Engineering field three-dimensional reconstruction physical scale estimation method with reference to standard component

Abstract

The invention discloses an engineering field three-dimensional reconstruction physical scale estimation method with reference to a standard component, and belongs to the technical field of building engineering digitization, three-dimensional reconstruction and intelligent construction, and the method comprises the steps: constructing a three-dimensional reconstruction physical scale estimation platform; a multi-view image sequence containing a standard component is collected, a scale-free three-dimensional point cloud and a camera pose are obtained through structure self-motion reconstruction SfM and multi-view stereo matching MVS reconstruction, a point cloud subset of the standard component is detected and positioned, a reconstruction scale size is obtained through geometric fitting, a scale factor is calculated in combination with a real physical size, and a three-dimensional point cloud is obtained. And carrying out global scaling on the scale-free model to finally obtain a real physical scale three-dimensional model. The method is suitable for various image acquisition devices and construction scenes, the scale recovery precision reaches the millimeter-to-centimeter level, the robustness is high, and the method can be widely applied to engineering measurement, BIM model alignment, digital twinning construction and other scenes and has remarkable engineering application value.

Description

Technical Field

[0001] This invention relates to the fields of digital construction engineering, three-dimensional reconstruction and intelligent construction technology, and in particular to a method for estimating the physical scale of three-dimensional reconstruction of engineering sites based on reference standard components. Background Technology

[0002] In the digitalization of construction engineering, using ordinary RGB cameras to acquire multi-view images and generating 3D point clouds or mesh models of scenes through SfM and MVS technologies has become the mainstream method for 3D modeling of engineering sites. This method boasts advantages such as low equipment cost, ease of operation, and wide adaptability, making it compatible with various image acquisition devices including surveillance cameras, mobile phones, ordinary cameras, and drones. It is widely used in basic scenarios such as engineering site visualization and progress recording. However, due to the inherent limitations of imaging principles, the above 3D reconstruction process can only preserve the projective geometric relationships of the images and cannot directly obtain the true physical scale of the scene, resulting in an inherent "scale uncertainty" problem. The reconstructed 3D model can only reflect the relative geometry of the scene, and its size has an unknown proportional relationship with the true physical size. This deficiency severely limits the in-depth application of this technology in core engineering scenarios with clear requirements for scale accuracy, such as engineering surveying, component inspection, and digital twin alignment.

[0003] To address the scale uncertainty issue in 3D reconstruction, existing technologies have developed various approaches, but all have significant limitations. One approach relies on placing targets, rulers, or artificial objects of known length in the scene, or using pre-placed calibration boards combined with manual measurements to provide a scale reference. These methods are not only cumbersome but also interfere with dynamic construction scenes, affecting construction efficiency. Furthermore, manually placed targets and calibration boards are easily moved, obstructed, or damaged in complex construction environments, making continuous and stable scale calibration difficult. Another approach integrates data from multiple sensors, such as LiDAR, depth cameras, and inertial measurement units (IMUs), utilizing the sensors' own depth measurement capabilities to obtain scale information. However, this approach requires additional high-end sensors, significantly increasing equipment costs, and is unsuitable for conventional engineering sites equipped only with ordinary cameras or surveillance videos. Some approaches attempt to infer the scene scale based on the identification of specific objects of known size in specific scenes. However, these methods are limited to specific scenes and objects, lacking adaptability to general engineering sites and failing to meet the diverse needs of construction scenarios.

[0004] It is worth noting that construction sites naturally contain a large number of standardized components. These components, as fundamental construction materials, possess significant characteristics such as fixed dimensions, wide distribution, clear image features, and stable three-dimensional geometry. Examples include scaffolding steel pipes (common outer diameters of 48mm or 48.3mm), wooden formwork (common sizes are 1220mm×2440mm and 915mm×1830mm), steel formwork (common sizes are 1800mm×900mm, 1500mm×900mm, and 1200mm×600mm), and plastic formwork. Their actual physical dimensions are clear and uniform, and they are widely present throughout the construction process, serving as natural "scale benchmarks." However, current technologies have not fully explored and utilized this unique resource advantage of construction sites. There is a lack of a technical solution capable of automatically identifying these standardized components and accurately restoring the scale of three-dimensional reconstruction based on them, leading to limitations in common multi-view... Figure 3 The engineering application value of reconstruction technology is greatly limited. Summary of the Invention

[0005] The purpose of this invention is to propose a physical scale estimation method for three-dimensional reconstruction of engineering sites based on reference standard components to solve the problems mentioned in the background technology: to break through the dependence of existing technologies on additional equipment and manual operation, improve the stability, automation and accuracy of scale estimation in dynamic construction environments, and provide reliable technical support for scenarios such as engineering site measurement, progress monitoring and digital twin alignment.

[0006] To achieve the above objectives, this invention provides a method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, comprising the following steps: S1: At the construction site, a three-dimensional reconstruction physical scale estimation platform is built, including an image acquisition module, a three-dimensional reconstruction module, a standard component detection module, a point cloud component extraction module, a geometric fitting module, a scale calculation module, and a model scaling module to realize the three-dimensional reconstruction physical scale estimation. S2: Acquire a sequence of images containing multiple views of the engineering site through the image acquisition module. The image sequence includes at least one standard component. S3: The 3D reconstruction module performs Structural Self-Motion Reconstruction (SfM) and Multi-View Stereo Matching (MVS) on the image sequence to obtain scale-free 3D reconstruction results, including 3D point clouds, camera pose, and scene geometry models. S4: The standard component is detected by the standard component detection module; the detection results are located in the 3D point cloud by the point cloud component extraction module based on the camera pose, and the point cloud subset of the standard component is extracted by combining cross-view geometric consistency constraints. S5: Perform geometric model fitting on a subset of point clouds using the geometric fitting module to obtain the reconstructed dimensions of standard components at the reconstruction scale; S6: Using the scale calculation module, calculate the scale factor required to restore the scaleless model to the real physical scale based on the actual physical size and the reconstructed size of the standard component; S7: Through the model scaling module, the scale factor is used to globally scale the scaleless 3D point cloud and camera pose to obtain a 3D model with a real physical scale. S8: Align the real physical scale 3D model with the Building Information Model (BIM) and perform least squares point cloud registration or component matching methods to achieve accurate integration of the real physical scale 3D model and the Building Information Model (BIM).

[0007] Preferably, in step S2, the image sequence consists of clear images from any viewpoint, ensuring sufficient overlap between multiple views to guarantee the effectiveness of reconstruction, and includes at least one standard component to ensure that the visual features of the standard component are clear in the image, avoiding severe occlusion or blurring; the standard component refers to an object with uniformly specified physical dimensions used on the engineering site, including scaffolding steel pipes, reinforcing bars, standard templates, or square timber.

[0008] Preferably, in step S3, the structural self-motion reconstruction (SfM) and multi-view stereo matching (MVS) reconstruction process for the acquired multi-view image sequence specifically includes: the execution of SfM uses the scale-invariant feature transform algorithm SIFT, the accelerated robust feature algorithm SURF, or the ORB algorithm to extract image feature points and complete matching, estimates the camera pose using the bundle adjustment method to generate a sparse point cloud, and then performs dense matching using the MVS algorithm to obtain a dense point cloud and a scene geometric model; the three-dimensional point cloud includes sparse point cloud and dense point cloud, and the camera pose includes camera intrinsic parameters and extrinsic parameters.

[0009] Preferably, in step S4, the detection of standard components adopts a deep learning-based target detection model and geometric shape detection using edge features, straight line features, or circular features.

[0010] Preferably, in steps S4 and S5, adaptive sampling based on point cloud density is performed on the point cloud subset to reduce the impact of noise points on geometric fitting. When the point cloud subset contains outliers or occluded points, a robust fitting method is used, including Random Sample Consensus Algorithm (RANSAC), Huber loss fitting, or weighted least squares fitting. Cross-view geometric consistency constraints refer to the constraint rules for filtering 3D point clouds based on multi-view projection geometry and camera pose. The core is that the point cloud points corresponding to the same standard component must meet the following conditions: the corresponding point clouds after back-projection from different views into 3D space fall within the same spatial region; the constructed point cloud subset should show continuous distribution characteristics in 3D space; and the overall shape should be consistent with the preset geometric characteristics of the standard component. Through constraints, background points and noise points can be eliminated, and the point cloud subset of the standard component can be accurately extracted.

[0011] Preferably, in step S5, the geometric model fitting adopts RANSAC-based cylindrical model fitting or cross-sectional circular fitting, and the fitting parameters include the cylinder axis direction vector, cylinder radius and cylinder center point; the fitting process adds structural prior constraints to the standard component.

[0012] Preferably, in step S6, the actual physical size of the standard component is the actual diameter of the corresponding standard component, and the reconstructed size of the standard component is the reconstructed diameter of the standard component obtained through cylindrical fitting, with a scale factor. Calculate using the following formula: ; in, The actual diameter of the standard component. The reconstructed diameter of the standard component; When multiple standard components exist, the scale factors calculated from each standard component are fused using a weighted least squares method to obtain the optimal global scale factor. The following formula is used to obtain it: ; in, For the first The reconstructed diameter of a standard component, This is the actual diameter of the standard component. As weight.

[0013] Preferably, in step S7, global scaling refers to multiplying the three-dimensional coordinates of all points in the scale-free three-dimensional point cloud by a scale factor, and multiplying the translation vectors in all camera poses by a scale factor, ultimately obtaining a three-dimensional model with a true physical scale consistent with the actual size of the engineering site.

[0014] Therefore, the present invention employs the above-mentioned method for estimating the physical scale of a three-dimensional reconstruction of an engineering site using a reference standard component, which has the following advantages: (1) No additional hardware investment required, lower application cost: Make full use of the standard components naturally existing on the engineering site as the scale benchmark. There is no need to set up targets, rulers and other manual auxiliary tools, nor is there a need for high-end sensors such as lidar and depth cameras. Scale restoration can be achieved by relying only on ordinary image acquisition equipment, which greatly reduces hardware costs and on-site deployment difficulty, and is compatible with the equipment configuration of conventional engineering sites. (2) High degree of automation and adaptable to dynamic construction environment: From image acquisition, scale-free reconstruction, component detection and three-dimensional positioning, to geometric fitting, scale factor calculation and model scaling, the whole process is automated. There is no need for manual participation in component identification, measurement or target placement, which avoids the errors and inefficiencies caused by manual operation. It is not affected by the dynamic changes in the construction scene and can achieve continuous and stable scale estimation. (3) Strong robustness and outstanding anti-interference ability: The robust algorithms such as RANSAC-based cylindrical fitting and weighted least squares can effectively resist the influence of interference factors such as point cloud noise, component occlusion, and outliers. At the same time, standard components are widely distributed on the engineering site and have stable geometric structures, which further ensures the stability and reliability of the scale estimation results and is suitable for complex construction environments. (4) Wide adaptability and flexible application scenarios: It is compatible with various image sources such as surveillance video, drone images, and mobile phone images, which can meet the image acquisition needs of different construction scenarios; the standard components cover a variety of types such as scaffolding steel pipes, steel bars, and formwork, which can adapt to the scale restoration needs of different construction stages and different scenarios, and have strong versatility. (5) High scale accuracy to meet core engineering needs: Through precise geometric fitting algorithms and multi-component fusion strategies, the scale recovery accuracy can reach the millimeter to centimeter level, which can accurately match core engineering scenarios with high accuracy requirements such as engineering measurement, component size detection, and BIM model alignment, providing reliable data support for engineering quality control, schedule management and digital transformation. (6) Significant engineering value, promoting deep integration of technologies: This invention breaks through the limitations of ordinary multi-view Figure 3 Overcoming the scale bottleneck of 3D reconstruction, this technology deeply integrates computer vision with intelligent construction, enabling 3D reconstruction to be directly applied to key engineering scenarios such as construction progress monitoring, safety management, and digital twin construction. This provides crucial technical support for the digital and intelligent development of engineering and has significant industrial application value.

[0015] 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

[0016] Figure 1 This is a flowchart illustrating the steps of a method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as mentioned in an embodiment of the present invention. Figure 2 This is a comparison of the experimental results of a method for estimating the physical scale of a reference standard component in a three-dimensional reconstruction of an engineering site, as mentioned in an embodiment of the present invention, and a traditional method. Figure 3 This is a schematic diagram of cylindrical model fitting in the geometric fitting of a standard component in a method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as mentioned in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present 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 present invention without inventive effort are within the scope of protection of the present invention.

[0018] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0019] Example like Figure 1-3 As shown, this invention provides a method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, comprising the following steps: S1: Construct a 3D reconstruction physical scale restoration platform, including an image acquisition module, a 3D reconstruction module, a standard component detection module, a point cloud component extraction module, a geometric fitting module, a scale calculation module, and a model scaling module.

[0020] S2: Multi-view image acquisition: Using existing image acquisition devices at the construction site, such as surveillance cameras, handheld mobile phones, ordinary digital cameras, or drones, a multi-view image sequence containing the area to be reconstructed at the construction site will be collected. The image sequence must consist of clear images from any perspective, ensuring sufficient overlap between the multiple views to guarantee the effectiveness of the reconstruction, and must include at least one standard component (selected from at least one of scaffolding steel pipes, wooden formwork, steel formwork, and plastic formwork), ensuring that the component's visual features are clear in the image and avoiding severe occlusion or blurring. For video source data, a suitable image sequence can be generated automatically by extracting frames at fixed frame intervals or based on the degree of scene change.

[0021] S3: Scale-free 3D reconstruction (executed by SfM / MVS): The SfM and MVS reconstruction process is performed on the acquired multi-view image sequences: First, the scale-invariant feature transform (SIFT), speed-up robust features (SURF), or oriented fast and rotated BRIEF (ORB) algorithms are used to extract image feature points and complete matching. The camera intrinsic and extrinsic parameters are estimated by bundle adjustment to generate a sparse point cloud. Then, based on the sparse point cloud and the camera pose, dense matching is performed by the MVS algorithm to obtain a dense point cloud or mesh model.

[0022] Alternatively, existing mature 3D reconstruction models (such as Depth Anything 3) can be used to perform 3D reconstruction on the input multi-view image sequence: by utilizing the model’s cross-view attention mechanism and depth-ray joint prediction capability, spatially consistent geometric structures can be recovered from the multi-view images, and scale-free sparse / dense point clouds, camera intrinsic and extrinsic parameters can be directly generated.

[0023] The final result is a scale-free 3D reconstruction, including sparse / dense point clouds, camera intrinsic and extrinsic parameters, and the shape of the scene reconstruction model.

[0024] S4: Standard component inspection and 3D positioning: First, standard component detection is performed on the image. This can be achieved using object detection models such as the YOLO series, semantic segmentation models such as Faster R-CNN and Mask R-CNN, or a combination of geometric shape detection methods such as edge detection and line detection, automatically identifying the component's location, category, bounding box, and detection confidence. Then, 3D localization is performed. Based on camera intrinsic and extrinsic parameters and a projection model, the detected component regions in the image are back-projected into a 3D point cloud. Combining multi-view cross-view geometric consistency constraints, point cloud points belonging to the same component are selected, background and noise points are removed, and a subset of the point cloud corresponding to the component is extracted. Optionally, density-based adaptive sampling can be performed on the point cloud subset to optimize the efficiency and accuracy of subsequent fitting.

[0025] S5: Geometric fitting of standard components: The appropriate fitting method is selected based on the standard component shape: For cylindrical components such as scaffolding steel pipes, a cylindrical model fitting based on the RANSAC algorithm is used to resist interference from outliers and occlusion points, and the fitting yields the axial direction vector and radius. For parameters such as the cross-section of the component, the radius can be obtained by circular fitting using the least squares method. Calculate the component diameter at the reconstructed scale based on the fitted radius. During the fitting process, prior structural constraints (such as the axis aligning with the direction of gravity) can be incorporated, or robust methods such as Huber loss fitting can be used to optimize the results.

[0026] S6: Scale factor calculation: Based on the detected component type, query the pre-stored standard component physical dimension database to obtain the actual physical diameter. ;based on and Through formula Calculate the scale factor If multiple standard components are detected, a weighted least squares or RANSAC robust estimation method is used, through optimization formulas. .(in The scaling factor is calculated comprehensively based on the fitting error or detection confidence level. This improves reliability.

[0027] S7: Global scaling of 3D models: The translation vectors in the scale-free 3D point cloud, mesh model, and camera extrinsic parameters are uniformly multiplied by the calculated scale factor. This allows for global scaling, ultimately resulting in a true physical-scale 3D model that matches the actual dimensions of the engineering site.

[0028] S8: Optional model applications: Align the real-scale 3D model with BIM and achieve precise fusion using least-squares point cloud registration or component matching methods; the entire process can be executed in real time by edge computing devices, GPU servers or cloud systems to realize online physical scale restoration of the construction site.

[0029] A specific implementation process is as follows: like Figure 1 As shown, an embodiment of the present invention uses a scaffolding steel pipe with a diameter of 48 mm as a standard component. A method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component includes the following steps: S1: Construct a 3D reconstruction physical scale restoration platform, including an image acquisition module, a 3D reconstruction module, a standard component detection module, a point cloud component extraction module, a geometric fitting module, a scale calculation module, and a model scaling module.

[0030] S2: Multi-view image acquisition: Video data of the scaffolding area was collected using four 2K resolution network surveillance cameras at the construction site. One minute of clear, unobstructed video was selected, and 30 multi-view images were obtained by extracting frames at 2 seconds per frame. The acquisition requirements were: the overlap area of ​​adjacent images should be 72%, and the 48mm diameter scaffolding steel pipes in each image should be clearly visible without serious obstruction, with a pixel ratio of ≥5%.

[0031] S3: Scale-free 3D reconstruction: The SfM / MVS process was performed on the 30 acquired images: First, the SIFT algorithm was used to extract feature points on each image, and cross-view feature matching was completed through the Fast Library for Approximate Nearest Neighbors (FLANN) matcher; then, the camera intrinsic and extrinsic parameters were estimated by the bundle adjustment method to generate the corresponding sparse point cloud; finally, based on the sparse point cloud and the camera pose, dense matching was completed through the PatchMatch MVS algorithm to obtain a dense point cloud that meets the standards for the next step of implementation.

[0032] S4: Standard component inspection and 3D positioning: First, the YOLOv8 model is used to detect scaffolding steel pipes in the image, and the bounding box, category and detection confidence are output. Then, based on the camera intrinsic parameters, extrinsic parameters and pinhole projection model, the steel pipe region in the image is back-projected to a dense point cloud. Combined with the cross-view geometric consistency constraint of three adjacent views, the point cloud points of the same steel pipe are selected. After removing the background points, a subset of steel pipe point clouds is obtained. Finally, the subset is adaptively sampled according to density to retain the core feature points.

[0033] S5: Geometric fitting of standard components: For the cylindrical shape of scaffolding steel pipes, the RANSAC cylindrical fitting algorithm is used to fit and obtain the axial direction vector and reconstruction radius of the steel pipe. The reconstructed diameter was calculated. .

[0034] S6: Scale factor calculation: Query the pre-stored standard component physical dimension database to obtain the actual diameter of the 48mm scaffolding steel pipe. ; through formula Calculate the scaling factor .

[0035] S7: Global scaling of 3D models: The scale-free dense point cloud, mesh model, and translation vector in the camera extrinsic parameters obtained from S2 are all multiplied by a scale factor. We obtained a 3D model of the actual physical scale that corresponds to the actual size on site at a 1:1 ratio (the diameter of the steel pipe point cloud is approximately 48mm after scaling).

[0036] S8: Optional model applications: The ICP point cloud registration algorithm was used to align the real-scale 3D model with the BIM design model of the area (scaffolding steel pipe diameter 48mm, spacing 1.5m). After registration, the position error was <0.8mm and the size error was <0.2mm. The model was then fused for construction progress verification.

[0037] This implementation uses 48mm diameter scaffolding steel pipes as a benchmark to achieve physical scale estimation for three-dimensional reconstruction of the engineering site, with a scale error of <0.2%, meeting the accuracy requirements of building construction.

[0038] This invention, by using naturally occurring 48mm diameter scaffolding steel pipes at the construction site as a reference, effectively solves the problems of existing multi-view systems without additional hardware investment through a complete technical path of "multi-view acquisition - scale-free reconstruction - component detection and positioning - geometric fitting - scale calibration". Figure 3 This invention addresses the issue of scale loss in 3D reconstruction while overcoming drawbacks such as occlusion of manual targets, high cost of high-end sensors, and insufficient reliability of single-component scale estimation. It minimizes the scale error of the 3D reconstruction model, meeting the stringent accuracy requirements of the construction industry. This invention can be widely applied to construction progress monitoring, scaffolding safety inspection, BIM model alignment, and digital twin modeling, and is particularly suitable for complex scenarios such as high-rise buildings and large construction sites. With the development of intelligent construction technology, this method will become a key supporting tool for the digital transformation of engineering projects, driving the construction industry towards precision and intelligence.

[0039] 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 estimating the physical scale of a three-dimensional reconstruction of an engineering site using a reference standard component, characterized in that, Includes the following steps: S1: At the construction site, a three-dimensional reconstruction physical scale estimation platform is built, including an image acquisition module, a three-dimensional reconstruction module, a standard component detection module, a point cloud component extraction module, a geometric fitting module, a scale calculation module, and a model scaling module to realize the three-dimensional reconstruction physical scale estimation. S2: Acquire a sequence of images containing multiple views of the engineering site through the image acquisition module. The image sequence includes at least one standard component. S3: The 3D reconstruction module performs Structural Self-Motion Reconstruction (SfM) and Multi-View Stereo Matching (MVS) on the image sequence to obtain scale-free 3D reconstruction results, including 3D point clouds, camera pose, and scene geometry models. S4: The standard component is detected by the standard component detection module; the point cloud component extraction module locates the detection results to the 3D point cloud based on the camera pose, and extracts the point cloud subset of the standard component by combining cross-view geometric consistency constraints. S5: Perform geometric model fitting on a subset of point clouds using the geometric fitting module to obtain the reconstructed dimensions of standard components at the reconstruction scale; S6: Using the scale calculation module, calculate the scale factor required to restore the scaleless model to the real physical scale based on the actual physical size and the reconstructed size of the standard component; S7: Through the model scaling module, the scale factor is used to globally scale the scaleless 3D point cloud and camera pose to obtain a 3D model with a real physical scale. S8: Align the real physical scale 3D model with the Building Information Model (BIM) and perform least squares point cloud registration or component matching methods to achieve accurate integration of the real physical scale 3D model and the Building Information Model (BIM).

2. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site using a reference standard component, as described in claim 1, is characterized in that: In step S2, the image sequence consists of clear images from any viewpoint, ensuring sufficient overlap between multiple views to guarantee the effectiveness of reconstruction. It includes at least one standard component to ensure that the visual features of the standard component are clear in the image and to avoid severe occlusion or blurring. The standard component refers to an object with uniformly specified physical dimensions used on the engineering site, including scaffolding steel pipes, reinforcing bars, standard templates, or square timber.

3. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site using a reference standard component, as described in claim 1, is characterized in that: In step S3, the Structural Self-Motion Reconstruction (SfM) and Multi-View Stereo Matching (MVS) reconstruction processes are specifically performed on the acquired multi-view image sequences. Specifically, the SfM process uses the Scale Invariant Feature Transform (SIFT), the Accelerated Robust Feature Transform (SURF), or the ORB algorithm to extract image feature points and complete matching. The camera pose is estimated using the bundle adjustment method to generate a sparse point cloud. Then, the MVS algorithm is used for dense matching to obtain a dense point cloud and a scene geometric model. The 3D point cloud includes sparse and dense point clouds, and the camera pose includes camera intrinsic and extrinsic parameters.

4. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as described in claim 1, is characterized in that: In step S4, the detection of standard components adopts a deep learning-based target detection model and geometric shape detection using edge features, straight line features, or circular features.

5. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as described in claim 1, is characterized in that: In steps S4 and S5, adaptive sampling based on point cloud density is performed on the point cloud subset to reduce the impact of noise points on geometric fitting. When the point cloud subset contains outliers or occluded points, robust fitting methods are used, including Random Sample Consensus Algorithm (RANSAC), Huber loss fitting, or weighted least squares fitting. Cross-view geometric consistency constraints refer to the constraint rules for filtering 3D point clouds based on multi-view projection geometry and camera pose. The core is that the point cloud points corresponding to the same standard component must meet the following conditions: the corresponding point clouds after back-projection into 3D space from different views fall within the same spatial region; the constructed point cloud subset should show continuous distribution characteristics in 3D space; and the overall shape should be consistent with the preset geometric characteristics of the standard component. Through constraints, background points and noise points can be eliminated, and the point cloud subset of the standard component can be accurately extracted.

6. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site using a reference standard component, as described in claim 1, is characterized in that: In step S5, the geometric model fitting adopts RANSAC-based cylindrical model fitting or cross-sectional circular fitting. The fitting parameters include the cylinder axis direction vector, cylinder radius, and cylinder center point. The fitting process adds structural prior constraints to the standard component.

7. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as described in claim 1, is characterized in that: In step S6, the actual physical size of the standard component is the actual diameter of the corresponding standard component, and the reconstructed size of the standard component is the reconstructed diameter of the standard component obtained through cylindrical fitting, with a scale factor. Calculate using the following formula: ； in, The actual diameter of the standard component. The reconstructed diameter of the standard component; When multiple standard components exist, the scale factors calculated from each standard component are fused using a weighted least squares method to obtain the optimal global scale factor. The following formula is used to obtain it: ； in, For the first The reconstructed diameter of a standard component, This is the actual diameter of the standard component. As weight.

8. The method for estimating the physical scale of a three-dimensional reconstruction of an engineering site based on a reference standard component, as described in claim 1, is characterized in that: In step S7, global scaling refers to multiplying the 3D coordinates of all points in the scale-free 3D point cloud by a scale factor, and multiplying the translation vectors in all camera poses by a scale factor, ultimately obtaining a 3D model with a true physical scale consistent with the actual size of the engineering site.

Images