves the corresponding accuracy of data at different collection time and different angles by synchronously collecting multi-view images, support reaction force and environmental parameters, and establishing material coordinate anchor point time sequence data, and reduces point cloud mismatch and surface ghosting.

Description

Technical Field

[0001] This invention relates to the field of 3D modeling technology, and in particular to a method for 3D model reconstruction and structural refinement enhancement for the display of cultural relics. Background Technology

[0002] With the development of digital museums, virtual exhibitions, and digital preservation of cultural relics, establishing three-dimensional models of cultural relics using multi-view images, photogrammetry, or point cloud reconstruction techniques has become an important means of recording the form and digitally displaying cultural relics. Existing methods typically obtain three-dimensional models of cultural relics through feature matching, spatial intersection, point cloud fusion, and surface mesh construction, and combine image textures to enhance local structures such as inscriptions, cracks, and decorations.

[0003] However, during the collection process, cultural relics are usually under support, clamping, suspension, or partial support, and their surface morphology is simultaneously affected by the load of the support frame and the weight of the relic itself. For easily deformable cultural relics such as paper, silk fabrics, leather, and thin-walled wooden artifacts, the support position and distribution of support force can easily cause local bending, stretching, or warping. Existing methods often directly take the collected stress pattern as the relic's inherent morphology, making it difficult to obtain the relic's true morphology after eliminating the effects of the support frame and gravity.

[0004] Meanwhile, artifact scanning typically requires a certain amount of time, and changes in ambient temperature, humidity, and illumination during the acquisition process may cause slow expansion or contraction or deformation of the materials. The image acquisition device may also experience slight vibrations or thermal drift. Existing methods struggle to effectively separate camera motion, the overall rigid motion of the artifact, reversible environmental deformation, and time-varying deformation within the materials, easily leading to surface ghosting, structural boundary shifts, and local morphological distortions during multi-moment point cloud fusion.

[0005] Therefore, this invention proposes a method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display. The information disclosed in the background section is only for enhancing understanding of the background of this disclosure and may therefore contain prior art information that is not common knowledge to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for the reconstruction and structural refinement of three-dimensional models for the display of cultural relics, thereby solving the technical problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for reconstructing and refining the structure of a 3D model for displaying cultural relics includes the following steps: S1. While the cultural relic is supported by the display stand, collect multi-view images, support reaction force, environmental parameters and collection time, extract material feature points and establish material coordinate anchor point time series data; S2. Determine the reference environment state and reference support state based on the time series data of material coordinate anchor points, separate the camera motion, the overall rigid body motion of the cultural relic and the material deformation, map the surface data to the reference environment state, and generate the reference state surface data. S3. Generate a load perturbation sequence based on the reference support state. When the resultant force, resultant moment and spatial position of the artifact are within the allowable range, transfer the support load between adjacent support points, collect the support reaction force increment and material coordinate displacement field, and form a load perturbation state association dataset. S4. Based on the reference state surface data and the load perturbation state associated dataset, determine the local stiffness, material orientation, support contact range and initial stress distribution, and construct the display frame constraint mechanics model through strategy optimization; S5. Using the constraint mechanics model of the display stand, the support load of the display stand and the self-weight load of the cultural relic are removed in stages to obtain the intrinsic three-dimensional model of the cultural relic. Based on multi-view geometric evidence, the structure is enhanced, and the enhanced model is restored to the acquisition state in the positive direction. After the projection consistency verification and mechanical response consistency verification, a refined three-dimensional model is output.

[0008] S1 specifically includes: while the cultural relic is supported by the display stand, configuring image, reaction force, and environmental acquisition devices; aligning multi-view images, support reaction forces, and environmental parameters according to time markers after calibration to form a multi-source synchronous observation dataset and display stand support structure data; correcting multi-view images, extracting material feature points that move together with the cultural relic materials, establishing cross-view correspondences according to texture, adjacency, and motion continuity to form candidate material coordinate anchor points; calculating the initial three-dimensional position of the candidate material coordinate anchor point based on calibration parameters and pixel positions, and determining the material coordinate anchor point and its confidence level through ray angle and reprojection error screening, associating time markers, environmental parameters, and support reaction forces to form time-series data of material coordinate anchor points.

[0009] S2 specifically includes: determining the reference acquisition time, reference environmental state, and reference support state based on the time series data of material coordinate anchor points, and establishing material coordinate state compensation parameters; determining the camera motion components and the overall rigid body displacement of the artifact based on the fixed coding mark and the three-dimensional position of the material coordinate anchor points, and subtracting the reversible deformation displacement of the environment and the time-deformation displacement of the material to form fractional displacement data; performing dense matching on multi-view images under the same time marker, calculating the three-dimensional position of surface points, mapping the surface point cloud at each time to the reference acquisition time and reference environmental state based on the fractional displacement data, and performing confidence-weighted fusion on surface points that meet the multi-view support conditions to form reference state surface data.

[0010] S3 specifically includes: determining the combination of adjacent support points based on the reference support state, reference state surface data, and display stand support structure data; setting the load transfer-out amount, load transfer-in amount, and compensation support amount; generating a load perturbation sequence when the resultant support force, resultant support moment, and spatial position of the artifact are within the allowable range; adjusting the support load according to the load perturbation sequence; acquiring support reaction force and multi-view images after the state stabilizes; calculating the support reaction force increment; and forming a material coordinate displacement field after deducting non-load perturbation displacement; associating the support reaction force increment, material coordinate displacement field, environmental state, contact slip marker, and recovery state deviation; eliminating invalid and insufficiently repeatable states; and forming a load perturbation state association dataset.

[0011] S4 specifically includes: dividing the surface region based on the reference state surface data, the reference support state, and the associated dataset of the load perturbation state; determining the local stiffness, material orientation, support contact range, and initial stress distribution to form initial mechanical parameters; using the initial mechanical parameters as the starting values, adjusting the support contact range, local stiffness, material orientation, and initial stress distribution sequentially based on the predicted displacement error, the predicted support reaction force error, and parameter constraints to obtain optimized mechanical parameters; establishing a surface mesh based on the reference state surface data, assigning optimized mechanical parameters and setting display frame constraints; correcting the model using the correction state set; and independently verifying the model through the verification state set to obtain the corrected display frame constraint mechanical model.

[0012] S5 specifically includes: based on the corrected constraint mechanical model of the display stand, the support load of the display stand and the self-weight load of the cultural relic are removed in stages. Under the condition of preserving the inherent residual stress of the material, the intrinsic three-dimensional model of the cultural relic and the intrinsic deformation mapping relationship are obtained; the area to be enhanced is determined according to multi-view geometric evidence, and the intrinsic three-dimensional model of the cultural relic is subdivided into constrained meshes and the vertices are adjusted to form a candidate refined three-dimensional model and a record of structural enhancement evidence; the candidate refined three-dimensional model is positively restored to the corresponding acquisition state, and projection consistency verification and mechanical response consistency verification are performed using the verification perspective and the verification state set, respectively. Based on the verification results, the structural enhancement or mechanical parameters are adjusted or optimized, and the refined three-dimensional model is output.

[0013] The beneficial effects of this invention are as follows: This invention improves the accuracy of data correspondence between different acquisition times and perspectives by simultaneously acquiring multi-view images, support reactions, and environmental parameters, and establishing time-series data of material coordinate anchor points, thereby reducing point cloud mismatch and surface ghosting. By separating camera motion, the overall rigid body motion of the artifact, reversible environmental deformation, and time-varying deformation within the material, surface data is uniformly mapped to a reference environmental state, reducing the interference of environmental changes on the 3D reconstruction results of the artifact.

[0014] This invention achieves incremental support reaction force and material coordinate displacement field without disassembling the artifact by implementing micro-perturbation of the support load under controlled conditions of resultant support force, resultant support moment, and spatial position of the artifact, thus reducing disturbance to the artifact during the inspection process. By inverting local stiffness, material orientation, support contact range, and initial stress distribution, and correcting the constraint mechanics model of the display frame, the accuracy of characterizing the actual stress state of the artifact is improved, avoiding calculation deviations caused by using uniform empirical parameters.

[0015] This invention obtains an intrinsic 3D model of the artifact by gradually removing the support load of the display stand and the self-weight load of the artifact while preserving the inherent residual stress of the material. This reduces mesh flipping, structural distortion, and topological relationship destruction caused by nonlinear unloading. By constraining local structural enhancement with multi-view geometric evidence and employing forward state recovery, projection consistency verification, and mechanical response consistency verification, it avoids mistakenly enhancing ink stains, dirt, or shadows into 3D structures, thus improving the realism and reliability of the refined 3D model. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for reconstructing and refining the structure of a three-dimensional model for displaying cultural relics, according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown in the figure, this embodiment provides a method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display, including the following steps: S1. While the cultural relic is supported by the display stand, collect multi-view images, support reaction force, environmental parameters and collection time, extract material feature points and establish material coordinate anchor point time series data; S2. Determine the reference environment state and reference support state based on the time series data of material coordinate anchor points, separate the camera motion, the overall rigid body motion of the cultural relic and the material deformation, map the surface data to the reference environment state, and generate the reference state surface data. S3. Generate a load perturbation sequence based on the reference support state. When the resultant force, resultant moment and spatial position of the artifact are within the allowable range, transfer the support load between adjacent support points, collect the support reaction force increment and material coordinate displacement field, and form a load perturbation state association dataset. S4. Based on the reference state surface data and the load perturbation state associated dataset, determine the local stiffness, material orientation, support contact range and initial stress distribution, and construct the display frame constraint mechanics model through strategy optimization; S5. Using the constraint mechanics model of the display stand, the support load of the display stand and the self-weight load of the cultural relic are removed in stages to obtain the intrinsic three-dimensional model of the cultural relic. Based on multi-view geometric evidence, the structure is enhanced, and the enhanced model is restored to the acquisition state in the positive direction. After the projection consistency verification and mechanical response consistency verification, a refined three-dimensional model is output.

[0019] S1 specifically includes the following sub-steps: S110. Keep the cultural relic to be rebuilt in its original display stand support state. Without dismantling the cultural relic, changing the installation position of the display stand and the display posture of the cultural relic, arrange at least two image acquisition devices with overlapping fields of view on the outside of the cultural relic. Set up reaction force acquisition devices between each support point and the load-bearing structure of the display stand, and set up temperature acquisition devices, humidity acquisition devices and illuminance acquisition devices in the exhibition space where the cultural relic is located.

[0020] The effective field-of-view overlap ratio of adjacent image acquisition devices is not less than 60%, and the optical axis angle is 10° to 70°, so that the same surface position in the area to be reconstructed can appear in at least two effective viewing angles at the same time; the effective field-of-view overlap ratio refers to the ratio of the area of ​​the artifact surface projection area jointly observed by the two image acquisition devices to the smaller effective projection area of ​​the two.

[0021] Before data acquisition, the focal length, principal point coordinates, and lens distortion parameters of each image acquisition device are determined using a calibration board. The rotation matrix and translation vector of each image acquisition device relative to the exhibition space coordinate system are determined using fixed coded marks set on the periphery of the display stand that do not contact the artifacts. After calibration, the position of the image acquisition device is fixed, and the focal length, exposure time, gain, and white balance are locked.

[0022] Each reaction force acquisition device is zero-point calibrated under no cultural relic load, and the proportional coefficient is determined by a standard load. The unidirectional bearing support point acquires the support force along the support direction, while the multi-directional limiting support point acquires the support force in three orthogonal directions, and each support point is assigned a unique support point identifier. The position of each support point, the support direction, the connection relationship of the display stand structure, the adjustable load transmission path, and the adjustment direction of the micro-displacement actuator are recorded to form the display stand support structure data.

[0023] Using the image acquisition time as the main timeline, each image acquisition device is activated by a unified hardware trigger signal. The reaction force acquisition device and the environmental acquisition device use the same system clock to record timestamps. When the acquisition time of the support reaction force or environmental parameters does not coincide with the image acquisition time, linear interpolation is performed using adjacent measurement values ​​before and after the image acquisition time.

[0024] in, Indicates the time of image acquisition The corresponding alignment measurement value, where x represents any measurement data among support reaction force, ambient temperature, ambient humidity, or ambient illuminance. and They respectively represent the locations The preceding and subsequent adjacent measurement times.

[0025] The multi-view images corresponding to the same image acquisition time, the aligned support reaction force of each support point, the aligned ambient temperature, the aligned ambient humidity, and the aligned ambient illuminance are assigned the same time identifier, and together with the calibration parameters and the support structure data of the display stand, form a multi-source synchronous observation dataset. Taking an image acquisition frequency of 20 frames / s and a humidity acquisition frequency of 1Hz as an example, a single humidity measurement value should not be directly copied to 20 consecutive frames of images. Instead, the aligned ambient humidity at each image acquisition time should be calculated based on adjacent humidity measurements.

[0026] S120. Based on the calibration parameters, perform lens distortion correction on the multi-view images in the multi-source synchronous observation dataset, and use the fixed grayscale reference area outside the display stand as the brightness benchmark to unify the average brightness and brightness range of different images.

[0027] Material feature points are defined as local image feature locations located on the surface of the material of the cultural relic, which can move together with the material area and be repeatedly located in continuous multi-view images. Material feature points include fiber cross points, weaving nodes, crack endpoints, pigment particle edges, natural spots and surface texture inflection points, but do not include fixed coded marks, shadow boundaries, specular highlight centers, occlusion boundaries and isolated noise points.

[0028] A local image region is established centered on the candidate location. Image processing algorithms (preferably SIFT, SURF, or ORB feature extraction algorithms) are used to calculate the gray-level gradient and local texture descriptor. Candidate material feature points with local texture descriptors higher than the texture threshold are retained. The texture threshold is determined by repeatedly acquiring images of the indeformable surface region at least 20 times, based on the mean and standard deviation of the local texture descriptor. According to the calibration parameters of each image acquisition device, texture-similar locations are found within the epipolar search region of other effective viewpoints. Cross-viewpoint correspondences are established based on local texture similarity, adjacency consistency, and motion direction consistency within continuous time markers.

[0029] Candidate material feature points must be identified by at least two valid viewpoints under the same time marker, remain traceable within no less than five consecutive time markers, and have no exchange of adjacency order with neighboring candidate material feature points without reasonable occlusion reasons, and not be located in areas of high saturation, motion blur, display stand occlusion edges, or near image boundaries.

[0030] When multiple texture similarities exist for the same candidate material feature point in another effective viewpoint, and the score difference between the highest similarity result and the second highest similarity result is lower than the matching discrimination threshold, the candidate material feature point is deleted; the matching discrimination threshold is the intersection of the correct matching score distribution and the incorrect matching score distribution in the repeated acquisition experiment.

[0031] Material feature points that pass the above verification are identified as candidate material coordinate anchor points, and their time stamp, distortion-free pixel position, local texture description, and adjacent candidate material coordinate anchor point identifiers are saved. Highlight centers on glazed artifacts will shift with changing viewing angles; even if they are bright, they should not be used as candidate material coordinate anchor points. Crack endpoints located within the same material region and maintaining a stable adjacency relationship can be retained.

[0032] S130. Based on the distortion-free pixel position of each candidate material coordinate anchor point in each effective viewpoint, and the focal length, principal point coordinates, rotation matrix, and translation vector of the corresponding image acquisition device, construct a spatial observation ray pointing from the optical center of the image acquisition device to the candidate material coordinate anchor point. When the angle between the maximum observation rays is less than 10°, the depth calculation conditions are deemed insufficient, and no initial 3D position is generated; when the angle between the maximum observation rays is not less than 10°, the least squares intersection method is used to determine the position with the minimum sum of squared distances of each spatial observation ray, and this position is used as the candidate initial 3D position.

[0033] Reproject the candidate initial 3D positions onto each effective viewpoint and calculate the average reprojection error:

[0034] in, This represents the average reprojection error of the i-th candidate material coordinate anchor point. This indicates the number of effective viewpoints observed at the coordinate anchor points of the candidate material. This indicates the measured pixel position in the j-th effective viewpoint. Indicates the initial three-dimensional position of the candidate. Let the projection function of the j-th image acquisition device be denoted as . This represents the Euclidean norm.

[0035] The reprojection error threshold is set as twice the average reprojection error calibrated by the image acquisition device. ,when If the material coordinate anchor point is selected, the corresponding candidate material coordinate anchor point will be determined as the material coordinate anchor point; otherwise, it will be deleted. The confidence level of the material coordinate anchor point is calculated according to the following formula:

[0036] in, This represents the normalized reprojection error of the i-th material coordinate anchor point. This indicates the confidence level of the material coordinate anchor point.

[0037] A time-series record is established for each material coordinate anchor point, recording the material coordinate anchor point identifier, time identifier, pixel position in each effective viewpoint, initial 3D position, average reprojection error, 3D positioning error obtained by combining reprojection error with spatial depth conversion, material coordinate anchor point confidence, local texture description, adjacent material coordinate anchor point identifiers, alignment environment parameters, and alignment support reaction force saved according to support point identifiers. The time-series data of material coordinate anchor points is formed by arranging them according to the time identifier.

[0038] Material coordinate anchor point time series data is used in S210 to determine the reference acquisition time and reference support state; material coordinate anchor point confidence is used in S220 rigid body fitting and S230 confidence weighted fusion; and display stand support structure data is used in S310 to generate load perturbation sequence.

[0039] S2 specifically includes the following sub-steps: S210: Obtain the material coordinate anchor point time series data, multi-source synchronous observation dataset and calibration parameters generated in S130, and establish a stable time window according to the time identifier. Each stable time window contains at least 20 image acquisition times.

[0040] Calculate the difference between the maximum and minimum values ​​of the alignment ambient temperature, alignment ambient humidity, alignment ambient illuminance, and alignment support reaction force of each support point within the time window. When each difference is not greater than twice the repeatability error of the acquisition device, and the change in ambient temperature is not greater than 0.2°C, the change in ambient humidity is not greater than 1%, the change in ambient illuminance is not greater than 5 lx, and the change in alignment support reaction force of any support point is not greater than 0.5% of the average support reaction force of that support point, the time window is determined as a candidate stable time window.

[0041] When multiple candidate stabilization time windows exist, the candidate stabilization time window with the smallest average reprojection error of the material coordinate anchor points is selected. The middle acquisition time of this window is determined as the reference acquisition time. The median of the environmental parameters and the median of the aligned support reactions at each support point within this time window are respectively determined as the reference environmental state and the reference support state. The reference environmental state is used to unify the material state at different acquisition times, and the reference support state is the baseline stress state of each support point when no load perturbation is applied.

[0042] For each material coordinate anchor point, adjacent points located within the same continuous surface region are selected from its neighboring material coordinate anchor points. The confidence weighting center of these adjacent points is used as the origin of the local material coordinate system, and the first principal direction of the spatial distribution of adjacent points and its orthogonal directions are used as coordinate axes. The three-dimensional position of the material coordinate anchor point is then transformed to this coordinate system to eliminate the effects of overall translation and rotation. A continuous surface region refers to a surface adjacency path between material coordinate anchor points that does not cross cracks, pores, or material boundaries.

[0043] Using only data showing that the aligning support reaction forces at each support point are within the allowable range of the reference support state, a local material coordinate variation relationship is established:

[0044] in, This represents the change in local material coordinates of the i-th material coordinate anchor point at the k-th acquisition time relative to the reference acquisition time. , , and These represent the changes in ambient temperature, ambient humidity, ambient illuminance, and time interval, respectively. , , and These represent the temperature response coefficient, humidity response coefficient, illuminance response coefficient, and time drift coefficient, respectively. This represents the displacement residual.

[0045] Using the confidence level of the material coordinate anchor points as weights, each response coefficient is obtained by minimizing the sum of squared displacement residuals at the effective acquisition time. The environmental response lag time is tested one by one within the range of 0 to 1 / 3 of the total acquisition time, and the lag time with the smallest sum of squared displacement residuals is selected. The response coefficients and the environmental response lag time are combined to form the material coordinate state compensation parameters.

[0046] When the variation range of a certain environmental parameter is less than twice the repeatability error of the corresponding acquisition device, the median of the response coefficient corresponding to the effective material coordinate anchor points within the same continuous surface area is used. Taking an environmental humidity variation of only 0.1% as an example, the humidity response coefficient obtained by direct fitting mainly reflects measurement noise and should not be used for humidity compensation.

[0047] S220. Based on the pixel position of the fixed coding mark in the corresponding image of each time mark, recalculate the rotation matrix and translation vector of each image acquisition device relative to the exhibition space coordinate system, and determine the rotation and translation changes relative to the reference acquisition time as camera motion components to eliminate the apparent displacement caused by support micro-vibration and thermal drift.

[0048] For the 3D positions of the material coordinate anchor points after correction for camera motion components, rigid body fitting is performed based on the confidence level of the material coordinate anchor points:

[0049] in, and Let these represent the overall rotation matrix and overall translation vector of the artifact at the k-th acquisition time, respectively. This represents the three-dimensional position of the i-th material coordinate anchor point after camera motion component correction. Indicates the three-dimensional position at the reference acquisition time. This represents the confidence level of the material coordinate anchor points, where n represents the number of material coordinate anchor points involved in the rigid body fitting. This represents the combination of parameters that minimizes the objective function. This represents the Euclidean norm.

[0050] The material coordinate anchor points involved in rigid body fitting should be distributed across at least three non-collinear surface regions. For each material coordinate anchor point, its total displacement is successively reduced by the camera motion component and the component derived from the equation. , The overall rigid body displacement of the cultural relic is calculated, and then the reversible environmental deformation displacement and time-drift deformation displacement are deducted based on the material coordinate state compensation parameters to obtain the remaining local displacement.

[0051] Reversible environmental deformation displacement is displacement caused by changes in the environmental state relative to a reference environmental state and can be reversed. Internal material time-varying deformation displacement includes time-drift deformation displacement and local time-varying deformation displacement. Local time-varying deformation displacement is the remaining displacement after deducting the aforementioned components, maintaining a consistent main direction over at least five consecutive time intervals, and exhibiting the same trend of change among adjacent material coordinate anchor points. If only a single material coordinate anchor point undergoes a sudden change without a corresponding change at adjacent anchor points, it is considered a measurement anomaly and deleted.

[0052] The camera motion components, the overall rigid body displacement of the artifact, the reversible deformation displacement of the environment, the time-deformation displacement of the material's interior, and the reference state correction displacement required to map to the reference state are written into the fractional displacement data. Taking the material expansion and contraction that occurs 10 minutes after the change in environmental humidity as an example, the reversible deformation displacement of the environment should be calculated according to the environmental response lag time determined in S210, and this expansion and contraction should not be misjudged as natural creep of the material.

[0053] S230. Using distortion-corrected multi-view images at the same time marker as input, perform dense matching using a semi-global matching (SGM) or multi-view stereo vision (MVS) algorithm within the epipolar constraint range according to calibration parameters; construct spatial observation rays for matching pixels supported by at least two effective viewpoints, and calculate the three-dimensional position of surface points using least squares intersection. Only retain surface points with a maximum observation ray angle of not less than 10°, an average reprojection error not exceeding twice the calibration average reprojection error, and multi-view local texture differences not exceeding the texture difference threshold, to form the surface point cloud for each time marker; the texture difference threshold is determined based on the difference distribution obtained from at least 20 repeated acquisitions of the non-deformable surface region.

[0054] For any surface point, the nearest material coordinate anchor point within the same continuous surface region is selected. The reference state correction displacement is then weighted and propagated based on the anchor point distance and the material coordinate anchor point confidence level, without crossing cracks, pores, occlusion boundaries, or material boundaries. Even if surface points on either side of a crack are spatially close, they must not share the same anchor point neighborhood; otherwise, the actual crack opening will be incorrectly smoothed.

[0055] The effective number of viewpoints, the angle between observed rays, the average reprojection error, the local texture consistency, and the average confidence of neighboring anchor points are normalized to 0 to 1, and then weighted and averaged according to their correlation with the 3D positioning error to obtain the surface point confidence, with the sum of each weight being 1. The surface point cloud at each time point is mapped to the reference acquisition time and the reference environmental state; for multiple surface points falling into the same 3D spatial fusion unit, the weighted average position is calculated according to the surface point confidence, and the texture observation with the highest surface point confidence is retained.

[0056] The size of the 3D spatial fusion unit is taken as the larger of the actual size corresponding to the image spatial resolution and the average 3D positioning error. If the distance deviation between a surface point and its neighboring surface points in the same region exceeds three times the median distance deviation of the region, it is considered a discrete point and deleted; surface points supported only by a single effective viewpoint are not included in the fusion result. This yields reference state surface data, including the fused surface point cloud, surface point confidence, source time identifier, continuous surface region identifier, material coordinate anchor point position, correlation between surface points and neighboring material coordinate anchor points, and the reference state correction displacement field of the surface point cloud at each time step.

[0057] The reference state surface data is used in S310 to determine the candidate contact area for support, in S410 to divide the surface area, and in S510 to implement reverse unloading; the fractional displacement data is used in S530 to forward recover the observation state corresponding to each source time marker.

[0058] S3 specifically includes the following sub-steps: S310: Obtain the reference support state generated in S210, the reference state surface data generated in S230, and the display stand support structure data generated in S110. Based on the actual geometric dimensions of the support components, the installation position of the support points, the support direction, and the reference state surface data, determine the candidate contact area for each support point; the candidate contact area refers to the surface area where the support component may come into contact with the cultural relic and transmit the support reaction force.

[0059] Two support points with a continuous load-bearing path on the display stand structure, and whose corresponding candidate contact areas are located in the same or adjacent continuous stress areas, are identified as adjacent support point combinations. A continuous stress area refers to an area where the load transfer path does not cross a fracture, detachment from the contact point, or an independent component. Combinations where the reference support reaction force is less than three times the resolution of the reaction force acquisition device, or combinations that may lose contact after adjustment or cannot be compensated for by other support points, are excluded.

[0060] The reduction in support force relative to the reference support state is defined as the load transfer-out amount, the increase in support force is defined as the load transfer-in amount, and the additional adjustment amount generated by the remaining support points to maintain the posture of the artifact is defined as the compensation support amount. The minimum effective load change is taken as 3 times the resolution of the reaction force acquisition device; an initial trial adjustment is first performed according to the minimum effective load change amount, the ratio of the measured displacement of the material coordinate anchor point to the increase of the support reaction force is calculated, and the expected displacement corresponding to each candidate load change amount is linearly extrapolated accordingly.

[0061] The maximum permissible load variation is the minimum of the following: 3% of the reaction force of the reference support at the pivot point, the upper limit of the load specified in the cultural relic protection requirements, the load that ensures the expected displacement does not exceed 5 times the average three-dimensional positioning error, and the load that does not cause any support point to lose contact. Within this range, at least two incremental load levels are set, and a recovery state is configured for each load level.

[0062] Calculate the resultant force and resultant moment of the supports based on the support reaction vector and spatial location of each support point:

[0063]

[0064] in, and Let Q represent the resultant support force and resultant support moment under the s-th load perturbation state, respectively, and let Q represent the number of support points. This represents the support reaction vector at the q-th support point under the s-th load perturbation state. This represents the position vector from the reference point in the exhibition space to the q-th support point. This represents the cross product of vectors.

[0065] The load transfer amount and compensation support amount are adjusted in a coordinated manner to ensure that the deviation of the resultant support force and the deviation of the resultant support moment do not exceed 0.5% of the reference value. When the reference resultant support moment is close to 0, the absolute moment threshold obtained by converting the repeated measurement error of the reaction force acquisition device and the position error of the support point is adopted. A load perturbation sequence is generated according to the load level from small to large, and a unique state identifier is assigned to each load perturbation state.

[0066] Taking a 1N reduction in the left support point and a 1N increase in the right support point as an example, although the resultant force of the support can remain unchanged, additional torque will still be generated due to the different positions of action, and the third support point must provide the compensating support.

[0067] S320. The micro-displacement actuator (e.g., piezoelectric ceramic micro-displacement stage or precision slide driven by stepper motor) set at the support point is adjusted along the corresponding support direction according to the load perturbation sequence. The minimum adjustment resolution of the micro-displacement actuator should be able to generate a support reaction force change that is not less than the minimum effective load change.

[0068] After each adjustment is completed, the current load is maintained and a load perturbation stabilization time window is established, which includes at least 10 consecutive image acquisition moments. When the difference between the maximum and minimum values ​​of the support reaction force at each support point within the time window is not greater than twice the repeated measurement error of the reaction force acquisition device, the average displacement change rate of the material coordinate anchor point is not greater than the ratio of the average three-dimensional positioning error to the image acquisition cycle, the environmental parameters are within the allowable range of the reference environmental state, and the overall translation and overall rotation of the cultural relic do not exceed twice the average three-dimensional positioning error and 0.05°, respectively, the current load perturbation state is determined to be stable.

[0069] The median of the support reaction force vectors within the load perturbation stabilization time window is taken as the actual support reaction force, and the increment of the support reaction force is calculated:

[0070] in, This represents the increment of the support reaction force at the q-th support point under the s-th load perturbation state. This represents the reference support reaction vector at the q-th support point under the reference support condition.

[0071] Multi-view images are acquired synchronously. The three-dimensional position of the material coordinate anchor point is determined according to S110 to S130. Based on the material coordinate state compensation parameters generated in S210 and the fractional displacement data generated in S220, the camera motion component, the overall rigid body displacement of the cultural relic, the reversible deformation displacement of the environment, and the time-deformation displacement inside the material are deducted. The remaining displacement is determined as the material coordinate displacement caused by the load perturbation.

[0072] For each material coordinate anchor point, the three-dimensional displacement vector, displacement amount, principal displacement direction, neighborhood displacement gradient, and material coordinate anchor point confidence level are recorded. The principal displacement direction is the normalized direction of the material coordinate displacement vector; the neighborhood displacement gradient is a coefficient matrix obtained by selecting the six closest material coordinate anchor points within the same continuous surface region, fitting the linear relationship between the three-dimensional displacement vector and the local material coordinates using the least squares method. If there are fewer than six material coordinate anchor points in the neighborhood, all available material coordinate anchor points within that region are used. This forms the material coordinate displacement field.

[0073] After each load perturbation state ends, the reference support state is restored. If the restored position deviation of any material coordinate anchor point is greater than twice the corresponding three-dimensional positioning error, or if a continuous displacement gradient migration occurs within the candidate contact area of ​​the support, the state is marked as a contact slip state. Immediately acquiring data after load adjustment will mix the unstable creep response into the material coordinate displacement field; therefore, only data within the stable time window is used.

[0074] S330. Using the status identifier as an index, establish a correspondence between the support point combination, load transfer out of the support point, load transfer into the support point, actual support reaction force vector of each support point, support reaction force increment, support resultant force deviation, support resultant moment deviation, environmental status, load perturbation stabilization time window, material coordinate displacement field, contact slip identifier, and recovery status deviation.

[0075] The corresponding load perturbation state will be determined as invalid if any of the following conditions are met: the deviation of the resultant force or the deviation of the resultant moment of the support exceeds the corresponding threshold; the overall translation or rotation of the cultural relic exceeds the allowable range; the environmental parameters exceed the allowable range of the reference environmental state; no load perturbation stabilization time window is formed; the number of effective material coordinate anchor points is less than 70% of the effective number in the reference state; the increment of the support reaction force is less than 3 times the resolution of the reaction force acquisition device.

[0076] The state with a recovery deviation exceeding twice the corresponding 3D positioning error is not used for local stiffness and material orientation inversion, but its contact slip marker and displacement gradient abrupt change location are retained for S410 to determine the support contact range. For each adjacent support point combination, at least one load level is selected and repeated twice; when the included angle between the two displacement directions of the same material coordinate anchor point is not greater than 15°, and the difference in displacement is not greater than 10% of the average of the two displacements, the median of the two results is taken; otherwise, it is marked as insufficient repeatability.

[0077] Taking two displacements with opposite directions under the same load condition as an example, this result may come from image matching errors, unstable responses, or contact slip, and should not be used for mechanical parameter inversion. The effective states are first grouped according to adjacent support points, then according to the load transfer direction, and finally sorted in ascending order of the absolute value of the support reaction force increment at the load transfer support point, forming a load perturbation state association dataset.

[0078] The incremental support reaction force and material coordinate displacement field are used to calculate local stiffness in S410. Data for different load directions are used to determine material orientation. Displacement gradient and contact slip indicators are used to determine the support contact range. Recovery state deviation is used to identify irreversible responses.

[0079] S4 specifically includes the following sub-steps: S410: Obtain the reference state surface data generated in S230, the reference support state generated in S210, the load perturbation state association dataset generated in S330, and the mass, thickness, and material composition data from the cultural relic protection archives or display stand installation records to establish the basis for mechanical parameter inversion. Thickness is preferentially determined using measured values ​​from the cultural relic protection archives; if no measured values ​​are available, non-contact optical thickness measurement results or the spatial distance between the front and back surfaces are used; if still unmeasurable, the median of known thicknesses within the same continuous surface area is used and marked as the thickness estimation area.

[0080] For single-material artifacts, the density is primarily determined by archival records. If no records are available, the overall equivalent density is calculated by dividing the artifact's mass by the model's volume. For multi-material artifacts, the regional density is primarily determined by archival records or material sample measurements. If these cannot be obtained separately, a unified overall equivalent density is assigned, and the region is marked as a low-distinction mechanical parameter area. Regions with a thickness less than 1 / 10 of the local surface feature size utilize shell structures, enclosed regions with non-negligible thickness utilize solid structures, and slender connecting parts utilize rod structures.

[0081] Surface regions are defined based on material boundaries, cracks, pores, structural boundaries, and continuous surface area markers. Each surface region must contain at least three non-collinear material coordinate anchor points; if there are fewer than three, it is merged with the adjacent surface region that has the same material properties and the smallest curvature difference. Local stiffness is the ability of a surface region to resist displacement caused by an equivalent load increment under given thickness, structural form, and support conditions; material directionality is the difference in local stiffness along different directions within the same region and its principal direction; the support contact range is the continuous surface range where the artifact and the support directly transmit the support reaction force.

[0082] For each support point, connected surface points with displacement greater than 3 times the average three-dimensional positioning error and continuous neighborhood displacement gradient are selected in the corresponding support candidate contact area to form the initial support contact range, and the boundary is corrected by the abrupt change position of displacement gradient in the contact slip state.

[0083] Only a load perturbation state with a recovery state deviation not exceeding twice the corresponding three-dimensional positioning error is used. The equivalent load increment and material coordinate displacement of the region are projected onto the local material coordinate system of the region. The initial local stiffness is determined by minimizing the difference between the predicted displacement and the measured displacement. When the same region has at least two effective load directions with an included angle of not less than 30°, the material directionality is determined according to the displacement response ratio of different directions. Otherwise, the region is set to a direction-independent state.

[0084] For example, when only horizontal load data is available, the local stiffness in the vertical direction cannot be assumed. Combining the self-weight load of the artifact, the reaction force of the reference support, and the aforementioned parameters, the initial stress distribution that ensures the mechanical equilibrium of the reference state surface mesh is determined. The initial stress distribution includes the external load-induced stress released as the loads from the display frame support and the artifact's self-weight are removed, as well as the inherent residual stress of the material that still maintains internal self-equilibrium under zero external load conditions.

[0085] For thin paper artifacts, thickness and local stiffness jointly affect bending displacement; therefore, local stiffness cannot be directly inverted before the source of thickness is determined. Initial mechanical parameters are composed of the initial support contact range, initial local stiffness, material orientation, and initial stress distribution.

[0086] S420. Execute strategy optimization using initial mechanical parameters as the starting value. Strategy optimization involves selecting the parameters to be adjusted, the surface area, and the adjustment direction based on the location, type, and sensitivity of the prediction error, and iteratively reducing the prediction error within the allowable range. The first stage corrects the support contact range based on the contact slip state; the second stage adjusts the local stiffness using small-amplitude load perturbation; the third stage adjusts the material directionality using different load directions; the fourth stage adjusts two types of initial stresses based on the equilibrium residual and the difference in load increase / decrease response; and the fifth stage jointly corrects four types of mechanical parameters to reduce parameter substitution.

[0087] Establish the target value for strategy optimization:

[0088] Where J represents the policy optimization objective value, This represents the confidence-weighted average of the squared values ​​obtained by dividing the difference between the predicted and measured material coordinate displacements by the corresponding 3D positioning error. This represents the average of the squared values ​​obtained by dividing the difference between the predicted and actual support reactions by the corresponding reaction force acquisition device's repeated measurement error. P represents the penalty value for parameter out-of-bounds errors, parameter abrupt changes without structural boundary support, and discontinuities in the support contact range. , and These represent the displacement error weight, support reaction force error weight, and parameter constraint weight, respectively. Each error is divided by the corresponding measurement error to eliminate unit differences.

[0089] In practice, since all errors have been normalized, , and The preferred value range is [0.1, 1.0]. As an example, all three values ​​can be uniformly set to 1.0; or, based on the different sensitivities of the actual artifact material to displacement and reaction force, they can be set to... =0.6、 =0.3、 =0.1.

[0090] The allowable range of local stiffness is determined based on the extreme value relationship between the regional equivalent load increment and the measured displacement. The allowable range of material directionality is determined based on the displacement response ratio in different directions. The support contact range is limited to the actual geometric projection range of the support. The allowable range of initial stress is determined based on the equilibrium conditions of the reference support state and the self-equilibrium conditions within the zero external load. The parameter change for each sensitivity test is 1% of the allowable range width; this change is maintained when the target value decreases for two consecutive times, and halved when the target value increases. After each parameter update, the entire correction load perturbation state is resolved.

[0091] The process stops when both the normalized displacement error and the normalized support reaction error are no greater than 1, or when the target value decreases by less than 0.5% for five consecutive iterations, thus obtaining the optimized mechanical parameters. For regions with insufficient observation, the median of parameters from adjacent regions of the same material is used as the initial value, and low-identity mechanical parameter region identifiers are retained.

[0092] S430: Generate a reference state surface mesh based on the fused surface point cloud. The edge length of the basic mesh is 1 to 3 times the size of the 3D fused unit. The mesh size is reduced at locations of abrupt curvature changes, around cracks, and at the boundaries of the support contact area. Set shell units, solid units, or rod units according to the structural form determined in S410, and assign optimized mechanical parameters, thickness, and density to the corresponding surface areas. The support reaction force is the force of the support frame measured by the reaction force acquisition device. The support reaction force vector of each support point is applied to the support contact area to form the support load of the support frame.

[0093] The display stand constraints are set according to the support point location, support direction, and support contact range. The self-weight load of the cultural relic is applied according to the density, model volume, and gravity direction. Cracks, holes, and material boundaries must not be directly connected through mesh smoothing. The load perturbation state association dataset is divided into a calibration state set and a verification state set. At least one repeatable load perturbation state is retained as a verification state for each adjacent support point combination. The verification state does not participate in the parameter adjustment of S420.

[0094] A finite element statics algorithm (e.g., using ANSYS, Abaqus, or an open-source finite element library) is employed to apply incremental support reaction forces to each element in the correction state set. Based on the correspondence between material coordinate anchor points and mesh elements, element node displacement interpolation is used to obtain the predicted material coordinate displacement. The contact node reactions within the corresponding support contact range are then summed to obtain the predicted support reaction force. When the error exceeds the limit in a certain surface region, only the optimized mechanical parameters of that region and its first-order adjacent surface regions are adjusted, but the entire constraint mechanical model of the display frame is resolved after each adjustment. For example, reducing the local stiffness of a thin-walled region may simultaneously change the reaction force at the far-end support point; therefore, only the local region should be calculated.

[0095] After calibration, the model is independently verified using a set of validation states. The model is considered to have passed validation when the predicted material coordinate displacement error is not greater than the corresponding three-dimensional positioning error, the predicted support reaction force error is not greater than the corresponding reaction force acquisition device repeated measurement error, and there is no local displacement error in a continuous surface area that exceeds twice the corresponding three-dimensional positioning error.

[0096] Output the corrected display stand constraint mechanical model, including the reference state surface mesh, structural element type, region thickness and density, optimized mechanical parameters, display stand constraint conditions, gravity direction, material coordinate anchor point correspondence with mesh element, verification state set and low-recognition mechanical parameter region identifier, for use by S510 and S530.

[0097] S5 specifically includes the following sub-steps: S510 takes the corrected display stand constraint mechanical model generated by S430, the reference environment state and reference support state generated by S210, the reference state surface data generated by S230, and the optimized mechanical parameters generated by S420 as inputs.

[0098] The intrinsic 3D model of the cultural relic is a zero-external-load 3D geometric model obtained by eliminating the deformation corresponding to the support load of the display stand and the self-weight load of the cultural relic under the condition that the inherent residual stress of the material is preserved, while the cultural relic is in a reference environment state. Reverse unloading only releases the external load-induced stress, without deleting the inherent residual stress of the material. The reference state surface mesh is used as the initial geometry, the support load of the display stand is formed by the reference support reaction force of each support point, and the self-weight load of the cultural relic is calculated according to the area density, model volume and gravity direction.

[0099] First, divide the display stand support load into no less than 10 unloading levels. The remaining display stand support load of the h-th unloading level is determined according to the following formula:

[0100] in, This indicates the support load of the display stand under reference support conditions. This represents the remaining support load of the display frame at the h-th unloading stage. This indicates the unloading stage number of the display stand support load, where h represents the unloading stage number, and .

[0101] The grid position, stress state, and contact state after balancing at the previous level are used as inputs for each unloading stage. After the support load of the display stand is reduced to 0, the self-weight load of the cultural relic is reduced to 0 using the same grading method, and the displacement corresponding to the final load must not be deducted all at once. To eliminate the overall rigid body degrees of freedom, the centroid of the reference state surface grid is used as the origin of the coordinate system, and the three principal inertial directions are kept to correspond to the coordinate system of the exhibition space. This numerical reference constraint does not generate actual support reaction force.

[0102] After each unloading stage is completed, the mechanical equilibrium residual, nodal displacement change, mesh cell orientation and non-adjacent surface spacing are calculated. When the mechanical equilibrium residual is no greater than 1% of the remaining external load of the stage, the maximum nodal displacement difference between two consecutive iterations is no greater than the average three-dimensional positioning error, the mesh cells are not flipped, no new self-intersections are generated on the non-adjacent surfaces, and the topological relationship of cracks, holes and material boundaries remains unchanged, the next unloading stage is entered.

[0103] If the conditions are not met, the current unloading amplitude is halved and recalculated; regions that still do not converge after three consecutive halvings are marked as low-confidence reverse unloading regions. Taking thin paper artifacts as an example, their displacement and load are usually not strictly linearly related, and eliminating gravity at once may cause local mesh flipping, so it is necessary to solve in stages.

[0104] After the reverse unloading is completed, the intrinsic three-dimensional model of the cultural relic and the intrinsic deformation mapping relationship are obtained. The intrinsic deformation mapping relationship records the corresponding identifiers of the reference state surface mesh nodes and the intrinsic three-dimensional model nodes of the cultural relic, the cumulative reverse displacement, the intermediate displacement of each unloading stage, the stress state after unloading, and the region identifier.

[0105] S520. Group the effective viewing angles according to the optical axis direction of the image acquisition device, classify the viewing angles with an optical axis angle of no more than 15° into the same main observation direction, and reserve at least one verification viewing angle from each main observation direction; the verification viewing angle does not participate in region identification, mesh subdivision, and vertex adjustment. Using the intrinsic 3D model of the cultural relic as the basis for 3D modeling, obtain the surface point confidence, source time identifier, and continuous surface region identifier saved in S230, and combine them with calibration parameters and the original multi-view images obtained from the S110 multi-source synchronous observation dataset, excluding the verification viewing angle, to identify the region to be enhanced.

[0106] Multi-view geometric evidence is the observation of consistent parallax, contour displacement, occlusion relationship or surface normal change at the same surface location in at least two valid viewpoints; color edges, stain boundaries, single-view shadows and pigment color differences do not constitute multi-view geometric evidence on their own.

[0107] A candidate region is determined to be enhanced if it is supported by at least two valid viewpoints, the edges of the back-projected structure fall into the same continuous surface area, the concavity and convexity directions are consistent, the structure size is greater than the average 3D positioning error, the confidence level of the surface points is not lower than the median of the region, and it does not belong to a low-confidence reverse unloading region. Taking paper ink as an example, even if there is a significant grayscale gradient, as long as there is no corresponding parallax or surface normal change between different viewpoints, it is only used as texture information and no concave structure should be generated.

[0108] For triangular meshes in the enhanced region whose side length is greater than the target mesh size, the mesh size is subdivided step by step. The target mesh size is taken as the larger of the following: the actual size corresponding to the original image pixels, the average 3D positioning error, and half the width of the multi-view structure edge.

[0109] Newly added mesh vertices are initially positioned by interpolation from the original triangular mesh, primarily adjusted using surface normals. When tangential deviations exist at corresponding positions on the edges of multi-view structures, edge alignment along the surface tangent is permitted, but the tangential movement must not exceed 1 / 4 of the target mesh size. After vertex movement, the projection residuals in at least two effective views should decrease simultaneously, and the increase in projection residuals in other effective views must not exceed the calibrated average reprojection error.

[0110] During the structural reinforcement process, the changes in the position of the material coordinate anchor points must not exceed the corresponding three-dimensional positioning error, the changes in the projection of the outer contour of the cultural relic in each effective view must not exceed the calibration average reprojection error, the connectivity of cracks, holes and material boundaries remains unchanged, and the position of the mesh nodes in adjacent non-reinforced areas does not change.

[0111] This yields candidate refined 3D models and a record of structural enhancement evidence, the latter including region identifiers, source images, number of valid viewpoints, multi-view geometric evidence types, vertex adjustment amounts, verification viewpoint identifiers, and structural enhancement confidence levels. Low-confidence reverse-unloading regions do not participate in structural enhancement; only the original geometry and confidence level are retained.

[0112] S530: Obtain the candidate refined 3D model generated by S520, the verification perspective and structural enhancement evidence record, and implement positive state recovery based on the intrinsic deformation mapping relationship generated by S510, the corrected display frame constraint mechanical model generated by S430, the fractional displacement data generated by S220, and the environmental state and support reaction force corresponding to each source time marker.

[0113] Forward state recovery starts from the candidate refined 3D model, applies the artifact's self-weight load and the corresponding display stand support load at the time of acquisition in the reverse order of reverse unloading, and restores the corresponding reversible deformation displacement of the environment and the time-varying displacement of the material's internal structure, generating a model with the same observation state as the original image. Zero-load models must not be directly projected onto the stress state image for comparison.

[0114] The forward-restored model is projected onto the corresponding verification viewpoint, and the deviation between the projected position of the model structure edge and the measured position of the structure edge in the image is calculated. If the average projection deviation is not greater than the calibrated average reprojection error, the projection consistency check is passed. The candidate refined 3D model is then forward-loaded into the verification state set, and the predicted displacement of the material coordinate anchor points and the predicted support reaction force of the support points are calculated. If the predicted displacement error is not greater than the corresponding 3D positioning error, the predicted support reaction force error is not greater than the corresponding reaction force acquisition device repetitive measurement error, and there is no abnormal migration of the support contact range, the mechanical response consistency check is passed.

[0115] When the projection error exceeds the limit but the mechanical response is qualified, the corresponding region is fed back to S520 to adjust the local mesh subdivision range or vertex displacement; when the mechanical response error directions are consistent in multiple verification states, the corresponding region is fed back to S420 to adjust and optimize the mechanical parameters; when the projection error occurs only in a single verification view and there is no other multi-view geometric evidence in the structural enhancement evidence record, the corresponding geometric enhancement is canceled and the texture information is retained; when both types of errors exceed the limit at the same time, the geometric structural enhancement of the corresponding region is canceled first, and then it is determined whether to adjust and optimize the mechanical parameters.

[0116] The number of feedback corrections shall not exceed 5; if the verification fails after 5 corrections, the model version shall be restored to the one that passed the mechanical response consistency verification most recently and marked as a low-confidence structural reinforcement region.

[0117] Once all regions involved in structural enhancement have passed the projection consistency check and mechanical response consistency check, the output includes a refined 3D model, texture map, structural enhancement evidence record, structural enhancement confidence, low-confidence reverse unloading region, low-confidence structural enhancement region, low-identity mechanical parameter region, and positive state recovery parameters for cultural relic display.

[0118] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.

[0119] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0121] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display, characterized in that, Includes the following steps: S1. While the cultural relic is supported by the display stand, collect multi-view images, support reaction force, environmental parameters and collection time, extract material feature points and establish material coordinate anchor point time series data; S2. Determine the reference environment state and reference support state based on the time series data of material coordinate anchor points, separate the camera motion, the overall rigid body motion of the cultural relic and the material deformation, map the surface data to the reference environment state, and generate the reference state surface data. S3. Generate a load perturbation sequence based on the reference support state. When the resultant force, resultant moment and spatial position of the artifact are within the allowable range, transfer the support load between adjacent support points, collect the support reaction force increment and material coordinate displacement field, and form a load perturbation state association dataset. S4. Based on the reference state surface data and the load perturbation state associated dataset, determine the local stiffness, material orientation, support contact range and initial stress distribution, and construct the display frame constraint mechanical model through strategy optimization.

2. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 1, characterized in that, Also includes: S5. Using the constraint mechanics model of the display stand, the support load of the display stand and the self-weight load of the cultural relic are removed in stages to obtain the intrinsic three-dimensional model of the cultural relic. Based on multi-view geometric evidence, the structure is enhanced, and the enhanced model is restored to the acquisition state in the positive direction. After the projection consistency verification and mechanical response consistency verification, a refined three-dimensional model is output.

3. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 1, characterized in that, S1 specifically includes: While the cultural relic is supported by the display stand, image, reaction force and environmental acquisition devices are configured. After calibration, multi-view images, support reaction force and environmental parameters are aligned according to time markers to form a multi-source synchronous observation dataset and display stand support structure data. Correct multi-view images, extract material feature points that move together with the cultural relics, establish cross-view correspondence based on texture, adjacency relationship and motion continuity, and form candidate material coordinate anchor points.

4. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 3, characterized in that, Also includes: The initial three-dimensional position of the candidate material coordinate anchor point is calculated based on the calibration parameters and pixel position. After filtering by ray angle and reprojection error, the material coordinate anchor point and its confidence level are determined. The time stamp, environmental parameters and support reaction force are associated to form the time series data of the material coordinate anchor point.

5. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 1, characterized in that, S2 specifically includes: Based on the material coordinate anchor point time series data, the reference acquisition time, reference environment state and reference support state are determined, and material coordinate state compensation parameters are established. Based on the fixed coding mark and the three-dimensional position of the material coordinate anchor point, the camera motion components and the overall rigid body displacement of the cultural relic are determined, and the reversible deformation displacement of the environment and the time-deformation displacement inside the material are deducted to form fractional displacement data. Dense matching is performed on multi-view images under the same time marker to calculate the three-dimensional position of surface points. Based on the fractional displacement data, the surface point cloud at each time moment is mapped to the reference acquisition time and reference environmental state. Surface points that meet the multi-view support conditions are fused with confidence weight to form reference state surface data.

6. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 1, characterized in that, S3 specifically includes: Based on the reference support status, reference status surface data and display stand support structure data, determine the combination of adjacent support points, set the load transfer out amount, load transfer in amount and compensation support amount, and generate a load perturbation sequence when the resultant support force, resultant support moment and spatial position of the cultural relic are within the allowable range. The support load is adjusted according to the load perturbation sequence. After the state stabilizes, the support reaction force and multi-view images are collected. The increment of the support reaction force is calculated, and the material coordinate displacement field is formed after deducting the displacement of non-load perturbation.

7. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 6, characterized in that, Also includes: By associating the incremental support reaction force, material coordinate displacement field, environmental conditions, contact slip indicators, and recovery state deviations, invalid and insufficiently repeatable states are eliminated to form a load perturbation state association dataset.

8. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 1, characterized in that, S4 specifically includes: The surface region is divided based on the reference state surface data, the reference support state and the associated dataset of the load perturbation state, and the local stiffness, material orientation, support contact range and initial stress distribution are determined to form the initial mechanical parameters; Starting with the initial mechanical parameters, the support contact range, local stiffness, material orientation, and initial stress distribution are adjusted sequentially based on the predicted displacement error, predicted support reaction force error, and parameter constraints to obtain the optimized mechanical parameters. A surface mesh is established based on the reference state surface data, optimized mechanical parameters are assigned and display stand constraints are set, the model is corrected using the correction state set, and the corrected display stand constraint mechanical model is obtained by independent verification through the verification state set.

9. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 2, characterized in that, S5 specifically includes: Based on the corrected display stand constraint mechanical model, the display stand support load and the self-weight load of the cultural relic are removed in stages. Under the condition of preserving the inherent residual stress of the material, the intrinsic three-dimensional model of the cultural relic and the intrinsic deformation mapping relationship are obtained. Based on multi-perspective geometric evidence, the areas to be enhanced are identified. The original 3D model of the cultural relic is then subjected to constrained mesh subdivision and vertex adjustment to form candidate refined 3D models and records of structural enhancement evidence.

10. The method for three-dimensional model reconstruction and structural refinement enhancement for cultural relic display according to claim 9, characterized in that, It also includes: forward restoring the candidate refined 3D model to the corresponding acquisition state, performing projection consistency verification and mechanical response consistency verification using the verification perspective and verification state set respectively, adjusting the structural enhancement or optimizing the mechanical parameters according to the verification results, and outputting the refined 3D model.

Images