Financial building high-precision modeling method based on three-dimensional reconstruction

Abstract

The embodiment of the invention provides a financial building high-precision modeling method based on three-dimensional reconstruction, and the method comprises the steps: extracting geometric information and texture information from a multi-view image of a financial building, and forming an initial three-dimensional model; analyzing complex curved surface and detail texture characteristics based on the extracted geometric and texture information, and determining parameter setting of a grid optimization algorithm; performing grid optimization on the initial three-dimensional model according to the determined parameter setting, retaining key details and reducing redundant data; and the optimized three-dimensional model is simplified, and the model resolution is reduced while the detail integrity is kept. Through the scheme of the embodiment of the invention, the problem of how to regulate and control the parameter setting of the grid optimization algorithm according to the complex curved surface and detail texture characteristics so as to solve the problem of detail loss in the high-resolution model simplification process can be solved.

Description

Technical Field

[0001] The present application relates to the technical field of computer vision and graphics, and more particularly to a high-precision modeling method for financial buildings based on three-dimensional reconstruction. Background Art

[0002] A high-precision modeling method for financial buildings based on three-dimensional reconstruction mainly uses multi-view images or laser scanning technology to accurately collect and reconstruct financial buildings, and uses computer graphics and computer vision algorithms to efficiently restore information such as the building's geometric features, texture details, and material properties, providing accurate and visual data support for architectural design, evaluation, and cultural relics protection. However, there is an important problem when dealing with such models: how to reasonably adjust the parameter settings of the mesh optimization algorithm according to the characteristics of complex surfaces and detailed textures to solve the problem of detail loss that may occur in the process of simplifying high-resolution models. This is because complex surfaces and rich textures require different algorithms and parameters to retain their unique appearance characteristics, and simple parameter settings cannot adapt to different situations, resulting in the loss of original details in the simplified model. Summary of the invention

[0003] In view of this, an embodiment of the present disclosure provides a high-precision modeling method for financial buildings based on three-dimensional reconstruction, which at least partially solves the problems existing in the prior art.

[0004] A high-precision modeling method for financial buildings based on three-dimensional reconstruction, comprising: Extract geometric information and texture information from multi-view images of financial buildings to form an initial 3D model; Analyze complex surfaces and detailed texture characteristics based on the extracted geometry and texture information, and determine the parameter settings of the mesh optimization algorithm; Optimize the mesh of the initial 3D model according to the determined parameter settings, retain key details and reduce redundant data; The optimized 3D model is simplified to reduce the model resolution while maintaining the integrity of the details.

[0005] Preferably, analyzing the complex surface and detailed texture characteristics based on the extracted geometry and texture information, and determining the parameter settings of the mesh optimization algorithm further includes: Get the proportion of complex surfaces in the image P_complex; Determine the mesh refinement coefficient K_initial; Adjust the mesh refinement factor K = K_initial * (P_complex + 1) based on the following formula to ensure that the detailed texture is not lost due to simplification; If K > K_max, then set K to K_max, where K_max is the set maximum refinement coefficient.

[0006] Preferably, based on the extracted geometric and texture information to analyze the characteristics of complex surfaces and detailed textures, the parameter setting of the mesh optimization algorithm further includes: Calculate the initial detail density D_init and weight it according to the material distribution M of the building. Set the texture fidelity coefficient F_textural to a range value from 1 to 10. Use the formula C_optim = F_textural * (M / max(M)) to adjust the mesh optimization threshold C_optim. If C_optim < threshold, increase the number of mesh points to maintain the details in the key area, where threshold is the set lower limit.

[0007] Preferably, on the basis of adjusting the mesh refinement coefficient, it is further limited and also includes the steps: Estimate the impact I_effect of model simplification on the details of the original complex surface. Set the initial impact tolerance R0 to ensure the minimization of detail loss. Use the formula R = exp(|I_effect|) to calculate the new tolerance R. Adjust the K_refinement parameter according to the R value to ensure that the number of meshes adapts to the changes in the surface characteristics, K_refinement = max(R*K, min_refinement). Where min_refinement is the minimum refinement degree.

[0008] Preferably, it also includes the steps: Analyze the key node density N_nodes at multiple locations in the simplified model. Set the expected node retention ratio R_expected. Apply N_preserve = N_original * R_expected to calculate the number of nodes to be retained. Evaluate the error of the simplified model according to the formula E_simplicity_loss = |N_preserved / N_expected - 1|, and when this error is greater than the set upper limit, reduce the amount of node removal to prevent the shape distortion caused by excessive simplification.

[0009] Preferably, the application of the formula is supplemented: The degree of detail loss is quantified by calculating the disparity V_disp between the simplified model and the original model at multiple angles; Given a reference value V_std and a sensitivity adjustment factor η, the intensity of the impact of parallax changes on user visual effects is controlled; Use the formula S_adjust = (V_std - V_new) / η to adjust the detail adjustment factor so that the appearance difference of the simplified model remains within the acceptable range for the human eye; If S_adjust>adjust_limit, the level of refinement is appropriately increased to ensure a consistent user experience. The upper limit of the adjustment factor here is adjust_limit.

[0010] Preferably, the following steps are additionally added: According to different building types, a specific detail priority matrix A_priority is set, in which each element reflects the importance weight w_i of a certain type of detail structure of that type; Measure the corresponding detail retention in the actual processing results and generate a score vector B_score; According to B_priority_weight = Sum(A_priority.*B_score), all scores are summed up to comprehensively evaluate the key retention level in the optimization process; Compare with the target standard value Standard. Once the difference is found to be beyond the allowable range, the correction parameters are fed back until it is close to the standard, that is, |B_priority_weight - standard_wt)|≤ delta_w.

[0011] Preferably, Combined with the lighting environment under different light source conditions, the maximum deviation Δv_max between mesh vertices is dynamically adjusted: Consider the difference factor f(L, I) of the total illumination distribution of natural light L_nat and indoor light I_in; Define the maximum deviation Δ_v_ideal reference value when there is no shadow in the ideal situation; Based on the comprehensive effect of f*L+ g*I illumination on Δ_v_ideal influence function H, the real-time corrected Δv_adj is obtained; Finally, it is judged that Δv = H(f*L, g*I; Δ_v_ideal). If Δv>Δv_max, Δv is reduced to not exceed Δv_max, thereby protecting the visual fidelity in complex environments.

[0012] Preferably, Added subdivision measures to address the problem of different treatment of sharp corners and smooth edges during mesh optimization: Increase the local density boost ratio P_boosted to 5 times the normal level for edges identified as sharp boundaries; For non-boundary parts, reduce to normal ratio P_decrease = P_base * ratio to reduce redundancy; By using the dichotomy principle of smooth continuity and sharp discontinuity, a detection mechanism is constructed to identify the interface between the two and implement differentiated optimization rules; Ensure the consistency of the visual characteristics of the boundaries before and after optimization by calculating C_shape_similarity = sqrt(sum(Di^2) / n). This similarity coefficient is used as the criterion. When C>limit, the current configuration will be reviewed to achieve simplicity and sufficient descriptiveness.

[0013] Preferably, To more accurately locate key details and perform special marking: Introduce the spatial positioning auxiliary tool set T_localization to find all geometric shapes S_unq and textures F_pat with unique attributes; Establish a flag bit array M_flags to carry the location information of these key elements; According to the formula P_special_points = Count(Find(T_localization * S_unique + T_localization * F_pattern)); the number of special node sets to be enhanced is determined; Ensure that each subsequent round of simplification iteration does not affect the actual structural properties expressed by this set of special points. If a violation is found, immediately adjust the layout of the adjacent triangle meshes to make up for the potential defects.

[0014] The disclosed embodiment provides a high-precision modeling method for financial buildings based on three-dimensional reconstruction, including: extracting geometric information and texture information from multi-view images of financial buildings to form an initial three-dimensional model; analyzing complex surfaces and detailed texture characteristics based on the extracted geometric and texture information, and determining the parameter settings of the mesh optimization algorithm; optimizing the mesh of the initial three-dimensional model according to the determined parameter settings, retaining key details and reducing redundant data; simplifying the optimized three-dimensional model to reduce the model resolution while maintaining the integrity of the details. Through the solution of the disclosed embodiment, it is possible to solve the problem of how to adjust the parameter settings of the mesh optimization algorithm according to the complex surfaces and detailed texture characteristics to solve the problem of detail loss during the simplification of high-resolution models. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0016] Figure 1 The present invention is a flowchart of a high-precision modeling method of financial buildings based on three-dimensional reconstruction. DETAILED DESCRIPTION

[0017] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0018] Next, with reference to the accompanying drawings, the specific implementation steps and examples of a high-precision modeling method for financial buildings based on three-dimensional reconstruction of the present invention are described. This method can not only accurately restore the geometric structure and texture details of financial buildings, but also optimize and simplify the amount of model data on the basis of ensuring the integrity of the model, greatly improving the application efficiency and effect of financial building design, restoration and display.

[0019] Extracting the geometric and texture information of financial buildings from multi-view images is the first step. In practice, this process first involves image acquisition and preprocessing. By using drones, ground cameras or other high-precision shooting devices to obtain all-round pictures of the building, these pictures cover the different heights, positions and perspectives of the building, ensuring that all important parts are clearly imaged. Then the collected photos are imported into the image registration software, and the spatial correspondence between the overlapping areas of all pictures is established with the help of feature point detection and matching technology. After that, the 3D coordinates of each pixel in the real world are calculated using the multi-view stereo vision algorithm or the principle of photometric consistency to restore the spatial form of the building's outer contour lines and the uneven parts of the facade. For example, the exterior of a landmark building in Beijing Financial Street is decorated with a large number of exquisite reliefs, and there are a large number of curved and folded structures on its surface that need to be accurately recorded; in addition, for the reflection of the building's glass curtain wall, additional light analysis correction is required to avoid miscalculation. After completing the geometric modeling, it is also necessary to synchronously process the visual features of the image itself, such as color, material, gloss, etc., and attach them to the corresponding mesh vertex attribute values, so that the entire initially constructed virtual entity can be closer to the appearance of the objective object.

[0020] In order to ensure that the generative model can express various special structural elements existing in the real world, it is necessary to deeply analyze the shape feature information collected previously and the composition mechanism of the pattern style it carries. For complex surfaces (such as arc tops) or areas with rich detailed textures (such as classical column carvings and imitation stone textures), curvature estimation, Fourier transform spectral domain decomposition or principal component analysis dimensionality reduction methods are used to identify those important indicator parameters representing key properties and combine them to form a set of characterization rules to guide the logical direction of filtering and downsampling operations in subsequent stages. Specifically, for lamellar tissues with higher degree of continuous undulation and coherence, the constraints on the size of adjacent units can be appropriately relaxed. On the contrary, if there are very significant mutations locally, the fine granularity should be maintained as much as possible, and different weight coefficients should be set according to the intensity of color contrast to quantify the distribution mechanism of their relative contribution ratio, so that the algorithms in subsequent links can automatically adjust their own operation strategies based on such prior knowledge to adapt to changes in specific scene requirements. For example, when dealing with a historic bank building in New York, we found exquisite Rococo-style friezes on the columns, which were small-scale details that needed special attention to preserve. Therefore, high-frequency segmentation was used to maintain the original phase offset to ensure that the smoothness of the lines and the sharpness of the edges were not damaged and the characteristics were not lost.

[0021] Then, we enter the mesh optimization process. The goal of this task is to achieve the best overall quality at the cost of minimum loss of core elements, so it requires taking into account two aspects: removing some unnecessary repeated connection edges or isolated hanging end nodes to reduce storage space without destroying the original shape characteristics. Based on the previously established standards, the corresponding numerical range is set to define the reasonable threshold limit, and the switching timing of the LOD (level-of-details) multi-level visual scheme is dynamically controlled based on this. At the same time, the adaptive subdivision methodology is introduced to solve the stretching deformation problem caused by non-proportional scaling. In addition, the genetic selection tournament confrontation elimination system under the framework of evolutionary programming theory can be integrated to randomly initialize an initial population of individual member candidate solutions. After several iterative rounds of screening, the best result set is finally retained as a new basic template library member. Continuously improve until convergence stops and the predetermined conditions are met. It is considered that the global best answer option is found and the optimization search activity ends. For example, if the unique shape of the top of the twin towers of the Jinmao Tower in Shanghai is simplified to a basic cone, it would obviously be too rough and simplistic, so it is necessary to strictly follow the characteristic curve equation summarized before to approximate the actual situation without deviating too much from the allowable error limit.

[0022] After completing the preliminary correction work, the last important task is to further simplify and compress the entire system architecture, but make sure that the information contained inside is still rich, detailed and without omissions. At this time, the specific approach should be determined by comprehensively weighing factors such as the graphics rendering capabilities of the hardware equipment and the memory bandwidth bottleneck in the actual engineering practice application scenario. A more common method is to use half edge collapse, that is, to select a line between two adjacent fixed points that are relatively close and have a large exchange value instead of the two original separate paths. Connect them and directly delete the redundant intermediate points, leaving only a few key control supports to maintain a roughly similar appearance. The effect can greatly improve the transmission and download speed and the smoothness of the real-time browsing interactive experience. However, under this premise, additional protection measures need to be added to ensure that those small but critical functional components are not accidentally deleted to avoid increasing the possibility of accidents that cause the entire system to crash and paralyze.

[0023] During the implementation of the entire method, how to adjust the parameter settings of the mesh optimization algorithm according to the characteristics of complex surfaces and detailed textures to effectively deal with the problem of detail loss during the simplification of high-resolution models is a core technical challenge. This modeling method analyzes and evaluates the geometric and texture information in the initial three-dimensional model, and formulates a personalized solution based on the characteristic requirements of the specific project. On the one hand, by carefully studying the microscopic geometric morphological characteristics of the object surface, a reasonable mesh refinement level is determined to prevent the simplification from causing excessive smoothness and flatness of the surface; on the other hand, through the study of detailed texture characteristics, special protection strategies are adopted for these key areas with cultural or artistic value during the simplification process. This approach ensures that the original charm and characteristics of the building can be maintained at the macro and micro levels even after a significant data reduction. Take the renovation of an old private club next to the Champs Elysees in Paris as an example. The exterior walls of this old building with a profound historical and cultural heritage are inlaid with many precious artworks, and their textures are very complex. In the process of high-precision digital twin modeling, the researchers customized a set of parameter control mechanisms for the unique patterns of each craft. This allows the model resolution to be reduced while protecting every historical detail worth cherishing to the maximum extent. This method not only helps to reduce the computing burden, but also provides accurate and reliable material resource support for future cultural relics protection and re-creation.

[0024] Next, the present invention describes a method for analyzing complex curved surfaces and detailed texture characteristics based on extracted geometric and texture information to determine parameter setting of a mesh optimization algorithm. In this method, data is obtained from a three-dimensional reconstructed image of a financial building to ensure precision and efficiency of modeling.

[0025] The first step is to obtain the proportion of complex surfaces in the image, P_complex. This step aims to quantify the proportion of complex surfaces in the image. The proportion P_complex ranges from 0 to 1 and represents the importance of the complex surface in the entire modeled object. For example, a modernist bank building may have many unique decorative details, such as curved facades, relief decorations and other complex structures, which will make P_complex higher; while a traditional block-style design may have a lower P_complex. After determining P_complex, the system has a more accurate understanding of complex architectural elements.

[0026] The second step is to determine the initial mesh refinement coefficient K_initial. This coefficient is used to control the density and refinement level of the initial mesh division of the model. K_initial is a constant greater than 0. The optimal value is set according to the specific scenario, generally between 0.5 and 2. In one embodiment, if the building shape is relatively regular, a smaller K_initial value is selected to reduce the computational burden; if there are a large number of detailed structures that need to be accurately represented, a higher value is selected to ensure the effect. At this time, K_initial has a certain flexibility to cope with different types of buildings.

[0027] The third step is to use the formula K = K_initial * (P_complex + 1) to adjust the final mesh refinement coefficient K. The meaning of this formula is to comprehensively consider the initial mesh refinement requirements and the actual complexity, so that the accuracy and efficiency of the model are balanced. As the complexity increases, the degree of mesh refinement also increases. For example, when the P_complex of a certain part is very close to 1 (indicating that the part is almost entirely complex), the adjusted K is greatly improved, thereby retaining this part of the features more finely without losing the overall expressiveness.

[0028] Step 4: If the adjusted K value exceeds the maximum allowable upper limit K_max, limit it not to exceed K_max. The purpose of setting K_max here is to prevent unnecessary waste of computing resources and performance problems caused by over-optimization. Usually K_max is defined as a reasonable empirical value, such as 3 or a slightly larger specific value. Specifically, no matter how complex the local area is, the grid should not be infinitely subdivided to avoid affecting the overall rendering and interactive experience. In this way, the rationality of 3D model construction is effectively improved without distortion.

[0029] Next, the further included method of analyzing complex surface and detailed texture characteristics based on the extracted geometry and texture information and determining parameter settings for the mesh optimization algorithm of the present invention is described.

[0030] Calculate the initial detail density D_init and weight it according to the material distribution M of the building. In this step, the initial detail density refers to the default model refinement without specific conditions. In order to accurately reflect the different requirements for the final quality of different parts, the material characteristics are taken into account in the density adjustment. For example, different materials such as stone and glass may result in different rendering requirements. For a financial building, if a certain area uses a highly reflective curtain wall material, the D_init of this part may get a larger weighted value after calculation to ensure the detail display.

[0031] Let the range of the texture fidelity coefficient F_textural be from 1 to 10. This value reflects the expected degree of preservation of the original texture. A higher coefficient means trying to reproduce all visible features as completely as possible, not just the large geometric forms. In this process, each specific value of F_textural depends on considerations of the importance and visual impact of the details of the building facade. For example, there may be special carved decorations near the main entrance of a financial building that require a higher texture resolution.

[0032] The formula C_optim = F_textural * (M / max(M)) is used to determine the optimal threshold setting in mesh optimization. The meaning of the formula is to calculate the optimal solution by multiplying the defined texture fidelity by the current maximum material weight ratio. In this example, if a certain part has both unique design elements and a high material value (such as precious metals or stones), it will be given a more stringent but reasonable optimization standard to maintain both performance and quality.

[0033] If C_optim < threshold, increase the number of grids in the key area. Here, threshold represents the pre-set lower limit. This means that when some important structural or decorative features are at risk of being ignored due to being too fine in the ordinary optimization scheme, the system can recognize this situation and ensure that these elements are properly processed by encrypting the number of points. For example, in an embodiment, it specifically involves the drawing of complex colonnade patterns inside a large financial center. After detecting that the local optimization result is too low, more data points will be automatically generated to capture the minute unevenness.

[0034] In short, these steps ensure that, on the premise of achieving efficient modeling, the realism and accuracy of the model are maximally improved, especially applicable to those important public facilities with unique artistic styles or symbolic meanings.

[0035] Next, describe the steps further defined based on adjusting the mesh refinement coefficient in the present invention: estimate the impact I_effect of model simplification on the details of the original complex surface; set the initial impact tolerance R0 to ensure the minimization of detail loss; calculate the new tolerance R using the formula R = exp(|I_effect|); adjust the K_refinement parameter according to the R value to ensure that the number of grids adapts to the changes in the surface characteristics, K_refinement = max(R*K, min_refinement). Here, min_refinement is the minimum refinement degree.

[0036] First, in this process, the effect of the 3D architectural modeling simplification operation on the subtle details of the original structure will be evaluated at the beginning and recorded as I_effect. This value can reflect the size and importance of the shape distortion caused by the simplification process. This step plays a key role in ensuring that the subsequent work can preserve the original appearance as much as possible. For example, when simplifying the exterior wall with finely carved decorations, the specific parameters of the difference before and after simplification need to be calculated as the basis for adjustment.

[0037] Next, define a starting level, the initial impact tolerance R0, the core purpose of which is to set the maximum accuracy loss within the acceptable range to avoid oversimplification that leads to significant deviations in visual or physical properties. This value is usually within a small positive range of [1.01,2.0]. By default, it is recommended to use a value close to the middle, such as 1.5, as the optimal option, in order to find a balance point and obtain a more reasonable calculation speed benefit without sacrificing too much quality.

[0038] Then, we use the mathematical expression R = exp(|I_effect|) to get a new and updated version of the impact tolerance ratio value R. The absolute value and natural logarithm are combined here because both positive and negative effects should be uniformly quantified and the difference should be appropriately amplified (especially for some sensitive occasions with small changes), so as to highlight the degree of error; at the same time, since exp(·) is an increasing function, it can output a larger positive integer result in any non-zero real number input, thereby achieving the purpose of enhancing the weight.

[0039] Finally, based on the latest parameters obtained in the previous steps, the factor controlling the density of model detail generation is adjusted, called K_refinement. Specifically, it is combined with R * K through a product operation, and the larger of the two is selected, while ensuring that it will not fall below the specified minimum min_refinement (for example, 0.05) to prevent unexpected loopholes under extreme conditions. Such a complete logical chain design not only takes into account the rationality of resource coordination at the global level, but also pays attention to the accuracy requirements of capturing local morphological features. For example, in one embodiment, the above strategy was used in the process of highly restoring the facade of a large financial exchange building, successfully solving the contradiction between maintaining reasonable resolution and detail fidelity.

[0040] Next, the density of key nodes in multiple locations of the simplified model is analyzed. This step aims to evaluate the detail preservation in different areas of the 3D reconstructed model, that is, to calculate the number of key nodes in a specific area. For example, in one embodiment, high-density key node positioning is performed for key structural sections such as the bank's internal lobby in order to accurately capture design elements.

[0041] Next, set the expected node ratio. This ratio refers to a target value R_expected that is planned to maintain the original node density after completing the 3D modeling of the financial building, ranging from 0 to 1. Specifically, if the original goal is to reduce resource consumption while maintaining visual quality, this value should not be too low or too high. It is generally recommended to be between 0.5 and 0.7 to find a balance between efficiency and effect, so that the quality of the model can be guaranteed while ensuring performance.

[0042] Then, N_preserve = N_original * R_expected is applied to calculate the number of nodes to be retained. In this formula, N_original represents the total number of nodes before processing, and the total number of nodes to be retained is obtained by multiplying the original number by the above-set ratio. This is intended to accurately control the specific number of points that should be retained during the actual execution process according to the ideal simplification ratio determined in the early stage.

[0043] The error of the simplified model is evaluated according to the formula E_simplicity_loss = |N_preserved / N_nodes - 1|. The parameter N_nodes is the actual number of nodes retained in the final result of simplification, and the formula output represents the deviation from the expected simplification. If this error exceeds the predefined limit, it means that the degree of simplification is too drastic and may destroy the shape of the original model. The parameters need to be adjusted to reduce the amount of deletion until the error returns to an acceptable range, thereby avoiding irreversible information loss or shape distortion caused by over-simplification. For example, if significant deformation is found when excessive deletion is performed in an attempt to improve the rendering efficiency of the exterior of a commercial building, the simplification intensity should be adjusted in time to ensure that the building characteristics are correctly reflected.

[0044] Next, the visual difference control process between the simplified three-dimensional financial building high-precision model and the original model is further refined by describing multiple steps.

[0045] First, the disparity V_disp is calculated at multiple angles to quantify the degree of detail loss. Here, disparity refers to the geometric difference between two 3D models at the same viewing angle. This step introduces multiple viewing angles to more accurately evaluate how much information is lost from the simplified 3D financial building model compared to the high-precision source model when viewed from various angles.

[0046] In one embodiment, for example, when generating a simplified digital 3D representation of a landmark financial building in the city center, it was found that the simplified model and the high-resolution original version had obvious contour deviations on some sides, that is, the disparity V_disp was not 0; this means that the user can perceive the difference between the two at that observation angle. In order to make the evaluation more systematic and quantitative, a series of measurements were performed after setting a sufficiently diverse set of viewing angles.

[0047] Next, a predetermined reference value V_std plus a parameter called the sensitivity adjustment factor η are used to jointly determine how to affect the changes in people's visual effects. This step is intended to establish a set of quantitative standards based on psychophysical principles - to determine an acceptable error limit. The reference value is the maximum allowable parallax value that is difficult for the human eye to detect under ideal conditions based on empirical values ​​or user satisfaction surveys; the sensitivity adjustment parameter η is used to correct the specific demand differences in different application scenarios. For example, public display purposes may tend to use a higher threshold to allow greater compression to speed up loading efficiency but ensure similar appearance; in areas that require more details, such as virtual roaming simulation training grounds, smaller values ​​will be used to ensure a high degree of authenticity and restore the original morphological characteristics.

[0048] Specifically, the expression S_adjust=(V_std-V_new) / η is used to dynamically calculate and adjust the relationship between the actual parallax distance target value after simplification of each specific observation direction, that is, the detail optimization index. In this formula, V_std represents the ideal parallax upper limit mentioned above, and its value range is roughly between 0 and 5 cm (for the building facade scale), depending on the designer's standard setting for the project's precision requirements; V_new refers to the new two-phase misalignment obtained after the angle currently being examined is simplified; as a factor appearing in the denominator, η can be regarded as a key variable used to adjust the steep slope of the curve or the smooth transition amplitude, and its value generally fluctuates between [0.1, 2]. It is usually recommended to use a position close to the low end such as 1 as the starting point to balance the accuracy loss while taking into account the computing resource overhead; the purpose of this setting is to construct a set of measurement and evaluation criteria that can reflect individual preferences and choices while not losing the scientific basis for guidance.

[0049] Finally, if the calculated result S_adjust exceeds the established adjustment limit adjust_limit, the modeling precision level should be appropriately increased to maintain the consistency of the final viewing experience of the end user, and all adjustments mentioned here shall not exceed the upper limit setting. In other words, once the tolerance limit is exceeded in any comparison scene, it is necessary to call back and do some reverse correction work to keep the entire output sequence in a reasonable and stable state. For example, when facing an ancient European bank building complex with a complex structure and a large number of carved decorative parts, the original historical atmosphere may be lost due to the excessive reduction of grid units, resulting in blurred patterns on the surface of the column. At this time, it is necessary to appropriately call back to increase the multi-faceted expression in order to reproduce those delicate and subtle details as much as possible without causing excessive data expansion. At the same time, in order to avoid falling into an endless loop or overfitting, the maximum amplitude of the adjustment factor is limited.

[0050] For example, in the specific implementation, if it is found that even a small change will still cause a significant difference that exceeds the allowable range, the most conservative method will be used to correct it until the condition is restored, and the adjustment factor will not increase indefinitely. Such restrictive measures ensure that the actual operability and economic cost considerations will not be ignored in the pursuit of perfection during the implementation of the plan.

[0051] Next, the following additional steps are described for the present invention: introducing a detail priority mechanism in the process of high-precision building modeling to enhance the realism of the model.

[0052] First, during the 3D reconstruction, a priority evaluation scheme was customized for the features of different building types, that is, a specific detail priority matrix A_priority was established based on the different treatment of financial institution buildings, general residential buildings, etc. Each element in this matrix represents the importance of a certain type of detail structure, such as window frames, exterior wall masonry textures, etc. (the value is usually set between 1 and 10, the closer to 1 means the feature is in a secondary position and the closer to 10 means the component is extremely critical), to reflect the importance weight of accurately expressing these components.

[0053] Secondly, after the initial construction is completed, we will turn to examine the reproduction effect of the above key parts in the final result, measure the preservation ratio corresponding to each importance level, and summarize it into a report card B_score reflecting the processing status. The score in this step depends on the quality of reconstruction. If a certain detail is perfectly reproduced, it can be awarded full marks. Otherwise, a lower score will be assigned according to the actual situation. Within this range, each value should have a reasonable upper and lower limit, and try to get as close as possible to the best score of ideal performance as a reference.

[0054] For example, in one embodiment, consider a large financial office building project. For such a complex modern building, the curtain wall system and sign design are particularly significant and important, so these two aspects will receive a higher w_i value. If the curtain wall is partially rebuilt and it is found that some areas have color deviations but the overall appearance is still representative, the corresponding position B_score can be around 8 or 9 points. If the font size of the sign is completely the same, then it can get a full score of 10.

[0055] Next, the total score of all scores is calculated according to B_priority_weight = Sum(A_priority .* B_score) to quantify the level of detail retention in the whole process; in this formula, ".*" represents the multiplication of two vectors to obtain a new result set, and then the specific comprehensive evaluation value is obtained by accumulation. The reason for this design is that we should not only pay attention to the quality of each individual detail, but also consider the overall visual effect of the combination of them. Each w_i in A_priority is regarded as a regulating factor to ensure that the most important part accounts for a higher proportion in the final score.

[0056] Finally, the total score of the retention of key features calculated in the previous article will be compared with the preset target baseline Standard. If the difference between the two exceeds the set tolerance delta_w, it is necessary to feedback and fine-tune the modeling parameters until the difference is minimized, that is, the target is reached or approached, and | B_priority_weight - standard_wt | ≤delta_w holds. This setting ensures that no matter how the optimization is carried out, the quality of other non-core elements will not be excessively sacrificed and the most critical components will not be ignored, and a good balance will be maintained. Through the process of continuous monitoring-evaluation-feedback in the entire closed loop, the finished product can be closer and closer to the original planning standard and the expressiveness of the part that users are concerned about can be improved as much as possible.

[0057] Next, the present invention is described. In the process of high-precision modeling of financial buildings, based on the 3D reconstruction technology, the maximum deviation Δv_max between mesh vertices is dynamically adjusted so as to ensure the visual fidelity of the model under different lighting environments.

[0058] First, the analysis is combined with the influencing factors of the environment under different light source conditions. This means identifying and quantifying the difference factors f(L, I) between the natural light L_nat and the indoor light I_in in the environment. For example, in one embodiment, when the morning sun shines obliquely and the fluorescent light shines vertically at night, the effects caused by the two are completely different, and it is necessary to consider how these variables jointly affect the final visual result. Specifically, L_nat ranges from zero (no sunlight) to thousands of lux (bright daytime); I_in may reach hundreds of lux and vary depending on the type of light source. The ideal situation is the best state when there are no obstructions.

[0059] Secondly, a benchmark is defined based on the above conditions - Δ_v_ideal, which is the maximum allowable deviation in an ideal state without any obstacles or covering. This value represents the maximum limit distance that the model vertex can move relative to the original position, maintaining structural integrity and accuracy without interference while providing sufficient flexibility for rendering.

[0060] Then establish the influence function H to reflect the actual effect of the comprehensive illumination (natural light L and indoor light I after the weighted combination of coefficients f and g) on ​​the maximum deviation. The f and g in the formula reflect the weight setting of the contribution ratio of different types of light sources to the overall illumination field; generally speaking, for indoor scenes, a greater importance may be given to the g parameter (for example, g can be between 0.6 and 0.9), while outdoor scenes may pay more attention to the role of f (f can be from 0.5 to 0.8). H inputs this information and calculates a maximum deviation Δv_adj that has been corrected in real time, which represents the upper limit of vertex displacement under the current lighting conditions.

[0061] The result is then compared to the previously set limit Δv_max: if the calculated Δv exceeds the preset limit, it is forced to be reduced to this value. The purpose is to ensure that no matter what the situation is (whether it is strong sunlight penetrating through the window or a dim underground vault), the model's performance can meet the requirements and present the building's characteristics without distortion. For example, when dealing with the interior of a large vault, due to the relatively closed space and insufficient light, it is necessary to strictly control the range of Δv value changes to achieve a more realistic presentation effect.

[0062] Next, the present invention is directed to a high-precision modeling method for financial buildings based on three-dimensional reconstruction, and in particular, describes the subdivision measures added during the processing in order to solve the problem of distinguishing between sharp corners and smooth boundaries during mesh optimization.

[0063] In the process of detecting and distinguishing boundaries, the algorithm mechanism is used to recognize and classify all edge types in the model. Once a sharp boundary edge is identified, a special boosting strategy will be applied to it, such as increasing the local density boost ratio P_boosted of this part to five times that of the normal situation, so as to ensure that the sharp features can be accurately reproduced and avoid the phenomenon of unnatural mutations in the smooth transition area; at the same time, for the non-boundary part that is not regarded as the boundary, the redundant points are reduced to optimize efficiency and save resources. Specifically, a reduction coefficient is used to adjust the number of generated nodes. The formula is P_decrease = P_base * ratio, where the parameter P_base represents the base point density, and ratio is the reduction factor. The default range is (0, 1], and the best choice depends on the actual needs. This ensures that the flat surface will not generate redundant vertices and edges due to excessively high sampling rates.

[0064] In order to achieve a harmonious transition between sharp edges and rounded contours, the mathematical logic of continuity is used to construct an effective interface identification scheme, which can correctly identify the location where the two types of surfaces contact and apply different optimization rules specific to each case to enhance the realism and delicacy of the final model.

[0065] Finally, to ensure that the original contour features are retained after mesh reprocessing, a set of evaluation indicators needs to be set to quantify the consistency of the comparison results, which will involve the step of calculating the shape similarity measure C_shape_similarity. Specifically, the calculation formula is C = \sqrt{\sum Di^2 / n}, where the denominator n represents the number of samples (usually refers to the number of patches or line segments involved in the selection for comparison), and Di represents the Euclidean distance difference between the corresponding elements. When the calculated C>limit, that is, it exceeds the given limit, it is prompted that the rationality of the existing refinement method and the relevant parameter settings should be carefully re-evaluated. In one embodiment, when applied to the reproduction of a famous bank headquarters building, it was found that some architectural structural features need to be highly restored, such as the carved patterns on the top of the column or the frame form of the exterior wall windows. These details are rich in distinct angle changes or clear contours, and this method is used to ensure that every detail is reproduced with high quality.

[0066] Next, the present invention is described. In the key steps of the high-precision modeling method for financial buildings, the key details are more precisely located and marked based on claim 9: First, the spatial positioning auxiliary tool set T_localization is used to find all geometric shapes S_unique and textures F_pattern with unique attributes, which means that the specially designed spatial positioning technology is applied to the three-dimensional model of the building to identify and record those unique components or textures that have a significant impact on the appearance and characteristics of the building.

[0067] In one embodiment, for example, there are complex decorative columns on the facade of a financial building. Their geometric features such as curvature radius are obviously different from other parts, and there may be carved patterns to increase the surface complexity. In this case, T_localization is used to accurately find these columns and carved areas.

[0068] Then, a flag bit array M_flags is set up to carry the position information of these elements, that is, through a binary or multi-bit array, each member represents whether a certain voxel is a unique geometric or pattern part previously determined.

[0069] For example, where the special cylindrical structure mentioned above is located, the corresponding point of the array will be set to a specific identifier (such as 1) so that subsequent operations can recognize it and give it special treatment.

[0070] The number of nodes to be enhanced is determined according to the formula P_special_points = Count(Find(T_localization*S_unique + T_localization*F_pattern)). The parameter S_unique is the unique geometric structure defined above; F_pattern is the texture line frame. The Count() function counts the total number of objects that meet the conditions; the range covers all units of the entire building; there is no fixed optimal value for this process, because the best effect depends on the characteristics of the building itself.

[0071] Finally, it is ensured that each stage of the simplification algorithm does not destroy the nodes that have been marked as important. At the same time, if errors occur, the adjacent triangular mesh structure is automatically adjusted to repair the deviation.

[0072] Specifically, taking a large-scale financial building modeling project as an example, ensuring that its classic door head and sculptural outline are not deformed during the process of simplifying the building's appearance is a manifestation of the effectiveness of this mechanism.

[0073] The high-precision modeling method of financial buildings based on three-dimensional reconstruction of the present invention includes the following detailed steps, and through innovative technical solutions, successfully solves the problem of how to adjust the parameter settings of the mesh optimization algorithm according to the characteristics of complex surfaces and detailed textures during the simplification of high-resolution models, avoiding the technical problem of key details being lost. The following is a detailed description of the entire technical solution: First, the method begins with extracting geometric and texture information from multi-view images of financial buildings. By using high-quality camera systems or multi-view stereo (MVS) technology and 3D scanning equipment to collect a large number of images or point cloud data from different perspectives, advanced feature matching algorithms and triangulation methods are used to accurately capture the structural form of various aspects of the building and the color of its exterior surface. This step is an important prerequisite for building a complete and detailed initial 3D model. In this process, it is ensured that the key feature points in each image are accurately located to establish the precise coordinate system required for subsequent modeling, and to generate a point set with sufficient density to represent the shape and color characteristics of the building's appearance.

[0074] Next, in order to obtain a more accurate and beautiful model, we further analyze these preliminary 3D object models based on the data obtained above. In particular, we carefully consider the small morphological changes in the complex curved structure parts such as decorative lines and sculpture reliefs. Because this part often contains important historical value or aesthetic elements that need to be restored and reproduced in the digital world and preserved for a long time. For this purpose, a feature detection subprocess designed specifically for such characteristics is used to identify and quantify its specific scale characteristics; for example, the curvature estimation tool is used to determine the degree of difference in surface smoothness, or the edge sharpening technology is used to enhance the recognition of contour line boundaries. According to the results of these specific feature analyses, the corresponding optimization strategy is established - this is the specific reference for determining how to set the next step of mesh adjustment parameters. This step fully considers the different geometric features of the model, so that it can make accurate estimates for the subsequent processing and reduce the risk of distortion caused by rough compression.

[0075] Then, the process enters the grid optimization stage. After the previous ring is completed, the grid density distribution is adjusted based on a relatively complete understanding of which areas need to be protected and which areas can be appropriately simplified. At this time, the set of parameter configurations that have been carefully evaluated and selected are used for operation. This set of rules is usually derived from empirical formulas accumulated through long-term research in the professional field combined with actual measurement data. They can help the computer automatically calculate and determine when dense connection points should be maintained in order to faithfully restore those subtle but important parts; conversely, when encountering relatively flat walls without special structures, redundant node connections can be boldly removed to reduce the overall file size without affecting the visual viewing experience, saving storage costs and improving subsequent computing efficiency.

[0076] Finally, even after completing the above series of operations, the work cannot be terminated directly, but the last and very necessary process is to continue to implement the degradation sampling processing method for the entire 3D image after re-trimming. What is said here is not as simple as randomly deleting the number of nodes. On the contrary, achieving moderate refinement while ensuring that the overall look and feel is not distorted is a task that is both full of artistic creativity and requires rigorous scientific calculations. On the one hand, with the help of the principle of topological connectivity, even the reduced graphics can still maintain the original topological relationship without causing defects such as cracks and deformation; on the other hand, the blank ratio should be reasonably planned according to the distance requirements of different observers so that the final result can show the best quality level whether it is macroscopically examined or locally enlarged, without leaving a too rigid impression. In addition, a variety of filtering processing methods are introduced to further enhance the rendering fidelity and make the material texture look exactly like the real object.

[0077] In summary, a high-precision modeling method for financial buildings based on 3D reconstruction always focuses on how to better present and maintain the most core and representative details of the target object while minimizing irrelevant redundant parts throughout the process, effectively overcoming the problem of detail loss and visual distortion caused by traditional methods, and bringing great advantages to cultural relics protection, virtual display and other application scenarios. Through this targeted method, it is largely guaranteed that even the 3D model of the building with reduced resolution can maintain sufficient recognition and real beauty, which has a far-reaching impact on an industry such as financial buildings that combines historical and cultural values ​​with technical practicality.

[0078] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present disclosure. It should be understood that the above description is only the specific implementation method of the embodiments of the present disclosure and is not intended to limit the protection scope of the embodiments of the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the protection scope of the embodiments of the present disclosure.

Claims

1. A high-precision modeling method for financial buildings based on three-dimensional reconstruction, characterized in that: include: Extract geometric information and texture information from multi-view images of financial buildings to form an initial 3D model; Analyze complex surfaces and detailed texture characteristics based on the extracted geometry and texture information, and determine the parameter settings of the mesh optimization algorithm; Optimize the mesh of the initial 3D model according to the determined parameter settings, retain key details and reduce redundant data; The optimized 3D model is simplified to reduce the model resolution while maintaining the integrity of the details.

2. A high-precision modeling method for financial buildings based on three-dimensional reconstruction according to claim 1, characterized in that: Based on the extracted geometry and texture information, the complex surface and detailed texture characteristics are analyzed, and the parameter settings of the mesh optimization algorithm are further determined, including: Get the proportion of complex surfaces in the image P_complex; Determine the mesh refinement coefficient K_initial; Adjust the mesh refinement factor K = K_initial * (P_complex + 1) based on the following formula to ensure that the detailed texture is not lost due to simplification; If K > K_max, then K is set to K_max, where K_max is the set maximum refinement coefficient.

3. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 1 is characterized in that: Based on the extracted geometry and texture information, the complex surface and detailed texture characteristics are analyzed, and the parameter settings of the mesh optimization algorithm are further determined, including: Calculate the initial detail density D_init and weight it according to the material distribution M of the building; Let the texture fidelity factor F_textural be a value in the range of 1 to 10; Use the formula C_optim = F_textural * (M / max(M)) to adjust the grid optimization threshold C_optim; If C_optim < threshold, increase the number of grid points to maintain key area details, where threshold is the lower limit.

4. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 3 is characterized in that: Further limitation based on adjusting the mesh refinement coefficient also includes the following steps: Estimate the impact of model simplification on the original complex surface details I_effect; The initial influence tolerance R0 is set to ensure that detail loss is minimized; The new tolerance R is calculated using the formula R = exp(|I_effect|); Adjust the K_refinement parameter according to the R value to ensure that the number of grids adapts to the changes in surface characteristics, K_refinement = max(R*K, min_refinement). Where min_refinement is the minimum refinement level.

5. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 4 is characterized in that: Also includes the steps: Analyze the key node density N_nodes in multiple locations of the simplified model; Set the expected retained node ratio R_expected; Apply N_preserve = N_original * R_expected to calculate the number of nodes to be kept; The simplified model error is evaluated according to the formula E_simplicity_loss = |N_preserved / N_expected 1|, and when this error is greater than the set upper limit, the amount of node removal is reduced to prevent shape distortion caused by over-simplification.

6. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 5 is characterized in that: The application of the formula is supplemented: The degree of detail loss is quantified by calculating the disparity V_disp between the simplified model and the original model at multiple angles; Given a reference value V_std and a sensitivity adjustment factor η, the intensity of the impact of parallax changes on user visual effects is controlled; Use the formula S_adjust = (V_std - V_new) / η to adjust the detail adjustment factor so that the appearance difference of the simplified model remains within the acceptable range for the human eye; If S_adjust > adjust_limit, the level of refinement is appropriately increased to ensure a consistent user experience. The upper limit of the adjustment factor is adjust_limit.

7. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 6 is characterized in that: The following additional steps are added: According to different building types, a specific detail priority matrix A_priority is set, in which each element reflects the importance weight w_i of a certain type of detail structure of that type; Measure the corresponding detail retention in the actual processing results and generate a score vector B_score; According to B_priority_weight = Sum(A_priority.*B_score), all scores are summed up to comprehensively evaluate the key retention level in the optimization process; Compare with the target standard value Standard. Once the difference is found to be beyond the allowable range, the correction parameters are fed back until it is close to the standard, that is, |B_priority_weight - standard_wt)|≤ delta_w.

8. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 7 is characterized in that: Combined with the lighting environment under different light source conditions, the maximum deviation Δv_max between mesh vertices is dynamically adjusted: Consider the difference factor f(L, I) of the total illumination distribution of natural light L_nat and indoor light I_in; Define the maximum deviation Δ_v_ideal reference value when there is no shadow in the ideal situation; Based on the comprehensive effect of f*L+ g*I illumination on Δ_v_ideal influence function H, the real-time corrected Δv_adj is obtained; Finally, it is judged that Δv = H(f*L, g*I; Δ_v_ideal). If Δv > Δv_max, Δv is reduced to not exceed Δv_max, thereby protecting the visual fidelity in complex environments.

9. The high-precision modeling method of financial buildings based on three-dimensional reconstruction according to claim 8 is characterized in that: Added subdivision measures to address the problem of different treatment of sharp corners and smooth edges during mesh optimization: Increase the local density boost ratio P_boosted to 5 times the normal level for edges identified as sharp boundaries; For non-boundary parts, reduce to normal ratio P_decrease = P_base * ratio to reduce redundancy; By using the dichotomy principle of smooth continuity and sharp discontinuity, a detection mechanism is constructed to identify the interface between the two and implement differentiated optimization rules; Ensure the consistency of the visual characteristics of the boundaries before and after optimization by calculating C_shape_similarity = sqrt(sum(Di^2) / n). This similarity coefficient is used as the criterion. When C>limit, the current configuration will be reviewed to achieve simplicity and sufficient descriptiveness.

10. The method for high-precision modeling of financial buildings based on three-dimensional reconstruction according to claim 9 is characterized in that: To more accurately locate key details and perform special marking: Introduce the spatial positioning auxiliary tool set T_localization to find all geometric shapes S_unq and textures F_pat with unique attributes; Establish a flag bit array M_flags to carry the location information of these key elements; According to the formula P_special_points = Count(Find(T_localization*S_unique + T_localization*F_pattern)); the number of special node sets to be enhanced is determined; Ensure that each subsequent round of simplification iteration does not affect the actual structural properties expressed by this set of special points. If a violation is found, immediately adjust the layout of the adjacent triangle meshes to make up for the potential defects.

Images