Building exterior deformation detection method and device based on binocular vision and storage medium

Abstract

The application provides a building appearance deformation detection method and device based on binocular vision and a storage medium, and belongs to the technical field of computer vision. The method is as follows: a circular target and a scale are arranged on the surface of a building body, a binocular industrial camera is used to shoot a building body image containing the circular target and the scale, and a camera pose is acquired; the circular target and the scale in the image are identified at a sub-pixel level; the circular target is used as an unknown point, a free network without control points is constructed in cooperation with the scale, and translation, rotation and scale constraint conditions of the free network are determined; under the constraint conditions, the three-dimensional coordinates of each circular target are solved by using the least square principle in combination with a camera imaging equation; in the fixed free network, the three-dimensional coordinates of the circular targets at different times are solved, and the deformation variable is calculated by subtracting the positioning results of adjacent two times. The method avoids the problem that the control points are unstable with the deformation of the building body in the building appearance deformation detection process, and realizes high-precision deformation detection of the building appearance.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method, apparatus and storage medium for detecting architectural deformation based on binocular vision. Background Technology

[0002] Currently, common technologies for monitoring building exterior deformation mainly include GNSS monitoring and visual monitoring. GNSS monitoring acquires coordinate information by directly deploying receivers on the building, but GNSS signals are affected by factors such as terrain, building obstruction, and weather. Traditional visual positioning measurement requires the deployment of control points to calculate camera pose and locate target points. Deploying control points is not suitable for deformation monitoring because control points need to be placed within or near the building's monitoring area. The movement of control points leads to a lack of uniformity in the control network during monitoring, and the deformation calculation cannot be performed on the same scale for each target point location result. Furthermore, measuring the coordinates of control points during monitoring consumes a lot of manpower and resources and may also face challenges such as location selection and measurement accuracy, which runs counter to the goal of automated monitoring.

[0003] Common problems faced by pan-tilt-zoom (PTZ) surveillance cameras in high-precision monitoring include: First, the camera resolution is low, failing to meet the millimeter-level accuracy requirements for monitoring building deformation; second, the camera attitude data acquired by the PTZ has low accuracy, necessitating the calculation of the camera's pose to achieve higher observation precision. Industrial cameras are generally used in high-precision monitoring. Therefore, it is necessary to design a method, device, and storage medium for detecting building exterior deformation based on binocular vision. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device and storage medium for detecting deformation of building exterior based on binocular vision, which solves the technical problems of low accuracy and inaccuracy in existing building exterior deformation detection technology, as well as the need to set up control points, which consume a lot of manpower and resources and have insignificant effects.

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

[0006] A method for detecting architectural appearance deformation based on binocular vision, the method comprising the following steps:

[0007] Step 1: Set up a circular target and ruler on the surface of the building, and use a binocular industrial camera to capture an image of the building containing the circular target and ruler, while simultaneously acquiring the pose of the industrial camera.

[0008] Step 2: Use the ellipse fitting algorithm to extract the marker points at both ends of the circular target and ruler in the image, and determine the center of the circle based on the fitting results;

[0009] Step 3: Using a circular target as an unknown point, construct a free net without control points using a scale, and determine the translational degrees of freedom, rotational degrees of freedom, and scale constraints of the free net;

[0010] Step 4: Under the constraints, combined with the camera imaging equation, the three-dimensional coordinates of each circular target are obtained by using the Moore-Penrose generalized inverse matrix through the least squares principle.

[0011] Step 5: In a fixed free network, calculate the three-dimensional coordinates of the circular target at different times, and calculate the deformation by subtracting the positions of two adjacent positioning results.

[0012] Furthermore, in step 1, a circular target with a black border and a white circular pattern is selected. The circular targets are evenly distributed on the surface of the building. A fixed-length scale is selected, and circular markers are fixed at both ends of the scale. The computer controls the binocular industrial camera to acquire two images of the same observation area from different perspectives, and the industrial camera pose is obtained through a dual-antenna attitude-fixing and orientation module.

[0013] Furthermore, in step 3, the target point set is treated as a whole rigid body, and the centroid coordinates of the target point set are constrained to remain unchanged, thus locking the overall translational degrees of freedom of the free net.

[0014] Treat the target point set as a rigid body, constrain the resultant external torque of the target point set about the three principal axes X, Y, and Z to be zero, and lock the overall rotational degree of freedom of the free net;

[0015] Based on a known length scale within the target area, the distance between the two endpoints of the scale is constrained to be equal to the known length, thus locking the scale factor of the free net.

[0016] Furthermore, with the centroid of the constrained target point set remaining unchanged, the sum of the coordinate corrections of all target points in the X, Y, and Z directions is zero.

[0017] Furthermore, when the net external torque of the constrained target point set about the three principal axes X, Y, and Z is all zero, the sum of the torque vector formed by the coordinate corrections of all target points and their corresponding approximate coordinates is zero.

[0018] Furthermore, in step 4, the collinearity condition equation for camera imaging is linearized to obtain the error equation:

[0019]

[0020] in, Represents the exterior orientation correction vector of the image. This represents the vector of interior orientation correction elements of the camera. This represents the three-dimensional coordinate correction vector of the measurement point. , and This represents the corresponding coefficient matrix. This represents the difference between the measured value and the true value.

[0021] The geometric expressions for the translational and rotational degrees of freedom of the constrained free net are transformed into the form of a linear constraint matrix by taking partial derivatives with respect to the object point coordinate corrections and the camera exterior orientation element corrections.

[0022]

[0023] in, Represents the linear constraint matrix. This represents a vector of unknowns consisting of camera exterior orientation element corrections and object point coordinate corrections.

[0024] Transform the scale constraint into an error equation:

[0025]

[0026] in, Represents the scale constraint matrix. This represents the difference between the measured value and the true value.

[0027] Using the camera pose measured by the dual-antenna attitude and orientation module as the externally measured camera exterior orientation element, the error equation is listed:

[0028]

[0029] in, Represents the identity matrix. This represents the difference between the measured value and the true value.

[0030] List the comprehensive error equation:

[0031]

[0032] Using the least squares principle, substitute the constraint equations into:

[0033]

[0034] Find the minimum residual vector, and calculate the camera's exterior orientation elements, interior orientation elements, and the coordinates of the measurement points.

[0035] Furthermore, in step 5, images are captured at regular intervals each day and transmitted in real time to a computer to calculate the coordinates of the measurement points. The image sequence acquired within a monitoring period is labeled as 1, 2, 3, ... N. The deformation between two observations is calculated by subtraction.

[0036]

[0037] in, , and These represent the coordinate differences between the previous time point and the next time point for the X, Y, and Z axes, respectively.

[0038] Furthermore, the specific process of transmitting the coordinates of the measurement point to the computer in real time is as follows: the computer, industrial camera and dual-antenna attitude and orientation module are connected through a pair of industrial-grade bridges, and the computer end and the industrial camera end are connected to the bridges by network cables, which supports long-distance shooting control and wireless data transmission.

[0039] A device for detecting deformation of building exterior based on binocular vision includes a binocular industrial camera module, a wireless transmission module, a target detection module, a target positioning module, and a deformation detection module;

[0040] The binocular industrial camera module is wirelessly connected to the target detection module via a wireless transmission module. The binocular industrial camera module is used to capture images of buildings containing circular targets and rulers, and simultaneously acquire the industrial camera pose. The wireless transmission module is used for computer control of camera shooting, transmitting the image data and camera pose data from the monitoring site to the computer.

[0041] The target detection module is connected to the deformation detection module via the target localization module. The target detection module is used to detect circular targets deployed on buildings and determine the subpixel-level coordinates of the circular targets in the image. The target localization module is used to construct a free net and calculate the three-dimensional coordinates of the circular targets under constraints. The deformation detection module is used to obtain the deformation at each monitoring time interval by subtracting the three-dimensional coordinates of the target points under the monitoring time series.

[0042] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the binocular vision-based building appearance deformation detection method according to any one of claims 1-8.

[0043] The present invention, by adopting the above-described technical solution, has the following beneficial effects:

[0044] This invention establishes a long-term stable free-grid coordinate system in one go by setting up a circular target and scale on-site and utilizing seven geometric constraints, thus avoiding the problems encountered during the detection of building exterior deformation.

[0045] The problem of control points becoming unstable due to building deformation is addressed, eliminating the need for traditional control point placement and re-measurement. Simultaneously, high-precision camera pose is acquired through a dual-antenna attitude and orientation module, and this pose is used as an external weighted observation, substituted into the collinearity condition equation and constraint equations for joint solution. This achieves millimeter-level self-checking and collaborative solution of external orientation soft constraints, ensuring more accurate and robust observation results. Furthermore, the solution process requires only one linear equation solution, eliminating the need for iterations and significantly reducing computational power requirements and time costs. Attached Figure Description

[0046] Figure 1 This is a flowchart of the method of the present invention;

[0047] Figure 2 This is a flowchart illustrating the building exterior deformation detection system based on binocular vision according to an embodiment of the present invention.

[0048] Figure 3 This is a schematic diagram of the arrangement of the circular target and scale according to an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the process for extracting the center of a circular target according to an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the target point object coordinate calculation process according to an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and preferred embodiments. However, it should be noted that many details listed in the specification are merely to provide the reader with a thorough understanding of one or more aspects of the present invention, and these aspects of the invention can be implemented even without these specific details.

[0052] The following is combined Figures 1 to 5 The present invention describes a building appearance deformation detection system based on binocular vision. Figure 1 This is one of the flowcharts for a binocular vision-based building exterior deformation detection system, such as... Figure 1 As shown, the present invention provides a building exterior deformation detection system based on binocular vision, comprising:

[0053] S1. Set up circular targets and rulers on the surface of the building. Use a binocular industrial camera to capture images of the building including the circular targets and rulers, and simultaneously acquire the camera pose.

[0054] Depending on the monitoring surface, circular targets should be arranged. If the building monitoring surface is a vertical and flat surface, adhesive backing and screws can be used for fixing. If there is an inclined monitoring surface or the building surface is complex, measuring blocks can be set on the building surface, and the circular targets can be fixed on the measuring blocks. The overall arrangement of the circular targets should be in a grid pattern, while avoiding overly concentrated arrangements. Circular targets of the same style should be fixed at both ends of the scale for easy image recognition.

[0055] Unlike traditional binocular vision positioning methods, this invention eliminates the need for control points. All circular targets are used as measurement points within a locked free network for calculation, thus saving the process of manually measuring control points in the field and avoiding measurement result deviations caused by unstable control points.

[0056] Based on the size of the building to be measured, the observation distance and the baseline length of the binocular camera are determined. The general observation distance of an industrial camera does not exceed 400 meters. Before capturing images, the dual-antenna attitude and orientation module is connected to the industrial camera via a network port to obtain the camera's pose.

[0057] S2. Perform sub-pixel level recognition of circular targets and rulers in the image.

[0058] Since circular targets are fixed at both ends of the scale, identifying the circular targets at both ends of the scale is equivalent to identifying the position of the scale. After filtering and binarizing the acquired image, the circular targets can be identified. To make the identification results more accurate, this invention designs a pre-selection stage, which calculates candidate regions through connected component analysis, then performs contour extraction and ellipse fitting within the candidate regions, and finally obtains the sub-pixel level center coordinates.

[0059] S3. Under the constraints, combining the camera imaging equation, and using the least squares principle, through...

[0060] The three-dimensional coordinates of each circular target can be obtained by solving the Moore-Penrose generalized inverse matrix.

[0061] The collinearity condition equation describes the geometric relationship between the image point coordinates, the object point object-side coordinates, and the projection center. It is a fundamental equation describing the photographic process. For example, formula (1) is the general form of the collinearity condition equation considering lens distortion:

[0062] (1)

[0063] in, These are like point coordinates. These are the camera's coordinates. These are elements of the rotation matrix. It's the camera's focal length. It's like the principal point coordinates. These are the three-dimensional coordinates of the target point. It's due to poor lens distortion.

[0064] Formula (1) constrains the basic projection relationship. According to the traditional photogrammetric positioning method, it is necessary to use control points to perform resection calculation to obtain the camera pose and incorporate it into the geodetic coordinate system to calculate the target point coordinates. Building deformation monitoring does not require incorporation into the geodetic coordinate system for positioning. Therefore, seven constraint equations are introduced to lock the translation, rotation and scale degrees of freedom of the free network, and the target point is located within the locked free network.

[0065] Assume the column coordinates of the target point in the free network are The correction for the target point column coordinates is The translational degrees of freedom of the free net are constrained using formula (2):

[0066] (2)

[0067] Formula (3) sets the sum of the coordinate corrections of all target points within the free net and the moment vectors formed by their corresponding approximate coordinates to zero, thereby constraining the translational degrees of freedom of the free net.

[0068] (3)

[0069] For images External orientation elements and measurement points The correction amount, whose constraint matrix representation is given by formulas (4) and (5):

[0070] (4)

[0071] (5)

[0072] The translation and rotation constraints of the free net are expressed by formula (6):

[0073] (6)

[0074] Assume the coordinates of the points at both ends of the scale are respectively and The length of the scale is The scale constraint of the free net is completed using the Euclidean distance calculation formula, as shown in formula (7):

[0075] (7)

[0076] Linearize the collinearity condition equation and the scale constraint equation of the free network, and list the error equation. Incorporate the acquired camera pose as a weighted observation into the error equation for joint solution to improve the positioning accuracy. Equation (8) is the comprehensive error equation:

[0077] (8)

[0078] in, Represents the exterior orientation correction vector of the image. This represents the vector of interior orientation correction elements of the camera. This represents the three-dimensional coordinate correction vector of the measurement point. , and This represents the coefficient matrix of the exterior orientation correction, interior orientation correction, and target point coordinate correction for the collinearity equation. Represents the scale constraint matrix. Represents the identity matrix. This represents the difference between the measured value of the scale length and the true value, and the difference between the measured value of the camera pose and the true value.

[0079] The objective function is derived using the least squares principle and the Lagrange multiplier method, as shown in formula (9):

[0080] (9)

[0081] Study the vector of corrections to unknowns when the objective function reaches its minimum. The value of is the final solution vector, and the covariance matrix is ​​solved simultaneously. The solution process does not require iteration; a single linear solution can yield a result that meets the required accuracy. This numerical method is typically equivalent to obtaining the solution through the Moore-Penrose generalized inverse matrix. Since the solution process is a single step, the initial values ​​of the target points must satisfy the geometric relationships in the world coordinate system. For example, the distance between the circular targets at both ends of the scale must be equal to the scale length, and the distance between adjacent circular targets arranged in a grid must also be equal to the grid side length.

[0082] S4. In a fixed free network, calculate the three-dimensional coordinates of the circular target at different times, and calculate the deformation by subtracting the coordinates at different times.

[0083] By setting an observation cycle for building deformation monitoring, images are captured during each observation cycle, the three-dimensional coordinates of the observation points are calculated, and the difference between the current and previous observation results yields the deformation of that point within one observation cycle. Furthermore, through long-term observation and analysis of the overall deformation trend, losses caused by building deformation can be effectively prevented.

[0084] This invention relates to a binocular vision-based building exterior deformation detection system. Circular targets and rulers are set on the building surface. A binocular industrial camera captures images of the building including the circular targets and rulers, simultaneously acquiring the camera pose. Sub-pixel-level recognition of the circular targets and rulers in the images is performed. Using the circular targets as unknown points, a free net without control points is constructed in conjunction with the rulers, and translation, rotation, and scale constraints are determined for the free net. Under these constraints, the three-dimensional coordinates of each circular target are obtained using the least squares principle and the Moore-Penrose generalized inverse matrix, based on the camera imaging equation. Within the fixed free net, the three-dimensional coordinates of the circular targets at different times are calculated, and the deformation is calculated by subtracting adjacent positioning results. This method avoids the problem of unstable control points due to building deformation during exterior deformation detection, achieving high-precision deformation detection of building exteriors.

[0085] Figure 2 This is a second schematic flowchart of the building appearance deformation detection system based on binocular vision according to an embodiment of the present invention, as shown below. Figure 2 As shown, a binocular industrial camera is used to capture images of the building under test, including a circular target and a ruler. At the same time, a dual-antenna attitude and orientation module is used to acquire the camera pose. The images and pose data are transmitted to a computer via a network bridge for image recognition and spatial coordinate positioning. The difference between two adjacent positioning results is used to calculate the deformation, which is then transmitted to the user platform for visualization.

[0086] In one embodiment, setting a circular target and a ruler on the surface of the building, and using a binocular industrial camera to capture images of the building including the circular target and the ruler, while simultaneously acquiring the camera pose, includes: selecting a target with a black border and a white circular pattern for the circular target, evenly distributing the circular targets on the building surface; selecting a ruler of fixed length, with circular markers fixed at both ends of the ruler; and using a computer-controlled binocular industrial camera to acquire two images of the same observation area from different perspectives, and obtaining a high-precision camera pose through a dual-antenna attitude and orientation module.

[0087] Figure 3 This is a schematic diagram of the arrangement of the circular target and scale according to an embodiment of the present invention, as shown below. Figure 3 As shown, the circular targets should be arranged in a grid pattern, while avoiding overly concentrated placement. Identical circular targets are fixed at both ends of the scale to facilitate image recognition.

[0088] In one embodiment, the subpixel-level recognition of the circular target and ruler in the image includes: extracting the marker points at both ends of the circular target and ruler using an ellipse fitting algorithm, and determining the center of the circle based on the fitting result.

[0089] Figure 4 This is a schematic diagram of the circular target center extraction process according to an embodiment of the present invention, as shown below. Figure 4 As shown, after Gaussian filtering and binarization of the acquired image, the circular target can be identified. First, candidate regions are calculated through connected component analysis, then Canny operator edge detection and ellipse fitting are performed within the candidate regions, and finally the sub-pixel level center coordinates are obtained.

[0090] In one embodiment, the step of obtaining the three-dimensional coordinates of each circular target under constraints, combining the camera imaging equation and utilizing the least squares principle through the Moore-Penrose generalized inverse matrix, includes: linearizing the camera imaging collinearity condition equation to obtain the error equation.

[0091]

[0092] in, Represents the exterior orientation correction vector of the image. This represents the vector of interior orientation correction elements of the camera. This represents the three-dimensional coordinate correction vector of the measurement point. , and This represents the corresponding coefficient matrix. This represents the difference between the measured value and the true value.

[0093] The geometric expressions for the translational and rotational degrees of freedom of the constrained free net are transformed into the form of a linear constraint matrix by taking partial derivatives with respect to the object point coordinate corrections and the camera exterior orientation element corrections.

[0094]

[0095] in, Represents the linear constraint matrix. This represents a vector of unknowns consisting of camera exterior orientation element corrections and object point coordinate corrections.

[0096] Transform the scale constraint into an error equation:

[0097]

[0098] in, Represents the scale constraint matrix. This represents the difference between the measured value and the true value.

[0099] Using the camera pose measured by the dual-antenna attitude and orientation module as the externally measured camera exterior orientation element, the error equation is listed:

[0100]

[0101] in, Represents the identity matrix. This represents the difference between the measured value and the true value.

[0102] List the comprehensive error equation:

[0103]

[0104] Using the least squares principle, substitute the constraint equations into:

[0105]

[0106] Find the minimum residual vector, and calculate the camera's exterior orientation elements, interior orientation elements, and the coordinates of the measurement points.

[0107] Figure 5 This is a schematic diagram of the target point object coordinate calculation process according to an embodiment of the present invention, as shown below. Figure 5 As shown, the input includes the pixel coordinates of the circular target extracted in the above process, the length of the scale, and the camera's intrinsic parameters. The object coordinates of the target point in the locked free net are calculated using the collinearity equation and the seven free net constraint equations. To improve positioning accuracy, the externally measured camera pose is substituted as a weighted observation in the calculation process, and the initial values ​​of the unknowns must conform to the geometric relationships between the circular targets in the world coordinate system. Finally, the object coordinates and covariance matrix of the target point are output.

[0108] In one embodiment, the step of solving the three-dimensional coordinates of a circular target at different times in a fixed free net, and calculating the deformation by subtracting the coordinates at different times, includes:

[0109] Images are captured at set times each day and transmitted in real time to a computer to calculate the coordinates of the measurement points. The image sequence acquired within a monitoring period is labeled 1, 2, 3, ... N. The deformation between two observations is calculated by subtraction.

[0110]

[0111] In one embodiment, the real-time transmission of the measured point coordinates to the computer for calculation includes:

[0112] The computer, industrial camera, and dual-antenna attitude and orientation module are connected via a pair of industrial-grade bridges. The computer and industrial camera are connected to the bridges via network cables, supporting long-distance shooting control and wireless data transmission.

[0113] The present invention also provides a device for detecting architectural deformation based on binocular vision, comprising:

[0114] The wireless transmission module is used for computer-controlled camera shooting, transmitting image data and camera pose data from the monitoring site to the computer. The target detection module is used to detect circular targets deployed on buildings and determine the sub-pixel coordinates of the circular targets in the image. The target localization module is used to construct a free network and calculate the three-dimensional coordinates of the circular targets under constraints. The deformation detection module is used to obtain the deformation at each monitoring time interval by subtracting the three-dimensional coordinates of the target points in the monitoring time series.

[0115] The building appearance deformation detection device based on binocular vision provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned building appearance deformation detection system based on binocular vision. For the sake of brevity, any parts not mentioned in the embodiment of the building appearance deformation detection device based on binocular vision can be referred to the corresponding content in the aforementioned building appearance deformation detection system based on binocular vision.

[0116] Furthermore, this application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, App application store, etc., which stores a computer program, and the program implements corresponding functions when executed by a processor. This embodiment of the computer-readable storage medium, when executed by a processor, implements a method based on binocular vision for detecting building exterior deformation.

[0117] Matters not covered in this invention are common knowledge.

[0118] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting architectural deformation based on binocular vision, characterized in that: The method includes the following steps: Step 1: Set up a circular target and ruler on the surface of the building, and use a binocular industrial camera to capture an image of the building containing the circular target and ruler, while simultaneously acquiring the pose of the industrial camera. Step 2: Use the ellipse fitting algorithm to extract the marker points at both ends of the circular target and ruler in the image, and determine the center of the circle based on the fitting results; Step 3: Using a circular target as an unknown point, construct a free net without control points using a scale, and determine the translational degrees of freedom, rotational degrees of freedom, and scale constraints of the free net; Step 4: Under the constraints, combined with the camera imaging equation, the three-dimensional coordinates of each circular target are obtained by using the Moore-Penrose generalized inverse matrix through the least squares principle. Step 5: In a fixed free network, calculate the three-dimensional coordinates of the circular target at different times, and calculate the deformation by subtracting the positions of two adjacent positioning results.

2. The method for detecting architectural appearance deformation based on binocular vision according to claim 1, characterized in that: In step 1, a circular target with a black border and a white circular pattern is selected. The circular targets are evenly distributed on the surface of the building. A fixed-length scale is selected, and circular markers are fixed at both ends of the scale. The computer controls the binocular industrial camera to acquire two images of the same observation area from different perspectives, and the industrial camera pose is obtained through a dual-antenna attitude-fixing and orientation module.

3. The method for detecting architectural appearance deformation based on binocular vision according to claim 1, characterized in that: In step 3, the target point set is treated as a rigid body, and the centroid coordinates of the target point set are constrained to remain unchanged, thus locking the overall translational degrees of freedom of the free net. Treat the target point set as a rigid body, constrain the resultant external torque of the target point set about the three principal axes X, Y, and Z to be zero, and lock the overall rotational degree of freedom of the free net; Based on a known length scale within the target area, the distance between the two endpoints of the scale is constrained to be equal to the known length, thus locking the scale factor of the free net.

4. The method for detecting architectural appearance deformation based on binocular vision according to claim 3, characterized in that: When the centroid of the constrained target point set remains unchanged, the sum of the coordinate corrections of all target points in the X, Y, and Z directions is zero.

5. The method for detecting architectural appearance deformation based on binocular vision according to claim 1, characterized in that: When the net external torque of the constrained target point set about the three principal axes X, Y, and Z is all zero, the sum of the torque vector formed by the coordinate correction of all target points and their corresponding approximate coordinates is zero.

6. The method for detecting architectural appearance deformation based on binocular vision according to claim 1, characterized in that: In step 4, the collinearity condition equation for camera imaging is linearized to obtain the error equation: in, Represents the exterior orientation correction vector of the image. This represents the vector of interior orientation correction elements of the camera. This represents the three-dimensional coordinate correction vector of the measurement point. , and This represents the corresponding coefficient matrix. This represents the difference between the measured value and the true value. The geometric expressions for the translational and rotational degrees of freedom of the constrained free net are transformed into the form of a linear constraint matrix by taking partial derivatives with respect to the object point coordinate corrections and the camera exterior orientation element corrections. in, Represents the linear constraint matrix. This represents a vector of unknowns consisting of camera exterior orientation element corrections and object point coordinate corrections. Transform the scale constraint into an error equation: in, Represents the scale constraint matrix. This represents the difference between the measured value and the true value. Using the camera pose measured by the dual-antenna attitude and orientation module as the externally measured camera exterior orientation element, the error equation is listed: in, Represents the identity matrix. This represents the difference between the measured value and the true value. List the comprehensive error equation: Using the least squares principle, substitute the constraint equations into: Find the minimum residual vector, and calculate the camera's exterior orientation elements, interior orientation elements, and the coordinates of the measurement points.

7. The method for detecting architectural appearance deformation based on binocular vision according to claim 1, characterized in that: In step 5, images are captured at regular intervals each day and transmitted in real time to a computer to calculate the coordinates of the measurement points. The image sequence acquired within a monitoring period is labeled as 1, 2, 3, ... N. The deformation between two observations is calculated by subtraction. in, , and These represent the coordinate differences between the previous time point and the next time point for the X, Y, and Z axes, respectively.

8. The method for detecting architectural appearance deformation based on binocular vision according to claim 7, characterized in that, The specific process of transmitting the coordinates of the measurement point to the computer in real time is as follows: the computer, industrial camera and dual-antenna attitude and orientation module are connected through a pair of industrial-grade bridges. The computer end and the industrial camera end are connected to the bridges with network cables, which supports long-distance shooting control and wireless data transmission.

9. A device for detecting architectural deformation based on binocular vision, characterized in that: It includes a binocular industrial camera module, a wireless transmission module, a target detection module, a target positioning module, and a deformation detection module; The binocular industrial camera module is wirelessly connected to the target detection module via a wireless transmission module. The binocular industrial camera module is used to capture images of buildings containing circular targets and rulers, and simultaneously acquire the industrial camera pose. The wireless transmission module is used for computer control of camera shooting, transmitting the image data and camera pose data from the monitoring site to the computer. The target detection module is connected to the deformation detection module via the target localization module. The target detection module is used to detect circular targets deployed on buildings and determine the subpixel-level coordinates of the circular targets in the image. The target localization module is used to construct a free net and calculate the three-dimensional coordinates of the circular targets under constraints. The deformation detection module is used to obtain the deformation at each monitoring time interval by subtracting the three-dimensional coordinates of the target points under the monitoring time series.

10. A computer-readable storage medium storing a computer program thereon, characterized in that: When the computer program is executed by the processor, it implements the binocular vision-based method for detecting architectural appearance deformation as described in any one of claims 1-8.

Images