A method and system for comparing building structural components for engineering supervision

Abstract

This invention relates to a method and system for comparing building structural components in engineering supervision, specifically within the field of building engineering supervision technology. The purpose of this invention is to quickly and accurately process noise from on-site measurement data, precisely identify building components, compare them with design drawings, and achieve automated matching of building components, thereby improving the efficiency and accuracy of engineering supervision. Key technical points include: Point cloud data acquisition: Using LiDAR to scan the construction area from multiple angles to acquire high-precision point cloud data; Point cloud data denoising: Employing a denoising-diffusion-recovery model to remove discrete points from the initial point cloud data in an end-to-end manner, further improving the quality of the point cloud data; Building component identification: Slicing the denoised point cloud data and using an improved YOLOv8 algorithm to detect and identify relevant building components; Sequence matching: Applying a similarity-based sequence matching method to match the sequence of building components with the sequence of components to be detected in the design drawings, achieving automated comparison. This invention can be well applied in the engineering supervision process.

Description

Technical Field

[0001] This invention belongs to the field of building engineering supervision technology, and relates to a method and system for comparing building structural components for engineering supervision. Background Technology

[0002] Currently, engineering supervision mainly relies on manual measurement and verification of various components during the construction process. However, this method suffers from low efficiency and susceptibility to errors caused by human factors. In recent years, the rapid development of high-precision measurement technologies such as lidar has provided a technological foundation and new development opportunities for achieving automated and high-precision matching of building components in the engineering supervision process.

[0003] To comprehensively improve the efficiency of safety supervision at construction sites and effectively reduce accident risks, various intelligent engineering supervision safety system solutions have been proposed in existing research. For example, patent CN118279833A discloses an engineering supervision safety inspection system based on three-dimensional visualization. This system mainly includes a visualization model construction unit, used to acquire point cloud data within the construction area and, combined with stereoscopic vision technology, construct a three-dimensional engineering visualization model; an engineering geographic location planning unit, which uses images of the construction area to calculate the distance between each location point and the ground, achieving area division and identifying potential risk areas; an engineering operation safety judgment unit, which uses drones to collect real-time images of construction personnel operating within risk areas, identifies and analyzes their operational behavior to determine the degree of safety violations; and a personnel safety risk prediction unit, which combines the visualization model and the degree of safety violations to predict the probability of accidents involving construction personnel and generates corresponding supervision and control instructions. This system improves the automation and intelligence level of safety supervision at construction sites while effectively monitoring the wearing of safety protective equipment by construction personnel, significantly reducing the risk of accidents during construction.

[0004] Patent CN117874900B proposes a method for supervising building construction projects based on BIM (Building Information Modeling) technology. This method mainly includes the following steps: First, point cloud data of the building is collected and divided into several empty and non-empty voxels. Then, based on the laser reflection intensity at multiple angles within the non-empty voxels, the anomaly degree of each 3D spatial point is calculated. Further, based on the anomaly degree and spatial coordinates, the initial reference point for each non-empty voxel is determined, and the initial reference point for each empty voxel is also obtained. Next, based on the initial reference point, the voxel, and its normal direction, the influence coefficient of each non-empty voxel on other voxels is calculated. Finally, based on the influence coefficient and the initial reference point, a complete BIM model is constructed. This method effectively reduces the offset problem that may occur during voxel downsampling and can reasonably fill in missing reference points in the point cloud data, thereby improving the completeness and accuracy of the generated 3D model. However, this technology does not discuss how to eliminate noise in the construction site measurement data or the identification of building components. Summary of the Invention

[0005] The technical problem to be solved by this invention is: the purpose of this invention is to provide a method and system for comparing building structural components for engineering supervision, so as to quickly and accurately process the noise of the measurement data at the construction site, accurately identify the building components, compare them with the design drawings, realize the automated matching of building components, and improve the efficiency and accuracy of engineering supervision.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] A method for comparing building structural components for engineering supervision, the implementation process of which is as follows:

[0008] Step 1. Point cloud data acquisition: Use LiDAR to scan the construction area from multiple angles to acquire high-precision point cloud data;

[0009] Step 2. Point cloud data denoising: The initial point cloud data is denoised using a denoising-diffusion-restoration model in an end-to-end manner;

[0010] Step 3. Construction of Building Component Sequence: The denoised point cloud data is sliced ​​and identified using an improved YOLOv8 algorithm to identify the monitored building components and their relative positions, thus forming a building component sequence.

[0011] Step 4. Construction of the sequence of components to be detected: For paper drawing files, the attribute information of each component is marked by manual annotation to form the sequence S of components to be detected;

[0012] Step 5. Sequence Matching: Apply a similarity-based sequence matching method to match the building component sequences. The system matches the sequence S of the components to be inspected in the design drawings to achieve automated comparison. Based on the sequence matching results, a matching report of the building components is generated.

[0013] Furthermore, the point cloud data denoising process is as follows: First, the 3D point cloud data is encoded into structured 2D latent features using the PointNet network. Then, the obtained 2D latent features are processed using a diffusion-denoising method to obtain clean latent features. Finally, a FoldingNet-based decoder is used to reconstruct high-quality 3D point cloud data from the denoised features.

[0014] Furthermore, the building component matching process is as follows: First, the denoised point cloud data is horizontally sliced ​​and the sliced ​​data is mapped to a Cartesian coordinate system xOy to form a building floor plan; second, the improved YOLOv8 is used to identify building components on the obtained building floor plan and record their relative positions.

[0015] Furthermore, the improved YOLOv8 is as follows: For the YOLOv8 Backbone framework, a spatial attention (SAM) module is added between each Conv and C2f module to enhance the ability of feature extraction to perceive spatial information; in addition, a coordinate attention (Coord Attention) module is added after the last convolutional Conv2d layer of the YOLOv8 decoupled Head module to capture long-distance spatial information of building components.

[0016] Furthermore, the sequence matching process is as follows: first, the sequence of building components is... First, align the length of the sequence with the component to be detected, S, and use an empty set as the padding segment; second, for the sequence... Perform pairwise matching with elements in S, take the highest similarity as the matching result, and use the Hungarian algorithm to optimize and obtain a unique mapping.

[0017] Furthermore, the sequence of building components The sequence S of the components to be detected is obtained as follows: For building component sequences, an improved YOLOv8 is used for detection. For any element... in These represent the label information, relative size information, and relative position information of the building components, respectively. For the sequence of components to be detected, the elements on the engineering drawings are manually marked one by one to form the corresponding elements s∈S. s={l,m,p}, where l,m,p represent the label information, relative size information, and relative position information of the component to be retrieved, respectively.

[0018] Furthermore, the pairwise matching method involves performing multi-level matching and fusion of element pairs in the building component sequence and the component sequence to be detected. The specific calculation can be formally represented as follows: in These represent the label information, relative size information, and relative position information of the building component, respectively; l, m, and p represent the label information, relative size information, and relative position information of the component to be retrieved, respectively; α is a balancing factor used to balance the weights of category information and other parameters.

[0019] A building structure component comparison system for engineering supervision is provided. The system has program modules corresponding to the steps of the above-mentioned technical solution. When the program modules are run, they execute the steps in the building structure component comparison method for engineering supervision.

[0020] The present invention has the following beneficial technical effects:

[0021] This invention comprises point cloud data acquisition, point cloud data denoising, building component detection, and sequence matching. Utilizing equipment such as LiDAR, point cloud data is acquired through multi-angle scanning. Addressing the noise issue in point cloud data, a point cloud data denoising method based on a diffusion model is proposed. Furthermore, the target recognition algorithm is improved by introducing spatial attention and coordinate attention modules to achieve accurate identification of building components. Finally, a matching algorithm is used to automatically match various components in the construction project with those on the design drawings. This invention provides a novel solution for engineering supervision in the construction field, enabling rapid and accurate matching of on-site measurement data with design drawings, achieving automated supervision, and thus improving the efficiency and accuracy of engineering supervision. Attached Figure Description

[0022] This invention can be more thoroughly understood by referring to the description provided below in conjunction with the accompanying drawings. These drawings, together with the following detailed description, constitute an important part of this specification, not only intuitively illustrating the technical framework and implementation process of the invention, but also further explaining the core principles and technical advantages of the invention through specific preferred embodiments. Through these drawings and descriptions, the implementation process of the technical solution of this invention can be clearly understood.

[0023] Figure 1 This is a flowchart of a method for comparing building structural components for engineering supervision, as described in this invention.

[0024] Figure 2 Framework diagram for denoising point cloud data;

[0025] Figure 3 A model diagram for identifying building components;

[0026] Figure 4 To display the matching graph of the matching process.

[0027] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0028] like Figures 1 to 4 As shown, the implementation process of the building structure component comparison method for engineering supervision according to the present invention will be explained in detail.

[0029] This invention provides a method for comparing building structural components in engineering supervision. Based on point cloud data, it matches building components, quickly and accurately processing noise from on-site measurement data to precisely identify building components and compare them with design drawings. This achieves automated matching of building components, improving the efficiency and accuracy of engineering supervision. The specific technical solution is as follows:

[0030] Step 1. Point cloud data acquisition: Use LiDAR to scan the construction area from multiple angles to acquire high-precision point cloud data;

[0031] Step 2. Point cloud data denoising: The initial point cloud data is denoised using a denoising-diffusion-restoration model in an end-to-end manner;

[0032] Step 3. Construction of the building component sequence: The denoised point cloud data is sliced ​​and identified using an improved YOLOv8 algorithm to identify building components and their relative positions, thus forming a building component sequence.

[0033] Step 4. Construction of the sequence of components to be detected: For paper drawing files, the attribute information of each component is marked by manual annotation to form the sequence S of components to be detected;

[0034] Step 5. Sequence Matching: Apply a similarity-based sequence matching method to match the building component sequences. The system matches the sequence S of the components to be inspected in the design drawings to achieve automated comparison. Based on the sequence matching results, a matching report of the building components is generated.

[0035] The following are the specific implementation steps of the present invention:

[0036]

Step 1

[0037] Using a laser point cloud device, the architectural space of the component to be measured is scanned to obtain three-dimensional point cloud data P containing noise. noisy .

[0038] [Step Two] Point Cloud Data Noise Reduction

[0039] To address point cloud data denoising, a point cloud denoising method based on the Denoising Diffusion Restoration Models (DDRM) is designed, which mainly includes point cloud data feature representation, DDRM denoising, and point cloud data reconstruction.

[0040] ① Point cloud data feature representation

[0041] Using the PointNet encoder based on a point cloud deep network, 3D point cloud data is encoded into 2D latent features. The encoding process preserves the geometric and topological information of the point cloud through multi-level local feature aggregation and global pooling, and can be formally represented as follows:

[0042] Z = Encoder(P) noisy ),

[0043] Among them, P noisy The input is noisy 3D point cloud data, and Z is a 2D latent feature.

[0044] ②DDRM noise reduction

[0045] DDRM mainly consists of two processes: a diffusion process and a denoising process. The diffusion process involves iteratively adding a certain amount of random noise to each round of diffusion. The denoising process does the opposite; in each round of denoising, a deep neural network learns how to remove noise and restore the original structure of the data. For the acquired two-dimensional latent feature Z, the DDRM denoising process can be formally represented as:

[0046] x0 = DDRM(Z),

[0047] Where x0 is the denoised potential two-dimensional feature.

[0048] ③ Point cloud data reconstruction

[0049] The reconstruction of point cloud data primarily utilizes a FoldingNet-based network to reconstruct high-quality 3D point clouds from denoised latent 2D features x0. The decoder, through multi-scale upsampling and geometric refinement modules, restores the surface details and density distribution of the point cloud, which can be formally represented as:

[0050] P clean =Decoder(x0),

[0051] Among them, P clean This is the reconstructed 3D point cloud data.

[0052]

Step 3

[0053] The construction of building component sequences is mainly used to identify building-related components, and mainly includes two parts: point cloud data slicing and component identification.

[0054] ① Point cloud data slicing

[0055] For the reconstructed high-quality 3D point cloud data P clean The data is projected onto a two-dimensional Cartesian coordinate system (xOy) using horizontal slicing to form the building plan I. Theoretically, for three-dimensional point cloud data P... clean By controlling the height of the horizontal slices, multiple building floor plans can be generated. Therefore, point cloud data slices can be formally represented as follows:

[0056] I = f xoy (f hor (P clean ,θ hor ),θ xoy ),

[0057] Among them, f xoy and f hor Two-dimensional mapping function and horizontal slicing function, θ xoy and θ hor These are the corresponding mapping parameters and slice height, respectively.

[0058] ② Component identification

[0059] To capture each building component in plan view I, given the object detection capabilities of YOLOv8 and considering the characteristics of the building components, an improved YOLOv8 (IYOLOv8) is employed for component detection. Specifically, the improved YOLOv8 adds a spatial attention (SAM) module between each Conv and C2f module in the YOLOv8 backbone framework to enhance the feature extraction's ability to perceive spatial information. Additionally, a coordinate attention module is added after the last convolutional Conv2d layer of the YOLOv8 decoupled head module to capture long-range spatial information of the building components. Therefore, the building component detection process can be represented as follows:

[0060]

[0061] in, Building components obtained through IYOLOv8 detection, θ iyolo These are the corresponding parameters. Additionally, it's important to note that the IYOLOv8 model used in the experiment was pre-trained from manually annotated architectural component floor plans, which better suits IYOLOv8's ability to detect architectural components.

[0062]

Step 4

[0063] For paper drawings, the attribute information of each component is marked manually. Specifically, the relative position information p, dimension information m, and category information l of the component are obtained from the drawing information. Then, all manually marked components are placed into the component sequence S to be detected.

[0064]

Step 5

[0065] Sequence matching is mainly used to determine the degree of matching between actual building components and those shown in architectural drawings. Specifically, for any sequence of building components... Given sequences s∈S, the similarity between them is calculated using the following formula:

[0066]

[0067] in Let l, m, and p represent the label information, relative size information, and relative position information of the building component, respectively; l, m, and p represent the label information, relative size information, and relative position information of the component to be retrieved, respectively; α is a balancing factor used to balance the weights of category information and other parameters; for a sequence... Perform pairwise matching with all elements in S, and take the one with the highest similarity as the matching result. At the same time, use the Hungarian algorithm for optimization to obtain a unique mapping.

[0068] Verification has shown that the method proposed in this invention solves the technical problem raised in this invention. Simulation experiments and practical applications have verified the technical effects and practicality claimed by this invention.

[0069] This invention proposes a method for comparing building structural components for engineering supervision. This method (algorithm) forms the core technology of this invention, and various products can be derived from it. Based on this method, a building structural component comparison system for engineering supervision is developed using a programming language. This system has program modules corresponding to the steps of the above-described technical solution. When the program modules run, they execute the steps of the aforementioned method for comparing building structural components for engineering supervision. The developed system (software) computer program is stored on a computer-readable storage medium. The computer program is configured to implement the steps of the aforementioned method for comparing building structural components for engineering supervision when called by a processor. In other words, this invention is materialized on a carrier, becoming a computer program product.

[0070] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0071] The computational programs (also referred to as programs, software, software applications, or code) of this invention include machine instructions of a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0072] It should be understood that the various processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, they are all within the protection scope of this invention.

Claims

1. A method for comparing building structural components for engineering supervision, characterized in that, The implementation process of the method is as follows: Step 1. Point cloud data acquisition: Use LiDAR to scan the construction area from multiple angles to acquire high-precision point cloud data; Step 2. Point cloud data denoising: The initial point cloud data is denoised using a denoising-diffusion-restoration model in an end-to-end manner; Step 3. Construction of Building Component Sequence: The denoised point cloud data is sliced ​​and identified using an improved YOLOv8 algorithm to identify the building components of the monitored building and their relative positions, thus forming a building component sequence. ; Step 4. Construction of the sequence of components to be inspected: For paper drawings, manually annotate the attribute information of each component to form the sequence of components to be inspected. ; Step 5. Sequence Matching: Apply a similarity-based sequence matching method to match the building component sequences. The sequence of components to be tested in the design drawings The system performs matching to achieve automated comparison and generates a matching report of building components based on the sequence matching results. The point cloud data denoising process is as follows: First, the 3D point cloud data is encoded into structured 2D latent features using the PointNet network. Then, the obtained 2D latent features are processed using a diffusion-denoising method to obtain clean latent features. Finally, a FoldingNet-based decoder is used to reconstruct high-quality 3D point cloud data from the denoised features. The improvements to YOLOv8 are as follows: For the YOLOv8 Backbone framework, a spatial attention (SAM) module is added between each Conv and C2f module to improve the ability of feature extraction to perceive spatial information; in addition, a coordinate attention (Coord Attention) module is added after the last convolutional Conv2d layer of the YOLOv8 decoupled Head module to capture long-distance spatial information of building components. The pairwise matching method involves multi-level matching and fusion of element pairs between the building component sequence and the component sequence to be detected. The specific calculation is formally represented as follows: ,in , , These respectively represent the label information, relative size information, and relative position information of the building components. These represent the label information, relative size information, and relative position information of the component to be retrieved, respectively. This is a balancing factor used to balance the weight of label information with other parameters.

2. The method for comparing building structural components for engineering supervision according to claim 1, characterized in that, The building component matching process is as follows: First, the noise-reduced point cloud data is horizontally sliced, and the sliced ​​data is mapped to the Cartesian coordinate system xOy to form a building floor plan. Secondly, the improved YOLOv8 is used to identify building components in the acquired building floor plan and record their relative positions.

3. The method for comparing building structural components for engineering supervision according to claim 1, characterized in that, The sequence matching process is as follows: First, the sequence of building components is matched. and the sequence of components to be detected First, perform length alignment, using an empty set as padding segments; second, for the sequence... and The elements in the dataset are matched pairwise, and the matching result with the highest similarity is taken. The Hungarian algorithm is then used to optimize the matching to obtain a unique mapping.

4. The method for comparing building structural components for engineering supervision according to claim 3, characterized in that, Building component sequence and the sequence of components to be detected The acquisition method is as follows: For building component sequences, an improved YOLOv8 is used for detection; for any element... , ,in , , These represent the label information, relative dimension information, and relative position information of the building components, respectively. For the sequence of components to be inspected, each element on the engineering drawings is manually marked to form the corresponding elements. , ,in These represent the label information, relative size information, and relative position information of the component to be retrieved, respectively.

5. A building structural component comparison system for engineering supervision, characterized in that: The system has a program module corresponding to the steps of any one of claims 1-4. When the program module is run, it executes the steps in the method for comparing building structural components for engineering supervision.

Images