A three-dimensional steel bar reconstruction method based on a set constraint of packets

Abstract

The application discloses a three-dimensional steel bar reconstruction method based on grouped set constraints and belongs to the technical field of building engineering informatization, which comprises the following steps: extracting the geometric features of multi-view images to generate an initial three-dimensional Gaussian set; based on the similarity of the main direction of each Gaussian element in the set, the set is divided into a plurality of geometric grouped sets; a comprehensive loss function containing multiple geometric constraints such as luminosity, perception and linear continuity, intra-group parallelism, inter-group orthogonality and morphological proportion is constructed, and the grouped sets are iteratively optimized; and the optimized grouped sets are fused to generate a final three-dimensional entity model. The three-dimensional steel bar reconstruction method based on grouped set constraints provided by the application realizes high-precision and topologically consistent three-dimensional reconstruction of steel bar structures that are spatially dense and mutually occluded, and improves the structural accuracy and engineering application value of the reconstruction result.

Description

Technical Field

[0001] This invention relates to the fields of building engineering information technology and computer vision technology, and in particular to a three-dimensional rebar reconstruction method based on group set constraints. Background Technology

[0002] With the deep integration of Building Information Modeling (BIM), the Internet of Things (IoT), and computer vision technologies, 3D reconstruction technology is playing an increasingly important role in the digital transformation of the construction industry. 3D models acquired through multi-view images and laser scanning can record the complex spatial geometry of the construction site with high fidelity, providing crucial data support for project progress monitoring, quality acceptance, and final archiving. Slender rod-shaped components such as reinforcing bars, prestressed tendons, and anchor bolts are core load-bearing units in modern concrete and geotechnical engineering. Their three-dimensional spatial shape, relative position, and topological connections directly determine the safety and durability of the engineering structure. However, because these components are typically slender, small in size, densely arranged, and often located in complex, occluded environments, traditional contact-based, single-point manual measurement methods suffer from low efficiency, strong subjectivity, and susceptibility to environmental interference, failing to meet the demands of large-scale, high-precision, and traceable digital quality acceptance. Therefore, there is an urgent need to develop an automated and intelligent 3D reconstruction method to achieve high-precision, instance-level 3D model reconstruction of such slender components, providing a solid foundation for subsequent structural safety analysis and digital delivery.

[0003] Currently, the main technical approaches for 3D reconstruction of construction sites include: point cloud reconstruction based on multi-view stereo matching (MVS), implicit representation reconstruction based on neural radiation field (NeRF), and the recently emerging 3D Gaussian Splatting method. While traditional MVS methods can generate dense point clouds, they are prone to generating significant noise and voids due to feature matching failures when dealing with slender targets such as steel bars with simple surface textures and high reflectivity. This leads to fractures in the reconstructed rod models and compromises geometric continuity. NeRF and its variants, through end-to-end neural network optimization, have achieved significant breakthroughs in view synthesis quality. However, their implicit representation makes it difficult to precisely constrain the geometric structure of the scene, especially when dealing with a large number of repetitive slender rods. They are prone to generating incorrect topological connections due to spatial confusion, resulting in model sticking and morphological distortion.

[0004] The latest 3D Gaussian sputtering technology, by introducing anisotropic 3D Gaussian primitives as scene representations, has shown great potential in rendering efficiency and detail representation. However, existing Gaussian sputtering reconstruction methods mainly rely on photometric consistency loss during optimization, lacking the ability to explicitly model high-level structural relationships within the scene. When faced with complex steel cages composed of main bars, stirrups, and tie bars, although this method can generate visually coherent images, it cannot fundamentally understand and enforce strong engineering prior constraints such as "parallelism within a group" and "orthogonality between groups" between steel bars. Furthermore, for densely arranged parallel steel bar arrays, due to the lack of instance differentiation mechanisms and structured geometric constraints, the reconstruction results are prone to instances being misjudged, orientation errors, and topological adhesion. Therefore, existing technologies struggle to achieve high-precision, topologically consistent reconstruction of slender components such as steel bars in high-density, complex occlusion construction scenarios. An innovative reconstruction method that integrates geometric primitive representation, multi-task learning, and structured prior constraints is urgently needed to overcome these shortcomings. Summary of the Invention

[0005] The purpose of this invention is to provide a three-dimensional rebar reconstruction method based on group set constraints. By introducing the inherent geometric prior knowledge of the rebar structure as an explicit optimization constraint, it aims to achieve high-precision, topologically consistent three-dimensional reconstruction of spatially dense and mutually obstructing rebar cages or rebar meshes, thereby significantly improving the automation level and data reliability of building concealed works quality inspection.

[0006] To achieve the above objectives, this invention provides a method for reconstructing three-dimensional reinforcement based on grouped set constraints, comprising the following steps:

[0007] S1. Collect multi-view images of the rebar binding area from the construction site;

[0008] S2. Input the acquired multi-view images into a pre-trained image encoder. The role of the image encoder is to focus on extracting features related to 3D geometry and generate a set of geometric center features for subsequent 3D reconstruction.

[0009] S3. The geometric features extracted by the encoder are fed into the Gaussian decoder. The decoder predicts a corresponding 3D Gaussian element for each pixel or image patch, generating an initial Gaussian set containing tens of thousands of Gaussian elements. ,in For a single Gaussian unit, These represent the center coordinates, opacity, rotation, scaling, and color parameters of a single Gaussian element, respectively.

[0010] S4. For the i-th Gaussian element Its rotation parameters and scaling parameters Together, they determine their principal direction vector in space; using a clustering algorithm, the Gaussian set is divided into multiple geometrically grouped sets based on the principal direction similarity of Gaussian elements. ;

[0011] S5. In the optimization process, geometric prior knowledge of the steel structure is introduced, and linear continuity constraints, intra-group parallel consistency constraints, inter-group orthogonality constraints and shape proportion constraints are applied to the Gaussian group set mentioned in step S4 to force the Gaussian set to converge to the real and regular steel structure.

[0012] S6. Construct a total loss function and optimize iteratively to ensure that all parameters of the Gaussian set simultaneously satisfy rendering realism and geometric regularity. Calculate the fusion weights for the iteratively optimized Gaussian set and finally output a topologically consistent 3D reconstructed steel reinforcement model.

[0013] Preferably, in step S5, the linear continuity constraint applied to the Gaussian grouping set is specifically as follows:

[0014] To ensure that the center directions of two adjacent Gaussian elements are aligned with the local principal direction of the Gaussian reinforcement, the reinforcement maintains linear consistency along the principal axis:

[0015] ;

[0016] in, , The center coordinates of the adjacent Gaussians. For Gorsky Yuan Local principal direction.

[0017] Preferably, in step S5, the intra-group parallel consistency constraint applied to the Gaussian grouping set is specifically as follows:

[0018] Traverse the same group All Gaussian pairs within Calculations are performed to ensure that the principal directions of all Gaussian elements within the same geometric group are consistent:

[0019] ;

[0020] in, and These are the two principal directions of the Gaussian unit.

[0021] Preferably, in step S5, the orthogonal constraint between groups applied to the Gaussian grouping set is specifically as follows:

[0022] Apply orthogonal (perpendicular) constraints to different geometric groups (i.e., different reinforcing bars) that may intersect in space:

[0023] ;

[0024] in, and These are the two principal directions of Gaussian elements from different groups.

[0025] Preferably, in step S5, the inter-group morphological ratio constraint applied to the Gaussian grouping set is specifically as follows:

[0026] Ensure that the shape of each generated Gaussian element conforms to the slender physical form of a steel bar:

[0027] ;

[0028] in, For Gorsky Yuan The three-dimensional scale vector, This is a preset proportional constant.

[0029] Preferably, in step S6, a total loss function is constructed, specifically as follows:

[0030] The calculation combines photometric loss, perceptual loss, and the weighted sum of constructed geometric constraints. By adjusting the weights, the final 3D steel reinforcement model is ensured to retain both the true details of the photograph and the accurate topological structure.

[0031] ;

[0032] in, For loss of photometric uniformity, For perceptual consistency loss based on learned features, each These are the weighting coefficients.

[0033] Preferably, the fusion weights in S6 Based on directional consistency and local visibility, Gaussian fusion is further performed according to the group topology relationship to generate a topologically consistent 3D steel reinforcement model.

[0034] Therefore, the present invention employs the above-mentioned three-dimensional reinforcement reconstruction method based on group set constraints, which has the following beneficial effects:

[0035] (1) Strong structural consistency and accurate topological relationship. In view of the structural characteristics of steel reinforcement arrangement, an innovative grouping method based on directional characteristics and geometric constraints of inter-group orthogonality and intra-group parallelism are proposed. It can directly introduce the structural prior knowledge in engineering design into the reconstruction optimization process, and force the model to maintain the orthogonal and parallel topological relationship of steel reinforcement mesh or steel reinforcement cage in macroscopic way, reducing the topological confusion and model adhesion problems caused by repetitive structures in existing technologies.

[0036] (2) Good geometric continuity and high model integrity. To address the problem of slender rods being prone to breakage during reconstruction, a linear continuity constraint was designed. This constraint acts on the interior of a single steel bar instance, ensuring that the geometric primitives constituting the rod are arranged closely and continuously along the principal axis, effectively eliminating the model breakage caused by weak textures or occlusion, and significantly improving the geometric integrity of the reconstructed model.

[0037] (3) It balances realism and structural accuracy. The optimization objective not only includes multiple geometric constraints to ensure structural accuracy, but also introduces photometric loss. With perceived loss The dual visual constraints ensure that the reconstruction results not only conform to engineering specifications in terms of topology, but also visually reproduce the real texture details such as rust and binding wires on the surface of the steel bars, achieving a unity of engineering precision and visual realism.

[0038] (4) It has strong adaptability to engineering sites. The core is to use powerful geometric priors for constraints, rather than simply relying on fragile image texture features. Therefore, it has stronger robustness to unfavorable conditions such as complex lighting, shadow changes and partial occlusion that are common in construction sites, and can achieve robust and reliable three-dimensional reconstruction in more complex real environments.

[0039] (5) Excellent scalability and application value: The present invention outputs an instance-level, structured 3D model, rather than an unordered point cloud; each steel reinforcement instance can be independently identified and parameterized, so that the reconstruction results can be easily compared with the reinforcement of the Building Information Model (BIM) or seamlessly integrated into the digital twin platform, providing a high-quality data foundation for subsequent automated acceptance of engineering quality, structural safety simulation analysis and full life cycle management.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a three-dimensional reinforcement reconstruction method based on group set constraints, as an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of directional grouping clustering in step S4 of an embodiment of the present invention;

[0043] Figure 3 This is a schematic diagram illustrating intra-group parallelism and inter-group orthogonal constraints in an embodiment of the present invention;

[0044] Figure 4 This is a comparison chart of the experimental results of the method of the present invention and the traditional method. Detailed Implementation

[0045] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0046] Please see Figure 1 A method for reconstructing three-dimensional reinforcement based on grouped set constraints, comprising:

[0047] S1. Multi-view image acquisition: Collect multi-view images of the rebar binding area to be reconstructed from the construction site as input data for the reconstruction algorithm.

[0048] S2, Geometric Feature Extraction: The acquired multi-view images are input into a pre-trained image encoder to extract depth features related to the three-dimensional geometric shape in each view image.

[0049] S3. Initial Gaussian Set Generation: The geometric features extracted in step S2 are input into the Gaussian decoder to predict the corresponding 3D Gaussian primitives for pixels or image patches in the input image, thereby generating an initial, unordered 3D Gaussian set. , Wherein is a single Gaussian element. These represent its center coordinates, opacity, rotation, scaling, and color parameters, respectively.

[0050] S4. Geometric Grouping: Group the initial Gaussian set generated in step S3. For the i-th Gaussian element... According to its rotation parameters and scaling parameters Determine its principal direction vector. Using a clustering algorithm, based on the similarity of the principal direction vectors, divide the initial Gaussian set into multiple geometrically grouped sets. Each group set This corresponds to a single, independent rebar instance.

[0051] S5. Constraint Optimization and Model Generation: A comprehensive loss function is constructed, incorporating photometric loss, perceptual loss, and multiple geometric constraint losses. The parameters of all Gaussian elements in the Gaussian set are iteratively optimized and updated until the loss function converges. The final output is a geometrically continuous and topologically consistent 3D rebar reconstruction model. Multiple geometric constraints include: linear continuity constraints, intra-group parallelism constraints, inter-group orthogonality constraints, and morphological proportion constraints.

[0052] ;

[0053] Linear continuity constraints Defined as:

[0054] in, For the neighborhood set of Gaussian elements, Let the coordinates be the center coordinates of two Gaussian elements within the neighborhood. For Gorsky Yuan The local principal direction. This constraint forces the Gaussian centers to align along the principal axis by minimizing the cross product of the vector connecting adjacent Gaussian centers and the local principal direction vector, thus maintaining the geometric continuity of individual reinforcement bars.

[0055] Intragroup parallel consistency constraints Defined as:

[0056] ;

[0057] in, and For the same geometric grouping set The principal direction of any two Gaussian elements within the same group. This constraint ensures that all Gaussian elements constituting the same reinforcing bar have consistent directions by minimizing the cross product of direction vectors within the same group, preventing model distortion.

[0058] Intergroup orthogonal constraints Defined as:

[0059] ;

[0060] in, and They belong to different geometric grouping sets. This constraint forces spatially adjacent rebar instances to maintain a perpendicular relationship by minimizing the dot product of direction vectors between different groups, in order to comply with engineering design specifications.

[0061] ;

[0062] Form proportion constraints Defined as:

[0063] in, For Gorsky Yuan The three-dimensional scale vector, and These are the scales of its shortest and longest axes, respectively. This is a pre-defined slender proportional constant that conforms to the physical morphology of steel bars. This constraint forces each Gaussian element to approximate a slender rod shape in terms of morphology, ensuring the geometric accuracy of the reconstructed model.

[0064] ;

[0065] Comprehensive loss function for:

[0066] in, This is the photometric consistency loss, used to ensure pixel-level similarity between the reconstructed model's rendered image and the input image; This is a perceptual consistency loss based on learned features, used to improve the texture realism of rendered images; For the aforementioned geometric constraint loss term; each These are the weighting coefficients for each type of loss.

[0067] S6. Gaussian Fusion and Model Generation: The optimized grouped Gaussian sets from step S5 are fused to generate the final continuous entity model. This step is performed based on the group topology, using a fusion weight determined by directional consistency and local visibility. The discrete Gaussian elements are fused into a continuous three-dimensional density field, and the surface mesh is extracted from it, ultimately outputting a geometrically continuous and topologically consistent three-dimensional steel reinforcement reconstruction model.

[0068] The overall process is as follows:

[0069] First, in step S1, an image sequence of the rebar binding area at the construction site is acquired from different perspectives using an industrial camera or handheld device. Then, in steps S2 and S3, the image sequence is input into a neural network consisting of an image encoder and a Gaussian decoder to generate an initial, unordered set of discrete three-dimensional Gaussian units, which preliminarily represents the three-dimensional geometry of the scene.

[0070] Next, in the core step S4, refer to Figure 2 As shown, the initial Gaussian set is geometrically grouped. By analyzing the rotation and scaling parameters of each Gaussian element, its principal direction vector in 3D space is extracted. Then, clustering algorithms such as DBSCAN are used to group Gaussian elements with similar principal directions into the same geometric group set, thereby organizing the disordered point cloud into multiple sets representing independent reinforcement instances.

[0071] Then, in step S5, the grouped Gaussian set is iteratively optimized. (Refer to...) Figure 3 As shown, in the optimization process, in addition to the standard photometric and perceptual losses, four key geometric constraint losses are introduced: linear continuity constraint to ensure that the Gaussian centers within each group are aligned in a straight line to prevent breakage; intragroup parallel consistency constraint to ensure that the principal directions of all Gaussian elements within the same group are aligned to prevent distortion; intergroup orthogonality constraint to force different spatially adjacent groups to maintain a perpendicular relationship; and morphological proportion constraint to ensure that each Gaussian element presents a slender shape that conforms to the physical morphology of steel bars.

[0072] Finally, in step S6, after the comprehensive loss function converges, Gaussian fusion is performed on the optimized and regularly arranged discrete Gaussian set. This step smoothly fuses Gaussian elements belonging to the same group into a continuous entity with a surface, based on the directional consistency of each Gaussian element (as a quality weight) and spatial local visibility (as a distance weight). By performing this operation on all groups and handling the topological relationships at the connections between groups, a high-fidelity, topologically consistent 3D reinforced concrete solid model is finally generated.

[0073] Example 1:

[0074] In one specific embodiment, the method of the present invention can be applied to the three-dimensional reconstruction of complex rebar-binding areas such as beam-column joints. Operators can use a handheld camera or a fixed multi-camera array to acquire multi-view image sequences of the joint (step S1) and input them into the reconstruction system described in this invention.

[0075] The system will execute steps S2 to S6. Compared with conventional reconstruction methods (such as conventional Gaussian sputtering) that do not apply the geometric constraints described in this invention, the final 3D model generated by this invention (step S6) exhibits significant advantages in theory and performance because this invention explicitly introduces linear continuity constraints and inter-group orthogonality constraints in the optimization process (step S5).

[0076] Specifically, this means that the model breakage phenomenon commonly caused by weak textures or occlusion in conventional methods will be effectively suppressed (thanks to...). Meanwhile, the cross-topological relationship (especially the orthogonal relationship) that should exist between the main reinforcement and the stirrups can be correctly maintained (thanks to...). This fundamentally overcomes the problems of model adhesion, structural confusion, and geometric errors that are easily generated by conventional methods, referring to... Figure 4 As shown, this invention can generate high-quality models with rigorous structure and consistent topology.

[0077] Example 2:

[0078] In another embodiment, this embodiment is applied to the automated and digital acceptance of steel reinforcement cages for subway shield tunnel segments. First, on the segment production line, the method of this embodiment is used to quickly reconstruct the tied steel reinforcement cages in three dimensions, generating a three-dimensional solid model with centimeter-level accuracy.

[0079] Then, the reconstructed model is registered and deviation analyzed in three dimensions with the preset BIM (Building Information Modeling) design model in the same coordinate system. By setting a deviation threshold (e.g., 5 mm), the system can automatically highlight areas in the reconstructed model where the actual position, spacing, or diameter of the reinforcing bars does not match the design, and can quickly identify missing or incorrectly tied reinforcing bars, automatically generating a digital acceptance report containing three-dimensional views and deviation data.

[0080] Therefore, the present invention adopts the above-mentioned three-dimensional rebar reconstruction method based on group set constraints. By comparing the high-precision reconstruction model with the design model, it realizes the automated, data-driven and visual acceptance of the hidden rebar project, which greatly improves the detection efficiency and the reliability of quality control.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for reconstructing three-dimensional reinforcement bars based on grouped set constraints, characterized in that, include: S1. Collect multi-view images of the rebar binding area from the construction site; S2. Input the acquired multi-view images into a pre-trained image encoder. The image encoder is used to extract features related to 3D geometry and generate geometric center features for 3D reconstruction. S3. The geometric features extracted by the encoder are fed into the Gaussian decoder. The Gaussian decoder predicts a corresponding 3D Gaussian element for each pixel or image patch, generating an initial Gaussian set containing several Gaussian elements. ,in For a single Gaussian unit, These represent the center coordinates, opacity, rotation, scaling, and color parameters of a single Gaussian element, respectively. S4. For the i-th Gaussian element Its rotation parameters and scaling parameters The principal direction vectors of Gaussian elements in space are jointly determined; a clustering algorithm is used to divide the Gaussian set into multiple geometrically grouped sets based on the similarity of the principal directions of Gaussian elements. ; S5. In the optimization process, geometric prior knowledge of the steel structure is introduced, and linear continuity constraints, intra-group parallel consistency constraints, inter-group orthogonality constraints and morphological proportion constraints are applied to the Gaussian group set in step S4. S6. Construct a total loss function and optimize iteratively to ensure that all parameters of the Gaussian set simultaneously satisfy rendering realism and geometric regularity. Calculate the fusion weights for the iteratively optimized Gaussian set and finally output a topologically consistent 3D reconstructed steel reinforcement model.

2. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 1, characterized in that, In step S5, the linear continuity constraint applied to the Gaussian grouping set is as follows: To ensure that the center directions of two adjacent Gaussian elements are aligned with the local principal direction of the Gaussian reinforcement, the reinforcement maintains linear consistency along the principal axis: ； in, For the neighborhood set of Gaussian elements, , The center coordinates of adjacent Gaussian elements, For Gorsky Yuan Local principal direction.

3. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 2, characterized in that, In step S5, the intra-group parallel consistency constraint applied to the Gaussian grouping set is as follows: Traverse the same group All Gaussian pairs within Calculations are performed to ensure that the principal directions of all Gaussian elements within the same geometric group are consistent. The parallel consistency constraint is as follows: ； in, and They are the same geometric grouping set The principal directions of any two Gaussian elements within the domain.

4. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 3, characterized in that, In step S5, the orthogonality constraint between groups applied to the Gaussian grouping set is as follows: Apply orthogonal constraints to distinct geometric groups that intersect in space: ； in, and Different geometric grouping sets The two principal directions of Gaussian elements.

5. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 4, characterized in that, In step S5, the inter-group morphological ratio constraint applied to the Gaussian grouping set is as follows: Ensure that the shape of each generated Gaussian element conforms to the slender physical form of a steel bar: ； in, For Gorsky Yuan The three-dimensional scale vector, This is a preset proportional constant. and These are the scales of the shortest and longest axes of the Gaussian element, respectively.

6. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 5, characterized in that, In step S6, a total loss function is constructed, specifically as follows: The weighted sum of photometric loss, perceptual loss, and constructed geometric constraint terms is calculated by adjusting the weights: ； in, For loss of photometric uniformity, For perceptual consistency loss based on learned features, each These are the weighting coefficients for each type of loss.

7. The three-dimensional reinforcement reconstruction method based on group set constraints according to claim 6, characterized in that: Fusion weights in S6 Based on directional consistency and local visibility, Gaussian fusion is further performed according to the group topology relationship to generate a topologically consistent 3D steel reinforcement model.

Images