Steel structure installation and adjustment method and system based on vertical column axis dynamic tracking

Abstract

The invention relates to the technical field of image processing, and discloses a steel structure installation and adjustment method and system based on column axis dynamic tracking. The method comprises the steps of collecting a steel structure column image to obtain contour data; segmenting the image and extracting column position features; matching actual and design positions of the stand column and calculating deviation; analyzing the deviation track to generate a displacement distribution diagram; calculating adjustment parameters to form instruction data; and constructing a virtual model to verify the adjustment effect and optimize parameters. Through comprehensive application of dynamic tracking, image processing and a multi-target optimization algorithm of the axis of the stand column, a complete closed-loop system from measurement, analysis, optimization to verification is established, and accurate control and intelligent adjustment of installation of the steel structure stand column are achieved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a steel structure installation and adjustment method and system based on dynamic tracking of column axes. Background Art

[0002] Steel structures are widely used in modern industrial plants, commercial buildings, and high-rise buildings due to their advantages, such as short construction periods, good seismic resistance, and high strength-to-weight ratio. During steel structure construction, columns are the main load-bearing components, and their installation accuracy is directly related to the safety and service life of the entire structure. Traditional methods of installing steel structure columns rely mainly on manual measurement and empirical adjustments. Typically, a theodolite or total station is used to measure the spatial position of the column, and deviations are manually calculated based on the measured data. Adjustments are then made using tools such as hydraulic equipment or jacks. In recent years, with the development of 3D laser scanning and BIM technologies, some advanced steel structure construction companies have begun to adopt digital measurement methods to improve measurement accuracy and efficiency. However, the adjustment process still relies mainly on manual experience and simple calculation tools.

[0003] However, existing technologies have significant shortcomings. First, traditional manual measurement and adjustment methods have limited accuracy and are unable to meet the increasingly stringent steel structure installation accuracy requirements. Second, while current digital measurement methods have improved measurement accuracy, they lack real-time dynamic tracking capabilities and cannot effectively monitor changes in column position during the adjustment process. Third, existing adjustment methods are usually static adjustments for individual columns, lacking consideration of the interrelationships of the overall structure, resulting in adjustments in one area potentially causing new deviations in other areas. Fourth, the lack of a systematic adjustment optimization algorithm results in low adjustment efficiency, and the adjustment results are highly dependent on the operator's experience. Finally, existing methods lack effective verification and feedback mechanisms, making it difficult to ensure the consistency and reliability of the final installation quality. These shortcomings have led to low steel structure installation accuracy, low efficiency, and increased costs, seriously affecting the overall quality and progress of steel structure projects. Summary of the Invention

[0004] The present application provides a steel structure installation and adjustment method and system based on dynamic tracking of column axes, which is used to establish a complete closed-loop system from measurement, analysis, optimization to verification through the comprehensive application of dynamic tracking of column axes, image processing and multi-objective optimization algorithms, so as to achieve precise control and intelligent adjustment of steel structure column installation.

[0005] In the first aspect, the present application provides a steel structure installation and adjustment method based on dynamic tracking of column axes, and the steel structure installation and adjustment method based on dynamic tracking of column axes includes: data collection on steel structure column images to obtain an image data set containing column contours, target point features, timing information and environmental parameters; image segmentation and feature extraction are performed based on the image data set to obtain a timing feature graph representing column position nodes and connection relationships; a computer vision matching algorithm is established based on the timing feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; a bidirectional spatiotemporal trajectory analysis is performed on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution graph; an image processing algorithm is applied to calculate column adjustment parameters based on the three-dimensional deviation distribution graph to obtain adjustment instruction data containing a spatial transformation matrix; a virtual guidance model is constructed based on the adjustment instruction data and real-time image feedback verification is performed to obtain adjustment accuracy evaluation results and model optimization parameters.

[0006] In a second aspect, the present application provides a steel structure installation and adjustment system based on dynamic tracking of column axes, the steel structure installation and adjustment system based on dynamic tracking of column axes comprising: The acquisition module is used to collect data from steel structure column images to obtain an image dataset containing column outlines, target point features, time series information, and environmental parameters; An extraction module, configured to perform image segmentation and feature extraction based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships; A matching module is used to establish a computer vision matching algorithm based on the time series feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; an analysis module, configured to perform a bidirectional spatiotemporal trajectory analysis on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map; a calculation module, configured to calculate column adjustment parameters by applying an image processing algorithm according to the three-dimensional deviation distribution map, and obtain adjustment instruction data including a space transformation matrix; The verification module is used to build a virtual guidance model based on the adjustment instruction data and perform real-time image feedback verification to obtain adjustment accuracy evaluation results and model optimization parameters.

[0007] In a third aspect, a computer device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned steel structure installation and adjustment method based on dynamic tracking of the column axis.

[0008] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned steel structure installation and adjustment method based on dynamic tracking of the column axis.

[0009] In the technical solution provided in the present application, by collecting data on the images of steel structure columns, an image dataset containing column contours, target point features, timing information and environmental parameters is obtained, which solves the problems of insufficient accuracy and incomplete data of traditional measurement methods; image segmentation and feature extraction are performed based on the image dataset to obtain a timing feature graph representing the column position nodes and connection relationships, thereby achieving an accurate description of the spatial positions and mutual relationships of the columns; a computer vision matching algorithm is established based on the timing feature graph to obtain the correspondence and visual deviation data between the actual column image and the design model, making the matching process between the column and the design position more accurate and reliable; a bidirectional spatiotemporal trajectory analysis is performed on the correspondence and visual deviation data to obtain the column displacement vector field and three-dimensional deviation distribution map, thereby fully grasping the dynamic characteristics of the column position change; an image processing algorithm is applied to calculate the column adjustment parameters based on the three-dimensional deviation distribution map to obtain adjustment instruction data containing the spatial transformation matrix, thereby achieving accurate conversion from measurement data to specific adjustment plans; a virtual guidance model is constructed based on the adjustment instruction data and real-time image feedback verification is performed to obtain adjustment accuracy evaluation results and model optimization parameters, forming a complete closed-loop verification and continuous optimization mechanism. It is particularly worth emphasizing that this solution fully utilizes the innovative application of various artificial intelligence algorithms in the field of steel structure installation, including computer vision matching algorithms to improve the accuracy of column identification and matching, bidirectional spatiotemporal trajectory analysis algorithms to effectively capture dynamic change characteristics, and intelligent adjustment strategy generation based on multi-objective optimization and reinforcement learning. The contribution of these algorithm features to the solution is reflected in: on the one hand, it solves the problem of complex column interactions that traditional methods cannot handle, and improves the coordination of overall structural adjustment; on the other hand, through machine learning to continuously optimize adjustment parameters, it realizes the accumulation and application of empirical knowledge, and significantly improves adjustment efficiency and accuracy. Overall, this solution achieves the precision, intelligence and automation of the steel structure column installation process by building a complete technology chain from data collection, feature extraction, matching analysis to adjustment verification, combined with the advantages of artificial intelligence algorithms, and effectively solves the technical problems of insufficient precision, low efficiency and unstable quality in traditional steel structure installation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 This is a schematic diagram of an embodiment of a steel structure installation and adjustment method based on dynamic tracking of column axes in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a steel structure installation and adjustment system based on dynamic tracking of column axes in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a method and system for installation and adjustment of steel structures based on dynamic tracking of column axes. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the steel structure installation and adjustment method based on dynamic tracking of the column axis includes: Step S101: Data collection is performed on steel structure column images to obtain an image dataset containing column outlines, target point features, time series information, and environmental parameters; Step S102: performing image segmentation and feature extraction based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships; Step S103: establishing a computer vision matching algorithm based on the time series feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; Step S104: performing bidirectional spatiotemporal trajectory analysis on the correspondence relationship and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map; Step S105: Calculate column adjustment parameters using an image processing algorithm based on the three-dimensional deviation distribution graph to obtain adjustment instruction data including a space transformation matrix; Step S106: construct a virtual guidance model based on the adjustment instruction data and perform real-time image feedback verification to obtain adjustment accuracy evaluation results and model optimization parameters.

[0014] It is understandable that the execution subject of this application can be a steel structure installation and adjustment system based on dynamic tracking of the column axis, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0015] Specifically, multi-view, high-definition cameras are deployed at the construction site to capture reflective targets on the pillar surfaces. Cameras are typically positioned at various locations across the construction site to ensure comprehensive coverage of each pillar. The captured images undergo geometric correction to eliminate lens distortion, creating a corrected image sequence. The system also records the timestamp of each frame, aligning images captured by different cameras at specific points in time. The images undergo background segmentation to extract pillar outlines and target features. Furthermore, data on ambient lighting, temperature, and vibration parameters is collected and correlated with the image data to form an environmental impact factor matrix. The resulting image dataset contains pillar outlines, target features, temporal information, and environmental parameters. Image segmentation and feature extraction are performed based on the image dataset. This step begins by applying adaptive watershed segmentation to the pillar outlines to generate a pillar region mask. The watershed algorithm treats the image as a terrain surface, where regions with high grayscale values form "ridges" and regions with low grayscale values form "valleys." This process segments the image into distinct regions through a "flooding" process. From the segmented column region mask, the Hough transform algorithm is used to extract column centerlines and calculate the axis equation. The Hough transform identifies linear structures in an image through voting in parameter space and is particularly suitable for extracting the axis of regular geometric structures such as steel columns. Based on the target point features and the column axis equation, column location node data, including 3D spatial coordinates and attitude angles, is generated. This column location node data is then used to construct a topological relationship graph for the steel structure, and a spatial distance threshold algorithm is used to identify the connections between adjacent columns. The column location node data and connection relationships are then correlated with time series information to generate a time series node feature sequence. This is then subjected to time domain filtering to eliminate abnormal jitter, ultimately synthesizing a time series feature graph.

[0016] To establish a computer vision matching algorithm based on time-series feature graphs, design drawings are imported into a computer and the column design axis data is extracted. This data is then converted into a design feature graph in the image space coordinate system through coordinate transformation. Feature descriptors are then extracted from the time-series and design feature graphs, and a feature index table is constructed using the locality-sensitive hashing algorithm. This algorithm hashes similar data points into the same "bucket," accelerating the feature matching process. A feature similarity matrix is calculated using the feature index table, and a modified Hungarian algorithm is used to solve the maximum weighted bipartite matching problem. The Hungarian algorithm is a combinatorial optimization algorithm used to solve assignment problems. In this method, the optimal match between the actual and designed columns is determined. The correspondence between the actual and designed columns is obtained from the matching results, generating a column pairing table. The positional and angular differences of each pair of columns in the pairing table are calculated to generate raw deviation data. This raw deviation data is then statistically filtered and spatially interpolated to form visual deviation data. To perform bidirectional spatiotemporal trajectory analysis on the correspondence and visual deviation data, a time-series data matrix is first constructed, and principal component analysis is used to extract the main change patterns. Principal component analysis can reduce data dimensionality and highlight key changes in the data. The column position trajectory is calculated forward in time, and random noise is eliminated using the Kalman filter algorithm to form a forward displacement vector. Kalman filtering is a recursive filtering algorithm that can estimate the state of a dynamic system from noisy measurements. Similarly, the column position trajectory is calculated backward in time, and Bayesian estimation methods are used to improve reverse prediction accuracy, forming a reverse displacement vector. The forward and reverse displacement vectors are weighted and fused to generate a column displacement vector field. The displacement vector field is spatially interpolated, and a continuous displacement field representation is constructed using radial basis functions. Radial basis function interpolation can handle irregularly distributed data points and generate a smooth, continuous surface. The spatial distribution of position deviations is calculated based on the displacement field and the original position data, generating a three-dimensional deviation distribution map.

[0017] To calculate column adjustment parameters based on the 3D deviation distribution map, a gradient analysis is first performed on the 3D deviation distribution map. Directional derivatives are then calculated to determine the deviation change rate and principal deformation directions. A multi-objective optimization function is constructed based on the deviation change rate, encompassing three sub-objectives: minimizing column position deviation, minimizing adjustment amount, and maximizing structural stability. A particle swarm optimization algorithm is used to solve the multi-objective optimization function, yielding a Pareto-optimal solution set that balances these objectives. Particle swarm optimization simulates the foraging behavior of bird flocks to find optimal solutions, making it suitable for multi-objective optimization problems. Column adjustment vectors are extracted from the Pareto-optimal solution set and then screened for feasibility based on engineering constraints. A rigid body transformation matrix is calculated for the screened column adjustment vectors, generating a spatial transformation matrix containing translation and rotation components. This spatial transformation matrix is then converted into engineering-executable adjustment instructions, generating adjustment instruction data. A virtual guidance model is constructed based on the adjustment instruction data and verified using real-time image feedback. A virtual outline of the column target position is constructed based on the adjustment instruction data and then overlaid onto the real-time image using augmented reality technology. Column adjustment operations are performed at the construction site, and changes in column position during the adjustment process are recorded using real-time image acquisition. The actual column position in the real-time image is aligned with the virtual outline, and the positional coincidence index is calculated. Based on the positional coincidence index, an adjustment accuracy assessment report is generated, including residual deviation and spatial distribution characteristics. Based on the residual deviation and historical adjustment data, the adjustment coefficient is updated using a reinforcement learning algorithm to generate model optimization parameters. The model optimization parameters are fed back into the column adjustment database, forming a closed-loop adjustment optimization mechanism and completing the entire steel structure installation and adjustment process.

[0018] In an embodiment of the present application, data collection is performed on images of steel structure columns to obtain an image dataset containing column contours, target point features, timing information, and environmental parameters, thereby solving the problems of insufficient accuracy and incomplete data in traditional measurement methods. Image segmentation and feature extraction are performed based on the image dataset to obtain a timing feature graph representing the column position nodes and connection relationships, thereby achieving an accurate description of the spatial positions and mutual relationships of the columns. A computer vision matching algorithm is established based on the timing feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data, thereby making the matching process between the column and the design position more accurate and reliable. Bidirectional spatiotemporal trajectory analysis is performed on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution graph, thereby fully understanding the dynamic characteristics of the column position change. An image processing algorithm is applied to calculate the column adjustment parameters based on the three-dimensional deviation distribution graph to obtain adjustment instruction data containing a spatial transformation matrix, thereby achieving accurate conversion from measurement data to specific adjustment plans. A virtual guidance model is constructed based on the adjustment instruction data and real-time image feedback verification is performed to obtain adjustment accuracy evaluation results and model optimization parameters, thereby forming a complete closed-loop verification and continuous optimization mechanism. It is particularly worth emphasizing that this solution fully utilizes the innovative application of various artificial intelligence algorithms in the field of steel structure installation, including computer vision matching algorithms to improve the accuracy of column identification and matching, bidirectional spatiotemporal trajectory analysis algorithms to effectively capture dynamic change characteristics, and intelligent adjustment strategy generation based on multi-objective optimization and reinforcement learning. The contribution of these algorithm features to the solution is reflected in: on the one hand, it solves the problem of complex column interactions that traditional methods cannot handle, and improves the coordination of overall structural adjustment; on the other hand, through machine learning to continuously optimize adjustment parameters, it realizes the accumulation and application of empirical knowledge, and significantly improves adjustment efficiency and accuracy. Overall, this solution achieves the precision, intelligence and automation of the steel structure column installation process by building a complete technology chain from data collection, feature extraction, matching analysis to adjustment verification, combined with the advantages of artificial intelligence algorithms, and effectively solves the technical problems of insufficient precision, low efficiency and unstable quality in traditional steel structure installation.

[0019] In a specific embodiment, the process of executing step S101 may specifically include the following steps: (1) Multi-view high-definition camera equipment is set up around the steel structure columns, and the reflective cursor target on the column surface is captured through image tracking algorithm; (2) Perform geometric correction on the acquired image sequence, eliminate lens distortion through binocular vision calibration algorithm, and obtain the corrected image sequence; (3) Extract timestamp information from the corrected image sequence and align the images captured by different cameras according to time points using a time synchronization algorithm; (4) Perform background separation on the time-aligned images and extract the pillar outline and target point features using an adaptive threshold segmentation algorithm; (5) Collect environmental light, temperature and vibration parameter data, and associate them with image data through data fusion algorithm to form an environmental impact factor matrix; (6) Integrate the column profile, target point features, time series information and environmental impact factor matrix to obtain the image dataset.

[0020] Specifically, multi-view high-definition camera equipment is set up around the steel structure columns. These cameras usually include multiple high-definition industrial cameras, which are arranged at different locations on the construction site to ensure that each steel structure column has sufficient viewing angle coverage. Reflective target points are pre-attached to the surface of the column. These target points are round or square marks made of special high-reflective materials and can maintain good visibility under different lighting conditions. Image tracking algorithm is a technology in computer vision that tracks the movement of target objects by matching feature points between consecutive frames. In this method, the image tracking algorithm extracts these target points from the background by identifying the brightness and shape characteristics of the reflective target points, and records their position coordinates in the image. These coordinate data are the basic data for the subsequent calculation of the column axis.

[0021] The captured raw image sequences often exhibit distortion, requiring geometric correction. Lens distortion, caused by the non-ideal characteristics of the camera's optical system, primarily involves radial and tangential distortion. The binocular vision calibration algorithm uses a calibration plate (typically a checkerboard pattern) to capture multiple images at different positions and angles, calculating the camera's intrinsic parameter matrix and distortion coefficients. The intrinsic parameter matrix contains the focal length and principal point coordinates, while the distortion coefficients describe the mathematical model of the distortion. Using these parameters, each frame is dedistorted to produce a corrected image sequence. In the corrected image, straight lines remain straight, and the angle and distance ratios are maintained, laying the foundation for subsequent accurate measurement. Extracting timestamp information from the corrected image sequence is a key step in establishing a temporal relationship. The timestamp is a precise record of the time each frame was captured, typically with millisecond accuracy. Because multiple cameras are used simultaneously on construction sites, slight time differences may exist between devices. The time synchronization algorithm analyzes the timestamps in the image sequences captured by all devices to establish a unified time reference system. The method uses a single camera as the master clock, and aligns the time series of other cameras to the master clock's time points using linear or spline interpolation. This process groups the images captured by all cameras at the same time together, forming a time-aligned multi-view image set.

[0022] Background segmentation of time-aligned images is a crucial step in extracting pillar information. The goal of background segmentation is to isolate pillars and target points from the complex construction site background. The adaptive threshold segmentation algorithm dynamically adjusts the threshold value based on the characteristics of local regions in the image, adapting to image segmentation requirements under varying lighting conditions. The algorithm first calculates the image's grayscale histogram and then iteratively finds the optimal threshold value to classify image pixels into foreground (pillars and target points) and background. Reflective targets, due to their high reflectivity, appear as areas of high brightness in the image, allowing them to be extracted by setting a high brightness threshold. The extracted pillar outlines typically appear as connected regions, while the target points appear as discrete highlights. These features are further refined through regional connectivity analysis and shape analysis to obtain accurate pillar outline and target point features.

[0023] Environmental factors significantly impact the installation accuracy of steel structures, necessitating the collection of data on parameters such as ambient light, temperature, and vibration. Ambient light is measured using light sensors, recording variations in light intensity across the construction site. Temperature is measured using temperature sensors, recording both ambient and column surface temperatures. Vibration is measured using accelerometers, recording the frequency and amplitude of vibrations at the construction site. Correlating these environmental parameter data with image data requires a data fusion algorithm. This algorithm first aligns all environmental parameters by timestamp, then calculates the correlation between these parameters and image features to model the impact of environmental factors on column position changes. This correlation analysis is performed using multivariate regression or neural network methods, ultimately generating an environmental impact factor matrix that describes the degree to which different environmental factors influence column position and shape.

[0024] The final step in the data acquisition phase is to integrate the pillar outlines, target point features, time series information, and the environmental impact factor matrix to form the final image dataset. This integration process requires the establishment of a unified data structure to organize the various data types according to temporal and spatial relationships. Image datasets are typically stored in multidimensional arrays or structured databases, containing the spatial coordinates of the pillars, target point location information, time series data, and corresponding environmental parameters. The data structure design must consider the needs of subsequent processing, such as feature extraction and matching analysis, to ensure data integrity and accessibility.

[0025] For example, in a large steel structure project, the construction team deployed four high-definition industrial cameras around each column, providing 360-degree coverage. Each column was affixed with 12 reflective targets arranged in a specific geometric pattern. An image tracking algorithm identified the positions of these targets and recorded the column's real-time status. The captured raw images exhibited approximately 3% distortion, which was corrected by calculating camera parameters using a binocular vision calibration algorithm. The resulting image distortion was reduced to less than 0.2%. The four cameras were synchronized with a 5 millisecond accuracy, ensuring temporal consistency across multiple viewpoints. Background segmentation extracted the column outlines and target points from a complex environment, achieving recognition accuracy exceeding 98%. Simultaneously, recorded environmental data revealed that for every 1°C change in ambient temperature, the column length changes by approximately 0.01 mm. This data was integrated into an environmental impact factor matrix. The resulting image dataset contained dynamic column tracking information.

[0026] In a specific embodiment, the process of executing step S102 may specifically include the following steps: (1) Adaptive watershed segmentation is performed on the pillar contours in the image dataset to obtain a pillar region mask map; (2) Extract the column centerline from the column area mask image and calculate the column axis equation using the Hough transform algorithm; (3) Generate column position node data based on the target point characteristics and the column axis equation, where each node contains three-dimensional space coordinates and attitude angles; (4) Use the column position node data to construct a steel structure topology diagram, and use the spatial distance threshold algorithm to identify the connection relationship between adjacent columns; (5) Associating the column position node data and connection relationship with the time series information to generate a time series node feature sequence; (6) Perform time domain filtering on the time series node feature sequence to eliminate abnormal jitter and synthesize the time series feature graph.

[0027] Specifically, the adaptive watershed segmentation algorithm treats the image as a terrain surface, where regions with high grayscale values form "ridges" and regions with low grayscale values form "valleys." The algorithm first calculates the image's gradient to produce a gradient image, then applies the watershed transform to this gradient image. In steel structure column image processing, the gradient calculation is performed using the Sobel operator or the Canny edge detector to highlight the column's edge features. To avoid over-segmentation, the algorithm also incorporates a marker control mechanism, using the rough position of the column outline extracted in the previous step as the initial marker point. The watershed algorithm simulates rising water levels, "flooding" the gradient image starting from the marker point. When different water areas are about to meet, a watershed line is established, forming the segmentation boundary. After processing, a column region mask is obtained: a binary image in which pixels with a value of 1 represent the column area and pixels with a value of 0 represent the background.

[0028] Extracting the column centerline from the column region mask and then calculating the column axis equation using the Hough transform algorithm is a key step in accurately locating the column. The Hough transform is a feature extraction technique used to detect lines, circles, or other parametric shapes in an image. To extract the column axis, the column region mask is first refined to obtain a single-pixel-wide skeleton line. The Hough transform is then applied to convert points in image space to parameter space, where lines are identified through a voting mechanism. The mathematical basis of the Hough transform is the parametric equation of a line:

[0029] in, Represents the distance from the straight line to the origin of the coordinate system. Indicates the angle between the normal line and the x-axis. During the processing, for each foreground pixel in the image, all possible ( , ) combination, and accumulate votes for the corresponding positions in the parameter space. The peak point in the parameter space corresponds to the straight line in the original image. In the column axis extraction, the parameter space is discretized into an accumulator matrix ,in and They are and The discretized value of . By finding the local maximum value in the accumulator matrix, the parameters of the column axis are determined. For the column axis in three-dimensional space, the following parametric equation can be used to express it:

[0030] in, is the position vector of a point on the axis, is the direction vector of the axis, and t is the parameter variable. The axis parameters in three-dimensional space are calculated by triangulating the two-dimensional axis extracted from multiple viewpoint images.

[0031] Generating column position node data based on target point features and column axis equations is the basis for building a steel structure spatial model. Target point features contain the position information of each target point in the image. Through the geometric relationship of multi-view cameras, the coordinates of the target point in three-dimensional space can be calculated. Column position node data is a data structure that describes the spatial position and posture of the column. Each node contains three-dimensional spatial coordinates (X, Y, Z) and posture angle ( ), representing the position of the column center point and the spatial orientation of the column axis, respectively. The node data calculation process is as follows: first, the spatial orientation of the column is determined based on the column axis equation. Then, the 3D coordinates of the target point are fitted using the least squares method to determine the column center position. The attitude angle is calculated from the angle between the direction vector of the column axis and the coordinate axis.

[0032] Constructing a topological relationship diagram of a steel structure using column location node data is an important process for analyzing the connection relationship between columns. A topological relationship diagram is a graph structure in which nodes represent columns and edges represent the connection relationship between columns. The construction process first creates a node set based on the column location node data, and then uses a spatial distance threshold algorithm to determine whether there is a connection relationship between the columns. The spatial distance threshold algorithm calculates the Euclidean distance between any two column nodes. When the distance is less than a preset threshold and meets the connection rules in the steel structure design, a connecting edge is established between the two nodes. The setting of the threshold takes into account the requirements for the length of connectors in the steel structure design code, as well as the influence of measurement errors. The topological relationship diagram not only contains spatial location information, but also records properties such as connection type and connection strength, comprehensively describing the spatial configuration of the steel structure.

[0033] Associating column position node data and connectivity with time series information to generate a time series node feature sequence is fundamental to analyzing the time-varying characteristics of steel structures. This time series node feature sequence records the position and orientation information of all columns at each point in time, as well as the connectivity between columns. During the association process, the column position node data is first sorted by timestamp. A snapshot is then created for each point in time, containing the status of all columns at that point in time. By comparing data from adjacent time points, the rate of change of column position and orientation is calculated, identifying motion trends and abnormal changes. The time series node feature sequence is stored in a multidimensional array structure, facilitating subsequent temporal and spatial analysis.

[0034] Performing time-domain filtering on the characteristic sequence of time series nodes, eliminating abnormal jitter, and synthesizing a time series feature graph are important steps in improving data quality. Time-domain filtering uses a low-pass filter to remove high-frequency noise and transient interference. The filtering process first performs a Fourier transform on the position and attitude data of each column node to obtain a frequency domain representation. A frequency threshold is then applied to filter out high-frequency components, and finally, an inverse Fourier transform is performed to restore the time domain signal. The filtered data retains the main trends in column position changes and eliminates abnormal jitter caused by factors such as measurement errors and environmental vibrations. The time series feature graph is a graphical representation of the filtered time series node characteristic sequence, visually displaying the changes in column position and attitude over time, as well as the dynamic changes in the connection relationships between columns.

[0035] In a specific embodiment, the process of executing step S103 may specifically include the following steps: (1) Import the design drawings into the computer and extract the column design axis data, which is then converted into a design feature diagram in the image space coordinate system through coordinate transformation; (2) Extract feature descriptors from the timing feature graph and the design feature graph, and construct a feature index table using the local sensitive hashing algorithm; (3) Calculate the feature similarity matrix using the feature index table and solve the maximum weighted bipartite matching problem using the improved Hungarian algorithm; (4) Obtain the correspondence between the actual columns and the designed columns from the maximum weighted bipartite matching results obtained by solving the maximum weighted bipartite matching problem, and generate a column pairing table; (5) Calculate the position difference and angle difference of each pair of columns in the column pairing table to generate the original deviation data; (6) Perform statistical filtering and spatial interpolation on the original deviation data to form visual deviation data.

[0036] Specifically, importing the design drawings into the computer and extracting the column design axis data is the basic work for steel structure installation and adjustment. The design drawings usually exist in CAD format and contain the design position and geometric parameters of each component of the steel structure. The import process first converts the CAD file into a processable data format, and then extracts the column design axis data through graphic parsing technology. The column design axis data contains the spatial position and direction information of each column in the world coordinate system, represented by three-dimensional coordinates and direction vectors. After the extraction is completed, these data need to be converted from the world coordinate system to the image space coordinate system. This process is called coordinate transformation. Coordinate transformation involves translation, rotation and scale transformation, which is achieved through transformation matrix. The transformed data forms a design feature map, which is a graphical representation that shows the projection of the column design position in the image space, which is convenient for comparison and matching with the actual captured image.

[0037] Extracting feature descriptors from the time-series feature graph and the design feature graph is a key step in image matching. A feature descriptor is a numerical vector that describes the characteristics of a local region in an image and is used to uniquely identify feature points within the image. In steel structure column matching, feature descriptors primarily include the column's geometric features (such as length, width, and orientation) and topological features (such as its connection to adjacent columns). The extraction process first identifies key points in the time-series feature graph and the design feature graph, then calculates a feature descriptor around each key point. The locality-sensitive hashing algorithm is a similarity search technique that can quickly find similar feature descriptors. This algorithm divides the high-dimensional feature space into multiple "buckets" so that similar feature descriptors fall into the same bucket. In its implementation, multiple hash functions are selected, each mapping a feature descriptor to an integer value. These integer values are combined to form a hash code, which determines the bucket to which the feature descriptor belongs. This method constructs a feature index table, significantly accelerating the feature matching process.

[0038] Calculating the feature similarity matrix using the feature index table is the step that quantifies the degree of column matching. The feature similarity matrix is a two-dimensional array, where each element represents the degree of similarity between the actual column and the designed column. During the calculation, all pairs of actual and designed columns are traversed, the feature index table is queried, feature descriptors sharing the same bucket are found, and the similarity between them is calculated. Similarity is typically measured using Euclidean distance or cosine similarity, with larger values indicating a higher degree of matching. The maximum weighted bipartite matching problem is a combinatorial optimization problem. The goal is to find a set of edges in a bipartite graph such that the sum of these edge weights is maximized, with each node associated with at most one edge. In steel structure column matching, the two sides of the bipartite graph represent the actual column and the designed column, respectively, and the edge weight is the similarity between the corresponding column pairs. The improved Hungarian algorithm is an efficient method for solving this problem. It iteratively searches for augmenting paths until the optimal match is found.

[0039] The matching process outputs a column pairing table, which is the correspondence between actual columns and designed columns obtained from the maximum weighted bipartite matching results. The matching result is a one-to-one mapping relationship, indicating the corresponding designed column for each actual column. The column pairing table is a structured representation of this mapping relationship, typically containing the actual column ID, the corresponding designed column ID, and the matching confidence. The matching confidence is a measure of the reliability of the match and is derived from the values in the similarity matrix. The mathematical representation of the pairing table can be described as follows:

[0040] in, It is a column matching table. represents the identifier of the kth actual column, Represents The identifier of the matching design column, Is a permutation function, indicating the optimal matching result, is the matching confidence, is the total number of actual columns. It is obtained by solving the following maximization problem:

[0041] in, The actual column and design columns The similarity between them.

[0042] Calculating the positional and angular differences for each pair of columns in the column pairing table to generate raw deviation data is a preparatory step for installation adjustments. This raw deviation data includes the differences in spatial position and orientation for each pair of matching columns. Positional differences are calculated by calculating the Euclidean distance between the center point of the actual column and the center point of the corresponding designed column, typically in millimeters or centimeters. Angular differences are calculated by calculating the angle between the actual column axis and the designed column axis, in degrees or radians. These deviation data reflect deviations during steel structure installation and provide a quantitative basis for subsequent adjustments.

[0043] The final step in data processing is to perform statistical filtering and spatial interpolation on the original deviation data to form visual deviation data. Statistical filtering aims to eliminate outliers and noise and improve data reliability. Common statistical filtering methods include median filtering, average filtering, and Gaussian filtering. The filtered data may have spatial discontinuities or insufficient sampling problems, which are solved by spatial interpolation technology. Spatial interpolation is a method of inferring unknown point data based on known point data. Commonly used interpolation algorithms include inverse distance weighted method, Kriging method, and radial basis function interpolation. The visual deviation data formed after processing is a visual representation, usually displayed in the form of a heat map or vector field, which intuitively shows the deviation distribution and trend of various parts of the steel structure.

[0044] In a large-scale steel structure project, the design drawings contained design data for 35 columns. Through import processing, the coordinate and orientation information of each column was extracted and converted to the image coordinate system to generate a design feature map. Time-series feature maps collected at the actual construction site contained 32 identifiable column structures. Feature descriptors were extracted from these two sets of feature maps. Each column descriptor contained geometric parameters and connectivity. These high-dimensional descriptors were mapped to a low-dimensional space using a locality-sensitive hashing algorithm to construct a feature index table. The resulting similarity matrix showed clear correspondence for most columns, but significant confusion existed in some areas. An improved Hungarian algorithm was applied to solve the maximum weighted bipartite matching problem, resulting in the optimal column correspondence. Thirty-one column pairs were extracted from the matching results, of which one actual column was not matched, and four designed columns were not detected in the actual construction. Positional and angular deviations were calculated for these 31 column pairs, revealing the maximum positional deviation of 38 mm, occurring at the edge of the structure, and the maximum angular deviation of 1.8 degrees, occurring in areas significantly affected by external forces. Median filtering is applied to these raw deviation data to remove outliers, and then spatial interpolation is performed using radial basis functions to generate complete visual deviation data. This deviation data clearly shows the main modes of structural deformation, providing data support for subsequent precise adjustments.

[0045] In a specific embodiment, the process of executing step S104 may specifically include the following steps: (1) Establish a time series data matrix based on the correspondence relationship and visual deviation data, and extract the main change patterns through principal component analysis; (2) Calculate the trajectory of the column position change along the time forward direction, eliminate random noise through the Kalman filter algorithm, and form a forward displacement vector; (3) Calculate the column position change trajectory in reverse time, improve the reverse prediction accuracy through the Bayesian estimation method, and form a reverse displacement vector; (4) Perform weighted fusion of the forward displacement vector and the reverse displacement vector to generate a cylindrical displacement vector field; (5) Perform spatial interpolation on the column displacement vector field and construct a continuous displacement field representation through radial basis functions; (6) Calculate the spatial distribution of position deviation based on the displacement field and original position data, and generate a three-dimensional deviation distribution map.

[0046] Specifically, establishing a time series data matrix based on the correspondence and visual deviation data is an important step in the dynamic tracking of the axis of steel structure columns. The time series data matrix is a multidimensional data structure that records the position deviation information of all columns at each time point. The construction process first arranges the visual deviation data in chronological order, and then organizes the data at each time point to form a matrix. The rows of the matrix represent the time series, and the columns represent the position and posture parameters of different columns. Principal component analysis is a dimensionality reduction technique used to extract the main change patterns from high-dimensional data. In steel structure monitoring, principal component analysis identifies the main directions of data change by calculating the eigenvalues and eigenvectors of the data covariance matrix. These main directions usually correspond to the overall deformation mode of the steel structure, such as overall offset, distortion or local deformation. The extracted main change patterns help to understand the essential reasons for the changes in column position and provide guidance for subsequent precise adjustments.

[0047] The fundamental method for tracking the dynamic behavior of columns is to calculate their positional trajectory forward along time. The forward calculation begins at the beginning of the time series and progresses chronologically, recording the changes in column position. This process can be described by the following mathematical model:

[0048] in, is the position vector of the column at time t, is the velocity vector, is the time step, is random noise. The Kalman filter is a recursive estimation algorithm used to estimate the state of a dynamic system from noisy measurement data. The Kalman filter eliminates random noise through a prediction-update cycle, producing a smooth forward displacement vector.

[0049] Reverse calculation of the column position change trajectory along time is a complementary method, starting from the end of the time series and proceeding in reverse order. The reverse calculation process is similar to the forward process, but in the opposite direction. The mathematical model is:

[0050] in, is the position vector of the column at the reverse time t, is the reverse velocity vector, is random noise. Bayesian estimation improves the accuracy of state estimation by combining prior knowledge with observed data. In reverse trajectory calculation, Bayesian estimation uses structural characteristics and historical data as prior information, combined with real-time measurements, to obtain a more accurate reverse displacement vector.

[0051] Weighted fusion of forward and reverse displacement vectors is a key step in improving trajectory estimation accuracy. The fusion process uses a weighted averaging method, assigning weights based on the reliability of the forward and reverse estimates to calculate the final column displacement vector field. The displacement vector field is a complete representation of the column's motion in space, containing the position, velocity, and acceleration of each column. Compared to single-directional trajectories, the fused vector field more accurately reflects the column's actual motion, reducing estimation bias and uncertainty.

[0052] Spatial interpolation of the column displacement vector field is a necessary step in constructing a continuous displacement field representation. Spatial interpolation is a method for estimating data at unsampled locations based on data from discrete sampling points. Radial basis functions are a commonly used interpolation function that uses the distance between sampling points as an independent variable to construct a smooth and continuous interpolation surface. In steel structure monitoring, radial basis function interpolation can estimate the displacement value at any location within the entire structural space based on limited column displacement data, forming a continuous displacement field representation. This continuous representation facilitates subsequent analysis and visualization, providing comprehensive reference information for structural adjustments.

[0053] The final step in generating a 3D deviation distribution map is calculating the spatial distribution of position deviations based on the displacement field and original position data. The calculation process first applies the continuous displacement field to the original position data to obtain the adjusted expected position. The expected position is then compared with the designed position, and the deviation value for each point is calculated. Finally, these deviation values are mapped into three-dimensional space to form a 3D deviation distribution map. This 3D deviation distribution map, typically displayed as a heat map or vector field, intuitively reflects the magnitude and direction of deviations at various locations within the structure and serves as a crucial reference for adjusting steel structure installations.

[0054] In a specific embodiment, the process of executing step S105 may specifically include the following steps: (1) Perform gradient analysis on the three-dimensional deviation distribution map and obtain the deviation change rate and main deformation direction through directional derivative calculation; (2) Construct a multi-objective optimization function based on the deviation change rate, which includes three sub-objectives: minimizing the column position deviation, minimizing the adjustment amount, and maximizing the structural stability; (3) Solve the multi-objective optimization function through the particle swarm optimization algorithm and obtain the Pareto optimal solution set that balances all objectives; (4) Extract the column adjustment vector from the Pareto optimal solution set and perform feasibility screening based on engineering constraints; (5) Calculate the rigid body transformation matrix for the screened column adjustment vector to generate a spatial transformation matrix containing translation and rotation components; (6) Convert the spatial transformation matrix into engineering executable adjustment instructions to form adjustment instruction data.

[0055] Specifically, gradient analysis of the three-dimensional deviation distribution map is the first step in calculating the column adjustment parameters. The three-dimensional deviation distribution map is a spatial representation of the deviation between the actual position and the designed position of the column, which contains coordinate and deviation value information. Gradient analysis obtains the spatial distribution characteristics of the deviation by calculating the rate of change of the deviation field in each direction. The directional derivative is the rate of change of a function in a specified direction. By calculating the derivatives in different directions, the direction and rate of change of the fastest deviation change can be determined. In the specific calculation process, the three-dimensional deviation distribution map is first discretized into grid points, and then the partial derivatives of the deviation value in the x, y, and z directions are calculated for each grid point to form a gradient vector. The size of the gradient vector represents the rate of change of the deviation, and the direction points to the direction where the deviation increases the fastest. By analyzing the gradient distribution of the entire space, the main deformation areas and deformation directions can be identified, providing a quantitative basis for column adjustment.

[0056] Constructing a multi-objective optimization function based on the rate of change of deviation is the key to solving the optimal adjustment solution. The multi-objective optimization function integrates three competing sub-objectives: minimizing column position deviation, minimizing adjustment amount, and maximizing structural stability. The goal of minimizing column position deviation aims to reduce the difference between the actual column position and the designed position, usually expressed as Euclidean distance. The goal of minimizing adjustment amount is to reduce the workload of the adjustment operation, usually measured by the column movement distance and angle change. The goal of maximizing structural stability is to maintain the overall stiffness and strength of the structure, usually evaluated by calculating the stress distribution and deformation energy of the adjusted structure. These three sub-objectives are often in conflict. For example, reducing position deviation may require a large adjustment amount, while too large an adjustment amount may affect structural stability. Therefore, it is necessary to construct a weighted comprehensive objective function to balance the importance of each sub-objective.

[0057] Solving multi-objective optimization functions using the particle swarm optimization algorithm is an effective method for obtaining equilibrium solutions. The particle swarm optimization algorithm is a heuristic optimization algorithm that simulates the foraging behavior of bird flocks to find the optimal solution. In the steel structure column adjustment problem, each particle represents a possible adjustment solution, containing the adjustment parameters of all columns. The algorithm first randomly initializes a swarm of particles and then iteratively updates the position and velocity of each particle. The update rule takes into account the particle's own historical optimal position and the global optimal position of the swarm, guiding the particles to move towards a more optimal solution space. In multi-objective optimization, the algorithm maintains the diversity and convergence of the solution set through non-dominated sorting and crowding calculation, ultimately forming a Pareto optimal solution set. A Pareto optimal solution set is a set of solutions where each solution cannot further improve the value of other objective functions without reducing the value of any objective function.

[0058] Extracting column adjustment vectors from the Pareto optimal solution set is a crucial step in transforming theoretical solutions into practical solutions. Each solution in the Pareto optimal solution set represents a possible adjustment scheme, but not all schemes are suitable for actual construction. The extraction process first selects several representative solutions from the solution set based on project priorities and decision preferences; these solutions are then converted into column adjustment vectors, each of which contains the displacement and rotation parameters of the column. Engineering constraints, such as the capacity limitations of the adjustment equipment, construction space restrictions, and structural safety margin requirements, are used to screen for feasible solutions. Feasibility screening eliminates infeasible solutions by checking whether each adjustment vector satisfies all constraints, retaining adjustment vectors that meet the actual project requirements.

[0059] Calculating the rigid body transformation matrix for the filtered column adjustment vectors is the basis for generating precise adjustment instructions. The column adjustment vector describes the distance the column needs to move and the angle of rotation, but the construction site requires a more specific transformation description. The rigid body transformation matrix is a mathematical tool used to describe the translation and rotation of an object in three-dimensional space. The calculation process first decomposes the column adjustment vector into translation and rotation components; then constructs the translation matrix and rotation matrix; and finally, multiplies the two matrices to obtain the complete spatial transformation matrix. The spatial transformation matrix is a 4×4 matrix, with the upper left 3×3 submatrix representing rotation and the upper right 3×1 submatrix representing translation. This matrix can accurately describe the transformation process of the column from its current position to the target position.

[0060] Converting the spatial transformation matrix into engineering-executable adjustment instructions is the final step in achieving precise adjustments. The spatial transformation matrix is a mathematical representation that needs to be converted into specific operational instructions that engineers can understand and execute. The conversion process first extracts the translation vector and rotation angle from the transformation matrix. These parameters are then converted into equipment operating parameters based on the characteristics of the adjustment equipment used on-site, such as the hydraulic jack's lifting height, lateral thrust distance, and pad thickness. Finally, adjustment instruction data is generated, including the adjustment direction, adjustment amount, and execution sequence for each column. Adjustment instruction data is typically presented in a table or graphical format to facilitate understanding and execution by construction personnel.

[0061] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Construct a virtual outline of the target column position based on the adjustment instruction data, and superimpose the virtual outline on the real-time image through augmented reality technology; (2) Implement column adjustment operations at the construction site and record the changes in column positions during the adjustment process through real-time image acquisition algorithms; (3) Perform image registration between the actual position of the column in the real-time image and the virtual outline, and calculate the position coincidence index; (4) Generate an adjustment accuracy assessment report based on the position coincidence index, including the residual deviation and spatial distribution characteristics; (5) Based on the residual deviation and historical adjustment data, the adjustment coefficient is updated through the reinforcement learning algorithm to generate the model optimization parameters; (6) Feedback the model optimization parameters to the column adjustment database to form a closed-loop adjustment optimization mechanism and complete the entire steel structure installation adjustment process.

[0062] Specifically, a three-dimensional virtual model of the column is created, including its geometry, dimensions, and spatial position. This virtual outline is constructed using computer-aided design technology, combining the column's design parameters with target position data to generate a spatial outline of the column at the target location. Augmented reality (AR) technology overlays virtual information onto a real-world view. In steel structure installation, this is achieved using head-mounted displays (HMDs) or mobile devices such as tablets. The AR system first identifies feature points in the on-site environment, establishes a coordinate correspondence between the real and virtual worlds, and then overlays the virtual column outline onto the real-time image based on this spatial correspondence. This intuitive visual guidance provides construction workers with a clear understanding of the target column position, significantly improving adjustment accuracy and efficiency. When adjusting columns at the construction site, a real-time image acquisition algorithm is required to record changes in column position during the adjustment process. This real-time image acquisition algorithm, based on computer vision technology, continuously captures a sequence of images of the column using multiple fixed or mobile cameras. The image acquisition frequency is typically set to multiple frames per second to ensure that even subtle changes in column position are captured. The captured images first undergo preprocessing, including denoising, brightness equalization, and geometric correction. Feature extraction algorithms are then used to identify landmarks or characteristic regions on the pillars. This feature extraction uses methods such as edge detection, corner detection, or feature descriptors to generate a characteristic representation of the pillars in the image. By matching features between consecutive frames, the movement of the pillars during adjustment is tracked, and their position changes are recorded. This position change data is stored as a time series for subsequent accuracy analysis and verification.

[0063] Image registration of the actual column position in the real-time image with the virtual contour is a key step in evaluating adjustment progress and accuracy. Image registration is the process of aligning two or more images. In this method, it involves establishing a spatial correspondence between the real-time column image and the virtual target contour. The registration process first extracts the column contour or feature points from the real-time image and then matches them with the corresponding features of the virtual contour. The matching algorithm typically uses feature point matching or contour matching to calculate the spatial transformation relationship between the two. Positional coincidence metrics quantify the degree of alignment between the actual column and the target position, including positional deviation, angular deviation, and contour coincidence ratio. Positional deviation is calculated by calculating the Euclidean distance between corresponding feature points, angular deviation by calculating the angle between the principal axis directions, and contour coincidence ratio by calculating the ratio of the overlapping area to the total area. These metrics comprehensively reflect the accuracy of column adjustment and provide a quantitative basis for subsequent fine-tuning.

[0064] Generating an adjustment accuracy assessment report based on the position coincidence index is a crucial step in evaluating the effectiveness of adjustments. This report systematically summarizes the results of column adjustments and includes two main components: residual deviation and spatial distribution characteristics. Residual deviation refers to the position and posture deviations that remain after adjustment and is calculated using various parameters in the position coincidence index. Spatial distribution characteristics describe the distribution of residual deviations throughout the structure, including their magnitude range, primary distribution areas, and changing trends. The report generation process begins with statistical analysis of the position coincidence index data, calculating the mean, maximum, minimum, and standard deviation of each deviation. Spatial cluster analysis then identifies concentrated areas and patterns of deviations. Finally, intuitive charts and textual explanations are generated to form a complete assessment report. The assessment report not only documents the final results of the adjustment but also provides data support for subsequent optimization and improvement. Using residual deviation and historical adjustment data, a reinforcement learning algorithm updates adjustment coefficients and generates model optimization parameters, an innovative approach to improving adjustment efficiency and accuracy. Reinforcement learning is a machine learning method that learns optimal decision-making strategies through trial and error and a reward mechanism. In steel structure column adjustment, a reinforcement learning algorithm treats adjustment operations as a decision-making process, using the reduction in residual deviation as a reward signal. By repeatedly trying different adjustment strategies, the algorithm learns the optimal adjustment method. Specifically, the algorithm first constructs a state space (current column position and posture) and an action space (possible adjustment actions). The learning model is then initialized based on historical adjustment data. Q-learning or deep reinforcement learning algorithms are then used to iteratively update the action-value function, ultimately yielding the optimal adjustment strategy for each state. Model optimization parameters include the adjustment coefficient matrix, operation sequence weights, and convergence thresholds, which directly impact the efficiency and accuracy of adjustment.

[0065] Feeding back the model optimization parameters to the column adjustment database to form a closed-loop adjustment optimization mechanism is an important step in achieving continuous improvement. The column adjustment database is a centralized repository for data related to steel structure installation adjustments, including historical adjustment records, model parameters, and empirical knowledge. The feedback process first formats the newly generated model optimization parameters into a form acceptable to the database; then compares and verifies them with the historical parameters in the database; and finally updates the parameter records in the database and marks them with timestamps and applicable conditions. The closed-loop adjustment optimization mechanism is a self-improving working mode. By continuously accumulating adjustment experience and optimizing adjustment strategies, the system can draw on past experience when handling new adjustment tasks, thereby improving adjustment efficiency and accuracy. This mechanism is particularly suitable for continuous steel structure installation projects. Subsequent adjustment operations can benefit from previous experience and gradually improve the overall installation quality.

[0066] The above describes the steel structure installation adjustment method based on the dynamic tracking of the column axis in the embodiment of the present application. The following describes the steel structure installation adjustment system based on the dynamic tracking of the column axis in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the steel structure installation and adjustment system based on dynamic tracking of the column axis includes: The acquisition module is used to collect data from steel structure column images to obtain an image dataset containing column outlines, target point features, time series information, and environmental parameters; An extraction module, configured to perform image segmentation and feature extraction based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships; A matching module is used to establish a computer vision matching algorithm based on the time series feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; an analysis module, configured to perform a bidirectional spatiotemporal trajectory analysis on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map; a calculation module, configured to calculate column adjustment parameters by applying an image processing algorithm according to the three-dimensional deviation distribution map, and obtain adjustment instruction data including a space transformation matrix; The verification module is used to build a virtual guidance model based on the adjustment instruction data and perform real-time image feedback verification to obtain adjustment accuracy evaluation results and model optimization parameters.

[0067] Through the collaborative efforts of the aforementioned components, data collection is performed on steel structure column images to generate an image dataset containing column contours, target point features, temporal information, and environmental parameters, addressing the issues of insufficient precision and incomplete data encountered by traditional measurement methods. Image segmentation and feature extraction are performed on the image dataset to generate a temporal feature graph representing the column position nodes and connection relationships, enabling an accurate description of the spatial positions and relationships of the columns. A computer vision matching algorithm is established based on the temporal feature graph to obtain the correspondence between the actual column image and the designed model, as well as visual deviation data, making the matching process between the columns and the designed positions more accurate and reliable. Bidirectional spatiotemporal trajectory analysis is performed on the correspondence and visual deviation data to obtain the column displacement vector field and three-dimensional deviation distribution map, fully understanding the dynamic characteristics of column position changes. An image processing algorithm is applied to calculate the column adjustment parameters based on the three-dimensional deviation distribution map, generating adjustment instruction data containing a spatial transformation matrix, achieving a precise conversion from measurement data to specific adjustment solutions. A virtual guidance model is constructed based on the adjustment instruction data and verified through real-time image feedback to obtain adjustment accuracy assessment results and model optimization parameters, forming a complete closed-loop verification and continuous optimization mechanism. It is particularly worth emphasizing that this solution fully utilizes the innovative application of various artificial intelligence algorithms in the field of steel structure installation, including computer vision matching algorithms to improve the accuracy of column identification and matching, bidirectional spatiotemporal trajectory analysis algorithms to effectively capture dynamic change characteristics, and intelligent adjustment strategy generation based on multi-objective optimization and reinforcement learning. The contribution of these algorithm features to the solution is reflected in: on the one hand, it solves the problem of complex column interactions that traditional methods cannot handle, and improves the coordination of overall structural adjustment; on the other hand, through machine learning to continuously optimize adjustment parameters, it realizes the accumulation and application of empirical knowledge, and significantly improves adjustment efficiency and accuracy. Overall, this solution achieves the precision, intelligence and automation of the steel structure column installation process by building a complete technology chain from data collection, feature extraction, matching analysis to adjustment verification, combined with the advantages of artificial intelligence algorithms, and effectively solves the technical problems of insufficient precision, low efficiency and unstable quality in traditional steel structure installation.

[0068] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0069] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0070] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0071] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0072] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0074] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A steel structure installation and adjustment method based on dynamic tracking of column axes, characterized in that: The steel structure installation and adjustment method based on dynamic tracking of column axes includes: Data collection is performed on steel structure column images to obtain an image dataset containing column outlines, target point features, time series information, and environmental parameters; Performing image segmentation and feature extraction based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships; A computer vision matching algorithm is established based on the time series feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; Performing bidirectional spatiotemporal trajectory analysis on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map; Calculating column adjustment parameters using an image processing algorithm based on the three-dimensional deviation distribution graph to obtain adjustment instruction data including a space transformation matrix; A virtual guidance model is constructed based on the adjustment instruction data and real-time image feedback verification is performed to obtain adjustment accuracy evaluation results and model optimization parameters.

2. The steel structure installation and adjustment method based on dynamic tracking of column axis according to claim 1 is characterized in that: The data collection of the steel structure column image is performed to obtain an image data set containing column contours, target point features, time series information and environmental parameters, including: Multi-view high-definition camera equipment is installed around the steel structure columns, and the reflective cursor target points on the column surface are captured through image tracking algorithms; Perform geometric correction on the acquired image sequence, eliminate lens distortion through binocular vision calibration algorithm, and obtain the corrected image sequence; Extract timestamp information from the corrected image sequence and align the images captured by different cameras according to time points using a time synchronization algorithm; Perform background separation on the time-aligned images and extract pillar outlines and target point features using an adaptive threshold segmentation algorithm; Collect environmental light, temperature and vibration parameter data, and associate them with image data through data fusion algorithms to form an environmental impact factor matrix; The image data set is obtained by integrating the pillar profile, target point features, time series information and environmental impact factor matrix.

3. The steel structure installation and adjustment method based on dynamic tracking of column axis according to claim 1 is characterized in that: The image segmentation and feature extraction are performed based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships, including: Performing adaptive watershed segmentation on the pillar contours in the image data set to obtain a pillar region mask map; Extracting the column centerline from the column area mask image, and calculating the column axis equation using a Hough transform algorithm; Generating column position node data according to the target point features and the column axis equation, wherein each node includes three-dimensional space coordinates and posture angles; Using the column position node data to construct a steel structure topology diagram, and using a spatial distance threshold algorithm to identify the connection relationship between adjacent columns; Associating the column position node data and connection relationship with the time series information to generate a time series node feature sequence; The time series node feature sequence is subjected to time domain filtering processing to eliminate abnormal jitter and synthesize the time series feature graph.

4. The steel structure installation and adjustment method based on dynamic tracking of column axis according to claim 1 is characterized in that: The method of establishing a computer vision matching algorithm based on the time series feature graph to obtain the corresponding relationship between the actual column image and the design model and visual deviation data includes: Import the design drawings into the computer and extract the column design axis data, which is converted into the design feature diagram in the image space coordinate system through coordinate transformation; Extracting feature descriptors from the timing feature graph and the design feature graph, and constructing a feature index table using a local sensitive hashing algorithm; Calculating a feature similarity matrix using the feature index table and solving a maximum weighted bipartite matching problem using an improved Hungarian algorithm; Obtaining the correspondence between the actual columns and the designed columns from the maximum weighted bipartite matching result obtained by solving the maximum weighted bipartite matching problem, and generating a column pairing table; Calculate the position difference and angle difference of each pair of columns in the column pairing table to generate original deviation data; Statistical filtering and spatial interpolation are performed on the original deviation data to form the visual deviation data.

5. The steel structure installation and adjustment method based on dynamic tracking of column axis according to claim 1 is characterized in that: The step of applying an image processing algorithm to calculate column adjustment parameters based on the three-dimensional deviation distribution map to obtain adjustment instruction data including a space transformation matrix includes: A time series data matrix is established based on the correspondence relationship and visual deviation data, and the main change patterns are extracted by principal component analysis; Calculate the column position change trajectory along the time forward direction, eliminate random noise through the Kalman filter algorithm, and form a forward displacement vector; The column position change trajectory is calculated in reverse time, and the reverse prediction accuracy is improved through the Bayesian estimation method to form a reverse displacement vector; Performing weighted fusion on the forward displacement vector and the reverse displacement vector to generate a column displacement vector field; Performing spatial interpolation on the column displacement vector field and constructing a continuous displacement field representation through radial basis functions; The spatial distribution of the position deviation is calculated according to the displacement field and the original position data to generate the three-dimensional deviation distribution map.

6. The steel structure installation and adjustment method based on column axis dynamic tracking according to claim 1 is characterized in that: The bidirectional spatiotemporal trajectory analysis is performed on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map, including: Performing gradient analysis on the three-dimensional deviation distribution map, and obtaining the deviation change rate and main deformation direction through directional derivative calculation; A multi-objective optimization function is constructed based on the deviation change rate, which includes three sub-objectives: minimizing column position deviation, minimizing adjustment amount, and maximizing structural stability; Solving the multi-objective optimization function by a particle swarm optimization algorithm to obtain a Pareto optimal solution set that balances each objective; Extracting column adjustment vectors from the Pareto optimal solution set and performing feasibility screening based on engineering constraints; Calculate the rigid body transformation matrix for the filtered column adjustment vector to generate a space transformation matrix containing translation and rotation components; The spatial transformation matrix is converted into an engineering executable adjustment instruction to form the adjustment instruction data.

7. The steel structure installation and adjustment method based on dynamic tracking of column axis according to claim 1 is characterized in that: The step of constructing a virtual guidance model based on the adjustment instruction data and performing real-time image feedback verification to obtain adjustment accuracy evaluation results and model optimization parameters includes: constructing a virtual outline of the target position of the column according to the adjustment instruction data, and superimposing the virtual outline on the real-time image through augmented reality technology; Implement column adjustment operations at the construction site and record column position changes during the adjustment process through real-time image acquisition algorithms; Performing image registration on the actual position of the pillar in the real-time image and the virtual outline, and calculating a position coincidence index; Generate an adjustment accuracy assessment report based on the position coincidence index, including residual deviation and spatial distribution characteristics; Based on the residual deviation and historical adjustment data, the adjustment coefficient is updated by a reinforcement learning algorithm to generate model optimization parameters; The model optimization parameters are fed back to the column adjustment database to form a closed-loop adjustment optimization mechanism to complete the entire steel structure installation adjustment process.

8. A steel structure installation and adjustment system based on dynamic tracking of column axes, used to implement the steel structure installation and adjustment method based on dynamic tracking of column axes as described in any one of claims 1 to 7, characterized in that: The steel structure installation and adjustment system based on dynamic tracking of column axes includes: The acquisition module is used to collect data from steel structure column images to obtain an image dataset containing column outlines, target point features, time series information, and environmental parameters; An extraction module, configured to perform image segmentation and feature extraction based on the image data set to obtain a time series feature graph representing column position nodes and connection relationships; A matching module is used to establish a computer vision matching algorithm based on the time series feature graph to obtain the correspondence between the actual column image and the design model and visual deviation data; an analysis module, configured to perform a bidirectional spatiotemporal trajectory analysis on the correspondence and visual deviation data to obtain a column displacement vector field and a three-dimensional deviation distribution map; a calculation module, configured to calculate column adjustment parameters by applying an image processing algorithm according to the three-dimensional deviation distribution map, and obtain adjustment instruction data including a space transformation matrix; The verification module is used to build a virtual guidance model based on the adjustment instruction data and perform real-time image feedback verification to obtain adjustment accuracy evaluation results and model optimization parameters.

9. A computer device, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, it implements the steel structure installation and adjustment method based on dynamic tracking of the column axis as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the steel structure installation and adjustment method based on dynamic tracking of column axes according to any one of claims 1 to 7.

Images