A shear wall recognition method based on recombination segmentation graph

Abstract

The present application relates to the technical field of shear wall identification, in particular to a shear wall identification method based on recombination segmentation graphics, which obtains indoor wall images from different points through an image acquisition device, and synchronously obtains ranging data and positioning data; processes the images through a convolutional neural network, automatically calibrates the wall connection and performs segmentation, combines the ranging data to calculate the height and length of the single wall, and maps the positioning and orientation data obtained when the wall image is acquired in a two-dimensional matrix. Through geometric shape analysis and traversal of the two-dimensional matrix by the shear wall structure library, the shear wall candidate single wall is identified. Cluster analysis of the spatial position and connection relationship of adjacent walls determines the shear wall candidate area, and generates a recombination graphic. The recombination graphic is analyzed by a semantic segmentation algorithm, and whether it conforms to the shear wall characteristics is judged by a BP neural network to reduce misjudgment and improve identification accuracy. Finally, the identification result is displayed through three-dimensional visualization, which is convenient for engineers to analyze.

Description

Technical Field

[0001] This invention relates to the field of shear wall recognition technology, and more particularly to a shear wall recognition method based on reconstructed segmented graphics. Background Technology

[0002] Shear walls are critical load-bearing components in buildings, especially in high-rise buildings and seismic design. They are responsible for bearing vertical loads and resisting horizontal seismic forces, ensuring the stability of the building structure. The design and layout of shear walls not only directly affect the overall structural safety of the building but also play a vital role in seismic performance. In practical engineering, the correct identification of shear walls is of great significance for the design, analysis, and construction safety of building structures.

[0003] In interior renovation and remodeling scenarios, the identification and preservation of shear walls become particularly important. As load-bearing walls, shear walls cannot be arbitrarily demolished. If they are mistakenly demolished or their existence is overlooked during renovation, and they are not properly preserved, the building structure may be vulnerable to damage. Therefore, accurately and quickly identifying and properly preserving shear walls during interior renovation is a crucial step in ensuring the safety and sustainability of buildings. However, traditional shear wall identification methods often rely on manual marking, determining the location of shear walls through manual examination of architectural drawings and on-site surveys. Such methods are highly dependent on human experience, time-consuming, and prone to errors, especially in complex interior environments where the characteristics of shear walls often resemble other walls, making quick and accurate differentiation difficult. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a shear wall recognition method based on reconstructed segmented graphics. By combining image acquisition and positioning / orientation data with deep learning algorithms, the method automatically identifies shear walls, effectively automating shear wall recognition, improving recognition accuracy, avoiding reliance on human experience, and increasing recognition efficiency.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A shear wall identification method based on reconstructed segmented graphics includes the following steps:

[0007] S1. Acquire several indoor wall images and distance measurement data from different points or orientations using image acquisition equipment, and simultaneously collect positioning and orientation data when acquiring indoor wall images;

[0008] S2. Use a convolutional neural network to mark the wall connection points in the indoor image, divide the wall connection points into several individual wall graphics, and calculate the height and length of the individual wall based on the distance measurement data.

[0009] S3. Initialize the two-dimensional matrix and, based on the height and length of the individual walls and the positioning and orientation data when the wall images were acquired, map several individual walls into the two-dimensional matrix to obtain the indoor planar matrix.

[0010] S4. Based on the indoor plane matrix, candidate shear wall areas are identified through geometric morphology analysis. Individual wall graphics are selected and recombined according to the candidate shear wall areas to obtain recombined graphics. The semantics of the recombined graphics are analyzed based on the semantic segmentation algorithm and used as input to the BP neural network to output the shear wall recognition result.

[0011] S5. Perform 3D visualization based on the identification results of the shear wall.

[0012] Furthermore, the step of calibrating wall joints in indoor images using a convolutional neural network includes the following steps:

[0013] The collected images of the interior walls are input into the trained convolutional neural network model;

[0014] Edge feature points in the wall image are extracted by convolutional layers and downsampled by pooling layers;

[0015] The extracted features at the wall joints are calibrated using fully connected layers in a convolutional neural network.

[0016] Furthermore, the training of the convolutional neural network includes the following steps:

[0017] Collect several indoor wall images of unlabeled wall joints, manually label the wall joints, generate corresponding labeled images, and form a pair of training sample sets;

[0018] Based on the visual characteristics of the wall connection, the network structure of convolutional layers, pooling layers and fully connected layers is customized, and the network weight parameters are initialized.

[0019] Unlabeled interior wall images are input into the convolutional neural network model, and corresponding labeled images are used as supervision signals. The network parameters are iteratively updated using the backpropagation algorithm and loss function.

[0020] Furthermore, the calculation of the height and length of a single wall based on the distance measurement data includes the following steps:

[0021] Extract the horizontal distance between the individual wall and the image acquisition device, the distance between the vertical endpoint of the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device from the ranging data.

[0022] The height of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the vertical endpoint of the individual wall and the image acquisition device, respectively.

[0023] The length of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device.

[0024] Further, S3 includes the following steps:

[0025] Based on the height and length data of individual walls, the dimensions of each individual wall are normalized.

[0026] Based on the orientation of the image acquisition device and the spatial orientation of the individual wall, a sub-two-dimensional matrix for mapping is determined, and the matrix coordinates of the individual wall coverage are determined based on the normalized length data of the individual wall.

[0027] Based on the matrix coordinates of the individual wall coverage, the height of the normalized individual wall is mapped to the matrix element values ​​in a two-dimensional matrix to obtain the indoor planar matrix.

[0028] Furthermore, the identification of candidate shear wall regions through geometric morphology analysis includes the following steps:

[0029] Based on the individual wall data in the interior plan matrix, calculate the geometric parameters of each individual wall, including length-to-height ratio, area and space occupancy.

[0030] The interior plan matrix is ​​traversed by a pre-configured shear wall structure library to extract the interior structural parts that conform to the shear wall structure.

[0031] In the interior structural section that conforms to the shear wall structure, individual walls whose geometric morphological parameters meet the standard are selected as candidate individual walls for shear walls by using preset geometric standard parameter thresholds;

[0032] The selected candidate individual walls are clustered according to their spatial location to identify candidate shear wall regions composed of several individual walls.

[0033] Furthermore, the step of selecting and recombining individual wall graphics based on shear wall candidate areas includes:

[0034] Based on the indoor planar matrix, adjacent individual walls that conform to the characteristics of shear walls are determined in the candidate shear wall area by spatial location and geometric morphology parameters;

[0035] Based on the spatial connection relationship between adjacent individual walls that conform to the characteristics of shear walls, several combinations of adjacent individual walls that conform to the characteristics of shear walls are obtained.

[0036] Based on the combination results, select individual wall graphics and reassemble and splice the edges of the graphics.

[0037] Furthermore, the step of parsing and reconstructing the semantics of the graphic based on the semantic segmentation algorithm and using it as input to the BP neural network to output the shear wall recognition result includes the following steps:

[0038] The reconstructed graphic is input into the semantic segmentation model to parse the semantic information contained in the reconstructed graphic, including walls, windows, balcony openings, and doorways;

[0039] By importing semantic information into the BP neural network model, the current judgment is made as to whether the reconstructed graphic conforms to the standard characteristics of a shear wall.

[0040] Furthermore, the formula for the BP neural network is as follows:

[0041] y = f(W·x + b);

[0042] Where y is the output result, including whether the current recombined graphic is a shear wall; W is the weight matrix; b is the bias term; f is the activation function; and x is the feature vector of the input layer, i.e., the combination of semantic information.

[0043] The beneficial effects of this invention are as follows: This invention acquires indoor wall images from different points or orientations using image acquisition equipment, and combines distance measurement data and positioning data to ensure multi-dimensional and accurate data acquisition, avoiding subjective errors caused by manual measurement and annotation. Compared with traditional methods, automated data acquisition can cover more details of the indoor environment, laying a more reliable foundation for subsequent analysis. Furthermore, the acquired images are processed using a convolutional neural network to automatically calibrate the wall connections, and the wall images are segmented based on this. The calculation of the height and length of individual walls based on distance measurement data ensures that the physical dimensions of the wall structure in the image match the actual environment. This step greatly improves the accuracy of recognition, especially when shear walls are similar in appearance to other walls, enabling more precise differentiation of shear wall characteristics. Individual walls in a two-dimensional matrix are identified through a combination of geometric morphology analysis and a shear wall structure library. Unlike traditional geometric rule analysis, this solution uses a pre-configured shear wall structure library to traverse the individual wall data, selecting candidate individual walls that meet the characteristics of shear walls from the indoor structural parts that conform to the shear wall structure by combining geometric standard parameters. By clustering analysis of the spatial location and connectivity of adjacent individual walls, candidate shear wall regions are identified, solving the problem of difficult feature extraction of shear walls in complex indoor environments. The semantic segmentation algorithm analyzes the semantic information of the reconstructed image and combines it with a backpropagation neural network to determine whether the reconstructed image conforms to the standard features of a shear wall, effectively reducing the possibility of misjudgment and optimizing the recognition accuracy of shear walls. Finally, 3D visualization provides engineers with an intuitive display of the shear wall recognition results. Attached Figure Description

[0044] Figure 1 This is a flowchart of step 1 of a shear wall identification method based on recombined segmented graphics in this invention.

[0045] Figure 2 This is a flowchart of step S3 in this invention. Detailed Implementation

[0046] Please see Figure 1-2 As shown, this invention relates to a shear wall identification method based on reconstructed segmented graphics, comprising the following steps:

[0047] S1. Acquire several indoor wall images and distance measurement data from different points or orientations using image acquisition equipment, and simultaneously collect positioning and orientation data when acquiring indoor wall images;

[0048] S2. Use a convolutional neural network to mark the wall connection points in the indoor image, divide the wall connection points into several individual wall graphics, and calculate the height and length of the individual wall based on the distance measurement data.

[0049] S3. Initialize the two-dimensional matrix and, based on the height and length of the individual walls and the positioning and orientation data when the wall images were acquired, map several individual walls into the two-dimensional matrix to obtain the indoor planar matrix.

[0050] S4. Based on the indoor plane matrix, candidate shear wall areas are identified through geometric morphology analysis. Individual wall graphics are selected and recombined according to the candidate shear wall areas to obtain recombined graphics. The semantics of the recombined graphics are analyzed based on the semantic segmentation algorithm and used as input to the BP neural network to output the shear wall recognition result.

[0051] S5. Perform 3D visualization based on the identification results of the shear wall.

[0052] In some embodiments, in step S1, a handheld mobile device integrating a high-resolution camera and a laser rangefinder is used to acquire images of indoor walls from multiple points and different angles. The handheld device simultaneously acquires distance data between the wall and the device via a laser ranging module, and obtains positioning and orientation information during image acquisition via a built-in inertial measurement unit (IMU). This information ensures accurate correspondence between the image and the physical space. The device continuously records its positional changes in space during acquisition, and through real-time data synchronization, ensures that each image and ranging data reflects the current device orientation and distance. In step S2, the acquired images and ranging data are transmitted to a computing unit, where a convolutional neural network is used to calibrate the wall connections in the images. Through multi-layer convolution and feature extraction of the images by the network, the system can automatically identify the wall connection points. Based on the depth information in the image and the ranging data provided by the handheld device, the system further calculates the height and length of each individual wall at each connection point, ensuring that the geometric dimensions of the individual walls match the actual physical dimensions. Next, in step S3, the system maps the acquired height, length, and device positioning and orientation data of the individual walls into a two-dimensional matrix to generate an indoor planar matrix. The spatial location information of the handheld device is transformed using coordinates to ensure that the accurate positions of all individual walls in the two-dimensional plane are consistent with the actual indoor environment. In step S4, this invention identifies candidate shear wall regions through geometric morphology analysis. First, based on the individual wall data in the indoor planar matrix, the geometric morphology parameters of each individual wall are calculated, including the length-to-height ratio, area, and spatial proportion in the entire planar matrix. These parameters provide a basis for subsequent screening to ensure the accurate extraction of shear wall features. Next, the system traverses the indoor planar matrix through a pre-configured shear wall structure library. The shear wall structure library contains various common shear wall structure patterns and their geometric features. The system compares each individual wall in the two-dimensional matrix one by one according to these structural patterns, extracting the parts that match the shear wall structure. In the indoor structural parts that conform to the shear wall structure, the system further filters individual walls whose geometric morphology parameters meet the standards through preset geometric standard parameter thresholds (such as length-to-width ratio, area range, etc.), marking them as candidate shear wall individual walls. After completing the screening of candidate individual walls, the system performs cluster analysis based on the spatial positional relationships of these individual walls. The purpose of cluster analysis is to identify the spatial connections between multiple individual walls and determine whether they collectively form a continuous shear wall structure region. Through this analysis, the system can identify candidate shear wall regions composed of several adjacent individual walls, laying the foundation for subsequent shear wall reorganization and further analysis. Finally, in step S5, the system displays the shear wall identification results through a 3D visualization module. Based on data collected by a handheld device and the shear wall identification results, the 3D visualization system intuitively displays the specific location, dimensions, and spatial relationships of the shear walls within the indoor environment.This process can be integrated with Building Information Modeling (BIM) systems to provide architects and construction workers with accurate shear wall location information, ensuring the reasonable preservation and use of shear walls during interior renovations and avoiding structural safety hazards caused by accidental demolition of shear walls.

[0053] Furthermore, the step of calibrating wall joints in indoor images using a convolutional neural network includes the following steps:

[0054] The collected images of the interior walls are input into the trained convolutional neural network model;

[0055] Edge feature points in the wall image are extracted by convolutional layers and downsampled by pooling layers;

[0056] The extracted features at the wall joints are calibrated using fully connected layers in a convolutional neural network.

[0057] In some embodiments, indoor images are processed using a convolutional neural network (CNN) to achieve accurate calibration of wall joints. First, indoor wall images captured by a handheld mobile device are input into a pre-trained CNN model. This model, optimized with a large amount of labeled training data, can automatically identify wall features in the image. The CNN employs a layered structure, with each layer extracting features from the image data at different levels. The first layer of the CNN is a convolutional layer, whose main function is to extract edge feature points from the image. Specifically, the convolution operation detects edge information in different directions by applying multiple convolutional kernels to the image. The convolutional kernel is a trained weight matrix that extracts local features, such as vertical, horizontal, or diagonal edges of the walls, while scanning the image. Through the convolution operation, low-level features in the original image are gradually transformed into high-level features, forming a more abstract representation, especially capturing edge features at wall joints. After processing by the convolutional layer, the extracted feature maps are input into a pooling layer for downsampling. The main function of the pooling layer is to reduce the computational cost by reducing the size of the feature maps and to enhance the model's noise resistance. The commonly used pooling method is max pooling, which can retain the most salient features in local regions while eliminating redundant details in the image. Downsampling not only reduces data dimensionality but also makes subsequent calculations more efficient. Furthermore, by reducing the size of the feature map, it can highlight the core features at wall joints, further improving recognition accuracy. After convolution and pooling operations, the extracted feature maps are fed into fully connected layers. The fully connected layers are responsible for combining the local features extracted by the convolutional layers into global features, ultimately calibrating the wall joints. The fully connected layers perform a linear transformation on all input features through a weight matrix, followed by a non-linear transformation through an activation function, finally outputting a calibration result. Specifically, after computation by the fully connected layers, the model can accurately identify the location of each wall joint in the image based on the wall edge features extracted by convolution and precisely calibrate them.

[0058] Furthermore, the training of the convolutional neural network includes the following steps:

[0059] Collect several indoor wall images of unlabeled wall joints, manually label the wall joints, generate corresponding labeled images, and form a pair of training sample sets;

[0060] Based on the visual characteristics of the wall connection, the network structure of convolutional layers, pooling layers and fully connected layers is customized, and the network weight parameters are initialized.

[0061] Unlabeled interior wall images are input into the convolutional neural network model, and corresponding labeled images are used as supervision signals. The network parameters are iteratively updated using the backpropagation algorithm and loss function.

[0062] In some embodiments, firstly, a large number of unlabeled interior wall images need to be acquired. These images are obtained from different locations and angles using handheld mobile devices, covering various interior scenes and wall features. After acquiring these images, annotators label the wall connections in each image. This process is implemented using software tools, where annotators precisely delineate wall connection points in each image, generating a labeled image corresponding to the original image. Each training sample set includes an original image and its corresponding labeled image, forming the basic dataset required for supervised learning. The number and depth of convolutional kernels need to be fine-tuned through multiple experiments. Increasing the number of convolutional kernels allows the neural network to extract more diverse features in different regions of the image, covering various possible connection forms such as vertical, horizontal, and slanted edges. When designing the network depth, shallow networks may not be sufficient to extract complex features; therefore, network design typically includes multiple convolutional layers. As the layers deepen, the feature representation gradually transitions from low-level pixel information to higher-level abstract features. To address the feature complexity of wall connections, the depth and width of the convolutional layers (i.e., the number of convolutional kernels in each layer) should be adjusted experimentally to ensure that the network's feature extraction capability is sufficiently powerful to adapt to various complex interior environments. Following the convolutional layers, pooling layers are introduced to reduce the dimensionality of the feature maps. The core purpose of pooling is to reduce data redundancy through downsampling while retaining the most important feature information. In this scheme, max pooling is commonly used to extract features at wall joints. Max pooling retains the maximum feature value in each local region, which effectively enhances the robustness of the network against noise or interference in indoor environments. Typically, the pooling kernel size is 2x2, meaning the feature map size is halved with each downsampling. Pooling ensures that the convolutional neural network can extract the most significant connection features from a large-scale input image without losing important geometric shapes at wall joints. After the pooling layers, the network enters a fully connected layer, which is responsible for integrating the extracted features into a global feature vector. In the fully connected layer, all feature maps generated by the preceding convolution and pooling processes are unfolded into a vector, where each element represents a specific feature in the image. Through linear transformation, all features are weighted and combined to form the model's output. To enhance the model's nonlinear expressive power, fully connected layers typically incorporate nonlinear activation functions such as ReLU (Rectified Linear Array). For wall joint calibration tasks, the output of the fully connected layer is the specific coordinates of the wall joint in the image or the calibration result for a specific region.

[0063] Network initialization is crucial for successful training. Network weights are typically set using Gaussian random initialization or He initialization. Gaussian initialization allows weights to start with small random values, ensuring stable gradient flow in the early stages of training and preventing vanishing or exploding gradients. He initialization dynamically adjusts the weights based on the number of input units in the network layer. This method performs well in training deep networks, especially when dealing with complex features at wall junctions, ensuring the convolutional layers' sensitivity to edges and corners. During training, with each iteration, the convolutional neural network gradually learns the salient features of wall junctions from the training data. To ensure the model's generalization ability, data augmentation techniques such as rotation, scaling, and flipping are often used to randomly transform the original image, increasing the diversity of training samples. This helps the model adapt to different environmental changes and enhances its robustness in recognizing wall junctions in various indoor scenes.

[0064] Furthermore, the calculation of the height and length of a single wall based on the distance measurement data includes the following steps:

[0065] Extract the horizontal distance between the individual wall and the image acquisition device, the distance between the vertical endpoint of the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device from the ranging data.

[0066] The height of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the vertical endpoint of the individual wall and the image acquisition device, respectively.

[0067] The length of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device.

[0068] In some embodiments, the height and length of individual walls are accurately calculated using distance measurement data to ensure that the accurate mapping of the wall structure in a two-dimensional plane matrix is ​​consistent with the physical environment. First, while a handheld mobile device acquires images of the walls from different positions and orientations, an integrated laser rangefinder records the distance data between each individual wall and the device. Specifically, the distance data includes the horizontal distance between the individual wall and the device, the distance between the vertical endpoint and the device, and the distance between the parallel endpoint and the device. These measurement points provide core parameters for subsequent height and length calculations. For calculating the height of an individual wall, the system first extracts the horizontal distance between the individual wall and the image acquisition device, and the distance between the vertical endpoint and the device. The horizontal distance refers to the distance between the wall baseline and the device's location, while the vertical endpoint distance represents the vertical distance between a point on the top or bottom of the wall and the device. Using these two key parameters, combined with the wall's tilt angle or the image acquisition device's elevation angle, the system can calculate the height of the individual wall using trigonometric function formulas (such as sine or cosine functions). Specifically, when the image acquisition device is in a horizontal or non-horizontal position, the vertical distance can be converted into wall height information using the actual measurement value obtained by the distance measuring device and the device's tilt angle. The length of a single wall is calculated using the horizontal distance between the wall and the image acquisition device, as well as the distances between the wall's parallel endpoints and the device. The distances to the parallel endpoints are typically measured using a laser rangefinder in horizontal scanning mode, which accurately measures the wall's horizontal length. By measuring the distances between the two parallel endpoints of the wall and the image acquisition device, and combining this with the horizontal distance, the system can further calculate the wall's actual length. In this process, the system also uses trigonometric formulas to convert these distance data into actual wall length values ​​based on the relative angles between the measurement points and the device.

[0069] Further, S3 includes the following steps:

[0070] Based on the height and length data of individual walls, the dimensions of each individual wall are normalized.

[0071] Based on the orientation of the image acquisition device and the spatial orientation of the individual wall, a sub-two-dimensional matrix for mapping is determined, and the matrix coordinates of the individual wall coverage are determined based on the normalized length data of the individual wall.

[0072] Based on the matrix coordinates of the individual wall coverage, the height of the normalized individual wall is mapped to the matrix element values ​​in a two-dimensional matrix to obtain the indoor planar matrix.

[0073] In some embodiments, firstly, based on the measured physical height and length data of individual walls, the dimensions of each individual wall are normalized. The purpose of normalization is to convert wall data of different scales into a uniform scale, enabling all individual walls to be represented in the same two-dimensional matrix and ensuring that its scale is consistent with the actual space. The normalization process uses a reference scale (such as the height or length of the largest or smallest wall) as a standard, scaling the actual dimensions of each individual wall according to this standard. The normalized data is represented in standard units, ensuring that even if there are significant differences in wall sizes, all individual walls can be accurately calculated and processed within the same matrix. After normalization, the system determines a sub-two-dimensional matrix for mapping based on the orientation of the image acquisition device and the spatial orientation of the individual walls. To ensure projection accuracy, the system calculates the projection angle and position of each individual wall in two-dimensional space based on the device's positioning and orientation information. This process relies on the orientation data of the image acquisition device, such as pitch angle and horizontal rotation angle, and uses a geometric transformation algorithm to convert the individual wall data in three-dimensional space into a projection matrix on a two-dimensional plane. The sub-2D matrix is ​​a subset of the main 2D matrix, specifically designed to represent the spatial extent and geometric features of a particular individual wall. To ensure projection accuracy, the system performs real-time correction on the coordinates of each individual wall, taking into account orientation variations and positional errors of the image acquisition device. Next, the system determines the coverage coordinates of each individual wall in the sub-2D matrix based on the normalized length data. The normalized length of each individual wall determines its horizontal extent on the 2D plane, i.e., its lateral coordinate range. The system determines the specific start and end coordinates of each individual wall in the matrix by matching the normalized length with the scaling factor of the 2D plane matrix. This process ensures that the horizontal occupancy of the individual walls is accurately reflected in the 2D matrix, avoiding coordinate misalignment problems caused by data deviations. Finally, based on the matrix coordinates covered by the individual walls, the system maps the normalized height of each individual wall to the matrix element values ​​in the 2D matrix. The value of each matrix element represents the normalized height of the individual wall at that location. By calibrating the vertical position of the matrix, the system can accurately reflect the height distribution of individual walls on the 2D plane. In this way, an interior planar matrix is ​​generated, in which the height and length of each individual wall are clearly represented in two-dimensional space, ensuring a one-to-one correspondence between the physical structure and the matrix structure. The resulting interior planar matrix not only contains the geometric information of the walls but can also be used for further processing operations such as shear wall identification and 3D reconstruction.

[0074] Furthermore, the identification of candidate shear wall regions through geometric morphology analysis includes the following steps:

[0075] Based on the individual wall data in the interior plan matrix, calculate the geometric parameters of each individual wall, including length-to-height ratio, area and space occupancy.

[0076] The interior plan matrix is ​​traversed by a pre-configured shear wall structure library to extract the interior structural parts that conform to the shear wall structure.

[0077] In the interior structural section that conforms to the shear wall structure, individual walls whose geometric morphological parameters meet the standard are selected as candidate individual shear walls by using preset geometric standard parameter thresholds;

[0078] The selected candidate individual walls are clustered according to their spatial location to identify candidate shear wall regions composed of several individual walls.

[0079] It should be noted that geometric morphology analysis is used to accurately identify candidate shear wall regions from the interior planar matrix. First, the system calculates the geometric morphology parameters of each individual wall based on the individual wall data in the two-dimensional planar matrix. These parameters include the length-to-height ratio, area, and spatial proportion. Specifically, the system calculates the length-to-height ratio of each individual wall using its length and height data; this parameter indicates the wall's shape characteristics. The area of ​​each individual wall is determined by the number of grid cells it occupies in the matrix; and the spatial proportion reflects the proportion of the individual wall within the entire interior structure. Next, the system traverses the interior planar matrix using a pre-configured shear wall structure library. This library stores geometric models and feature patterns of various standard shear walls, covering different types of building and structural forms. The system compares the geometric parameters of each individual wall with the models in the shear wall structure library. Specifically, the system traverses the matrix by moving the structural window, step by step. Specifically, the structural window starts from a starting position in the planar matrix and moves row by row and column by column, moving one step distance at a time to ensure that all individual walls in the entire planar matrix are covered. During traversal, the geometric parameters of all individual walls within a window, including length-to-height ratio, area, and spatial proportion, are extracted. For each structural window, the system compares the geometric parameters of the individual walls it contains with models in the shear wall structure library. The shear wall structure library stores various standard geometric models of shear walls, which are predefined according to different architectural design and construction requirements. During comparison, the system first calculates the overall geometric features of all individual walls within the window, including their spatial arrangement and interconnections. Then, the system uses a matching algorithm to compare these features one by one with the shear wall models in the structure library, identifying sub-matrices that conform to the shear wall standards. When an individual wall within a structural window successfully matches a model in the structure library, the system records the specific location of that sub-region and further filters the walls within the sub-region based on preset geometric standard parameter thresholds. These geometric standard parameter thresholds include upper and lower limits for the length-to-height ratio, minimum and maximum area values, ensuring that the system can exclude individual walls that do not conform to the shear wall structural characteristics. This process is similar to classifying wall features. By excluding individual walls that do not conform to the geometry of shear walls, those that do conform to the geometric standards of shear walls are retained as candidate individual shear walls. To further accurately identify shear walls, the system performs cluster analysis on the selected candidate individual walls. The system further integrates and analyzes these candidate individual walls through cluster analysis. Cluster analysis is based on the relative spatial relationships of these individual walls. By calculating the spatial distance and geometric similarity between individual walls, it determines which individual walls belong to the same shear wall. Specific clustering algorithms can use K-means clustering or hierarchical clustering. These algorithms group adjacent individual walls into one class based on their geometric shape and spatial distribution.For example, if two adjacent individual walls are similar in length-to-height ratio and area and are close to each other, the system can identify a shear wall candidate region composed of multiple individual walls through cluster analysis.

[0080] Furthermore, the step of selecting and recombining individual wall graphics based on shear wall candidate areas includes:

[0081] Based on the indoor planar matrix, adjacent individual walls that conform to the characteristics of shear walls are determined in the candidate shear wall area by spatial location and geometric morphology parameters;

[0082] Based on the spatial connection relationship between adjacent individual walls that conform to the characteristics of shear walls, several combinations of adjacent individual walls that conform to the characteristics of shear walls are obtained.

[0083] Based on the combination results, select individual wall graphics and reassemble and splice the edges of the graphics.

[0084] In some embodiments, the system first determines, within the shear wall candidate area, which individual walls possess geometric features consistent with shear wall characteristics, such as length-to-height ratio, area, and degree of matching with standard shear wall structures, based on the spatial location and geometric parameters of each individual wall. Specifically, the system calculates the distance of each individual wall relative to other individual walls using position coordinates in a two-dimensional plane matrix, and combines this with its geometric features (such as length and width) to determine whether they might belong to the same shear wall structure. During this process, the system identifies which individual walls are physically adjacent and have similar geometric features through spatial proximity calculations. After determining adjacent individual walls that conform to shear wall characteristics, the system combines them by calculating the spatial connection relationships between these individual walls. Spatial connection relationships mainly consider the boundary alignment, directional consistency, and geometric similarity of adjacent individual walls. The system uses coordinate data in the matrix and determines whether adjacent individual walls are continuous through adjacency matrix or topological analysis. For example, when the boundary lengths of two individual walls are close and their connection points are linearly arranged in the plane, the system determines that they are interconnected and combines these adjacent individual walls together. The combined individual walls can present a larger continuous wall surface, which conforms to the overall structural characteristics of shear walls. After completing the spatial connection analysis of adjacent individual walls, the system selects suitable individual walls for recombining and splicing the graphic edges based on the results of these combinations. In this step, the system splices adjacent wall graphics together according to the boundaries of the previously combined individual walls to form a complete shear wall graphic. To ensure the accuracy of splicing, the system smooths the edges of the individual walls to eliminate misalignment problems caused by minor deviations between different walls. During the splicing process, the system repositions these spliced ​​walls in a two-dimensional plane matrix according to the normalized dimensions of each individual wall and outputs the recombined shear wall graphic. In this way, the system can generate a recombined shear wall graphic based on the initially identified candidate areas of individual walls through recombining and splicing.

[0085] Furthermore, the step of parsing and reconstructing the semantics of the graphic based on the semantic segmentation algorithm and using it as input to the BP neural network to output the shear wall recognition result includes the following steps:

[0086] The reconstructed graphic is input into the semantic segmentation model to parse the semantic information contained in the reconstructed graphic, including walls, windows, balcony openings, and doorways;

[0087] By importing semantic information into the BP neural network model, the current judgment is made as to whether the reconstructed graphic conforms to the standard characteristics of a shear wall.

[0088] In some embodiments, the reconstructed image is first input into a semantic segmentation model for semantic parsing. The semantic segmentation model also uses a convolutional neural network (CNN) learning architecture, capable of pixel-level classification of the input image. Specifically, the semantic segmentation model classifies each pixel in the reconstructed image, determining its category. In this way, the system can parse key semantic information related to shear walls in the reconstructed image, including structures such as walls, windows, balconies, and doorways. For example, by recognizing the edge features and textures of the wall, the system can accurately locate the wall's outline and distinguish surrounding building structures such as window sills and balconies, which is crucial for differentiating shear walls from non-load-bearing walls. During semantic segmentation, the model learns and distinguishes different structures in a building progressively based on a pre-trained dataset. For instance, elements such as windows and balconies typically have unique geometric features in images, and the semantic segmentation model combines these features to distinguish these areas from walls. The output of semantic segmentation is typically a label matrix, where the label of each pixel represents the category of wall, window, balcony, or other structure to which the pixel belongs. Next, the system inputs the parsed semantic information into a backpropagation (BP) neural network model. The input to the BP neural network model is a structural feature vector extracted from the semantic segmentation model. This feature vector includes the spatial distribution of the walls, the semantic categories of surrounding structures, and the relative positional relationships of these elements. In this way, the BP neural network can identify shear walls in the reconstructed graphic based on the input feature vector and the standard shear wall features learned during training. After receiving the semantically segmented feature data, the BP neural network calculates the degree of matching between the feature data and the shear wall features through forward propagation. The weights and bias parameters of the BP neural network have been optimized during training based on a large amount of labeled shear wall data. The system uses these parameters to process the input data in multiple layers, gradually determining whether the reconstructed graphic conforms to the standard features of shear walls. The BP neural network uses activation functions to non-linearly process the output of each layer, ensuring that the model can accurately identify complex shear wall structures. Finally, based on the output of the BP neural network, the system determines whether the reconstructed graphic contains structures that conform to the shear wall standards. This result is usually output as a classification probability, indicating the degree of matching between the reconstructed graphic and the shear wall features. In this way, the system can accurately distinguish shear walls from other walls or building structures, and finally output the identification results of shear walls.

[0089] Furthermore, the formula for the BP neural network is as follows:

[0090] y = f(W·x + b);

[0091] Where y is the output result, including whether the current recombined graphic is a shear wall; W is the weight matrix; b is the bias term; f is the activation function; and x is the feature vector of the input layer, i.e., the combination of semantic information.

[0092] Specifically, the BP neural network receives a set of feature vectors x from the input layer. These feature vectors are the result of reconstructing a graphic using a semantic segmentation algorithm, containing semantic information about the structures of walls, windows, balconies, and doorways. Each input feature represents the semantic features of a different region in the reconstructed graphic, describing its geometry, spatial location, and category. These input features are linearly combined with a weight matrix W in the network. The weight matrix is ​​a set of pre-trained parameters used to measure the importance of each input feature in the entire network. Each input feature is multiplied by a corresponding weight value; the larger the weight, the greater the influence of the feature on the identification of shear walls. The calculation of the weight matrix is ​​optimized using a backpropagation algorithm to ensure that the final output can accurately identify shear walls. Next, the system adds a bias term b to each node. This bias term is used to adjust the network's calculation results to ensure a more accurate output. The bias term provides a basic activation level for each node, allowing the network to output reasonable results even if all input features are zero. Subsequently, the network passes these linearly combined results to the activation function. The activation function is a non-linear transformation that enables the network to handle more complex data patterns. In shear wall recognition, commonly used activation functions include ReLU (Rectified Luminous Interval) and the Sigmoid function. These functions transform the network's linear output into non-linear probability values, used to determine whether the currently reconstructed image conforms to the standard characteristics of a shear wall. Ultimately, the network's output layer calculates a result, y, representing the network's judgment of whether the currently reconstructed image is a shear wall.

[0093] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for identifying shear walls based on reconstructed segmented graphics, characterized in that, Includes the following steps: S1. Acquire several indoor wall images and distance measurement data from different points or orientations using image acquisition equipment, and simultaneously collect positioning and orientation data when acquiring indoor wall images; S2. Use a convolutional neural network to mark the wall connection points in the indoor image, divide the wall connection points into several individual wall graphics, and calculate the height and length of the individual wall based on the distance measurement data. S3. Initialize the two-dimensional matrix and, based on the height and length of the individual walls and the positioning and orientation data when the wall images were acquired, map several individual walls into the two-dimensional matrix to obtain the indoor planar matrix. S4. Based on the indoor plane matrix, candidate shear wall areas are identified through geometric morphology analysis. Individual wall graphics are selected and recombined according to the candidate shear wall areas to obtain recombined graphics. The semantics of the recombined graphics are analyzed based on the semantic segmentation algorithm and used as input to the BP neural network to output the shear wall recognition result. S5. Perform 3D visualization based on the identification results of the shear wall; S3 includes the following steps: Based on the height and length data of individual walls, the dimensions of each individual wall are normalized. Based on the orientation of the image acquisition device and the spatial orientation of the individual wall, a sub-two-dimensional matrix for mapping is determined, and the matrix coordinates of the individual wall coverage are determined based on the normalized length data of the individual wall. Based on the matrix coordinates of the individual wall coverage, the height of the normalized individual wall is mapped to the matrix element values ​​in the two-dimensional matrix to obtain the indoor planar matrix. The process of identifying candidate shear wall regions through geometric morphology analysis includes the following steps: Based on the individual wall data in the indoor plan matrix, calculate the geometric parameters of each individual wall, including length-to-height ratio, area and space occupancy. The interior plan matrix is ​​traversed by a pre-configured shear wall structure library to extract the interior structural parts that conform to the shear wall structure. In the interior structural section that conforms to the shear wall structure, individual walls whose geometric morphological parameters meet the standard are selected as candidate individual shear walls by using preset geometric standard parameter thresholds; The selected candidate individual walls are clustered according to their spatial location to identify candidate shear wall regions composed of several individual walls; The step of selecting and recombining individual wall graphics based on shear wall candidate areas includes: Based on the indoor planar matrix, adjacent individual walls that conform to the characteristics of shear walls are determined in the candidate shear wall area by spatial location and geometric morphology parameters; Based on the spatial connection relationship between adjacent individual walls that conform to the characteristics of shear walls, several combinations of adjacent individual walls that conform to the characteristics of shear walls are obtained. Based on the combination results, select individual wall graphics and reassemble and splice the edges of the graphics.

2. The shear wall identification method based on recombined segmented graphics according to claim 1, characterized in that, The method of locating wall joints in indoor images using a convolutional neural network includes the following steps: The collected images of the interior walls are input into the trained convolutional neural network model; Edge feature points in the wall image are extracted by convolutional layers and downsampled by pooling layers; The extracted features at the wall joints are calibrated using fully connected layers in a convolutional neural network.

3. The shear wall identification method based on recombined segmented graphics according to claim 2, characterized in that, The training of the convolutional neural network includes the following steps: Collect several indoor wall images of unlabeled wall joints, manually label the wall joints, generate corresponding labeled images, and form a pair of training sample sets; Based on the visual characteristics of the wall connection, the network structure of convolutional layers, pooling layers and fully connected layers is customized, and the network weight parameters are initialized. Unlabeled interior wall images are input into the convolutional neural network model, and corresponding labeled images are used as supervision signals. The network parameters are iteratively updated using the backpropagation algorithm and loss function.

4. The shear wall identification method based on recombined segmented graphics according to claim 1, characterized in that, The calculation of the height and length of a single wall based on distance measurement data includes the following steps: Extract the horizontal distance between the individual wall and the image acquisition device, the distance between the vertical endpoint of the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device from the ranging data. The height of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the vertical endpoint of the individual wall and the image acquisition device, respectively. The length of the individual wall is calculated by the horizontal distance between the individual wall and the image acquisition device, and the distance between the parallel endpoint of the individual wall and the image acquisition device.

5. The shear wall identification method based on recombined segmented graphics according to claim 1, characterized in that, The process of parsing and reconstructing the semantics of the image based on the semantic segmentation algorithm and using it as input to the BP neural network to output shear wall recognition results includes the following steps: The reconstructed graphic is input into the semantic segmentation model to parse the semantic information contained in the reconstructed graphic, including walls, windows, balcony openings, and doorways; By importing semantic information into the BP neural network model, the current judgment is made as to whether the reconstructed graphic conforms to the standard characteristics of a shear wall.

6. The shear wall identification method based on recombined segmented graphics according to claim 1, characterized in that, The formula for the BP neural network is as follows: ； Where y represents the output result, including whether the currently reconstructed graphic is a shear wall; b is the weight matrix; b is the bias term; For activation functions; The input layer's feature vectors are combinations of semantic information.

Images