Glass curtain wall construction error detection method based on computer vision

Abstract

The invention relates to the technical field of machine vision, and further relates to a glass curtain wall construction error detection method based on computer vision. The method comprises the following steps: step 1, acquiring an original image of the glass curtain wall through a camera; acquiring surface reflection characteristics of the glass curtain wall by adopting a polarization imaging technology; performing intensity reflection compensation on the original image based on the surface reflection characteristics to obtain a compensation image; 2, extracting a feature vector from the compensation image, and calculating a matching metric between points in the compensation image according to the feature vector; step 3, performing two-dimensional reconstruction correction according to the matching measurement between the points in combination with camera calibration parameters to obtain a two-dimensional point coordinate correction value corresponding to each point; and 4, according to the two-dimensional point coordinate correction value, comparing with a design standard template, analyzing deformation and evaluating an error. According to the invention, the error evaluation is more accurate.

Description

Technical Field

[0001] The invention belongs to the technical field of machine vision, and in particular relates to a glass curtain wall construction error detection method based on computer vision. Background Art

[0002] As an important part of modern architecture, glass curtain walls are widely used in high-rise buildings, commercial complexes and landmark buildings. Its main advantages include good light transmittance, aesthetics, and excellent thermal and sound insulation. However, due to the particularity of the glass curtain wall structure, the control of its construction errors has a vital impact on the overall quality, safety and aesthetics of the curtain wall. In the actual construction process, due to installation deviations, material processing errors, temperature stress deformation, and the influence of the structural support system, the glass curtain wall may produce partial or overall deformation. If this error is not effectively controlled, it may lead to problems such as poor sealing, decreased structural stability, and increased light pollution. Therefore, how to efficiently and accurately detect the construction errors of glass curtain walls has become a key technical problem in the quality control of curtain wall projects.

[0003] Traditional manual measurement mainly relies on measuring tools such as levels, steel rulers, and theodolites. Construction workers manually measure the key nodes and reference lines of the curtain wall and compare them with the design standards. Although this method has certain applicability, it has the following major problems: Limited measurement accuracy: Manual measurement relies on the experience of construction workers and is easily affected by environmental factors (such as wind, lighting conditions, etc.), resulting in large measurement errors, usually at the millimeter level, which makes it difficult to meet high-precision requirements. Inefficiency: For large-area curtain walls, manual measurement takes a long time, and the measurement points are limited, making it difficult to achieve comprehensive inspection of the entire curtain wall. Difficult data processing: Manual measurement data is usually recorded in tables, making it difficult to form an intuitive error distribution diagram, which is not conducive to subsequent analysis and construction adjustments.

[0004] With the development of optical measurement technology, laser scanning measurement has been gradually applied to curtain wall construction error detection. This method uses a laser scanner to collect point clouds on the curtain wall surface, and calculates the deviation between the curtain wall point cloud data and the design model to achieve error analysis. This method has high measurement accuracy (up to millimeter level) and can obtain the overall morphological information of the curtain wall. However, this technology still has the following problems: High equipment cost: Laser scanners are expensive and have high operating costs. They are not suitable for small and medium-sized projects or projects with limited construction budgets. Complex data processing: The large amount of point cloud data generated by laser scanning requires complex post-processing, including denoising, registration, reconstruction and other processes, which requires a large amount of calculation and places high demands on the computing equipment at the construction site. Susceptible to glass reflection: Glass curtain walls have the characteristics of high reflectivity, which causes laser scanning to produce signal loss or interference on the glass surface, affecting the accuracy of the data, especially under complex lighting conditions, the stability of the measurement results is difficult to guarantee. Summary of the invention

[0005] The main purpose of the present invention is to provide a glass curtain wall construction error detection method based on computer vision, which reduces the mirror reflection interference of the glass curtain wall through polarization imaging, uses a high-precision feature matching algorithm to ensure point stability, combines structured light measurement to calculate depth information, and uses an optimized error weight calculation model to make error evaluation more accurate. Compared with the prior art, the present invention not only improves the detection accuracy of glass curtain wall construction errors, reduces the impact of ambient light and material reflection on measurement, but also improves calculation efficiency and can adapt to different types of glass curtain wall materials.

[0006] In order to solve the above problems, the technical solution of the present invention is achieved as follows: A glass curtain wall construction error detection method based on computer vision, the method comprising: Step 1: Obtain the original image of the glass curtain wall through a camera; use polarization imaging technology to obtain the surface reflection characteristics of the glass curtain wall; perform intensity reflection compensation on the original image based on the surface reflection characteristics to obtain a compensated image; Step 2: Extract feature vectors from the compensated image, and calculate the matching metric between each point in the compensated image based on the feature vectors; Step 3: Perform 2D reconstruction correction based on the matching metrics between each point and the camera calibration parameters to obtain the 2D point coordinate correction value corresponding to each point; Step 4: Compare the 2D point coordinate correction values ​​with the design standard template, analyze the deformation and evaluate the error.

[0007] Furthermore, the surface reflection characteristics of the glass curtain wall are obtained through the following formula: ; in, Indicates the incident angle of the glass curtain wall surface and phase angle Surface reflectivity under represents the refractive index of air; Indicates the refractive index of the glass curtain wall, ranging from 1.5 to 1.7; represents the incident angle of the light; represents the phase angle; Indicates the angle of the principal axis of polarized light.

[0008] Furthermore, the original image is subjected to intensity reflection compensation based on the surface reflection characteristics by the following formula to obtain a compensated image: ; in, Indicates that the compensated image is at any point The strength of the place; The original image at any point The strength of the place; is the reflectivity of the internal interface of the glass curtain wall; Ambient light at any point The strength of the place.

[0009] Furthermore, in step 2, the feature vector is extracted from the compensated image by the following formula: ; in, The scale is Gaussian filter; express The gradient of Represents any point The eigenvector at .

[0010] Furthermore, in step 2, the matching metric between each point in the compensated image is calculated based on the feature vector by the following formula: ; in, Indicate point The eigenvector at ; Indicate point The eigenvector at ; Indicates that the compensated image is at point The strength of the place; Indicates that the compensated image is at point The strength of the place; Represents L1 norm operation; Represents the sensitivity parameter of image intensity matching; if the glass curtain wall is high-transparency flat glass, The value range is 8 to 12; if the glass curtain wall is frosted glass, The value range is 12 to 18; if the glass curtain wall is coated glass, The value range is 15 to 20; if the glass curtain wall is colored glass, The value range is 18 to 25; Indicate point With point The matching metric of ; by calculating the mean of the matching metrics between all points, the average matching metric is obtained .

[0011] Further, step 3: perform two-dimensional reconstruction correction based on the matching metric between each point and the camera calibration parameters to obtain the two-dimensional point coordinate correction value corresponding to each point: ; in, For any point The X-axis coordinate of For any point The Y-axis coordinate of Indicates the depth value; is the X-axis coordinate of the camera's principal point; Indicates the Y-axis coordinate of the camera's principal point; Indicates the position of the camera optical center; Indicates the location of the projector; Indicates the focal length of the camera.

[0012] Furthermore, the depth value Calculated by the following formula: ; ; ; in, Represents the angle between the camera's line of sight and the optical axis; It represents the angle between the projector's line of sight and the projection axis; Indicates the baseline distance between the camera and the projector; Represents the wavelength of structured light.

[0013] Furthermore, in step 4, the following formula is used to compare the two-dimensional point coordinate correction value with the design standard template to analyze the deformation and evaluate the error: ; ; in, Indicates that the design standard template is at coordinates The standard value at the point of is the difference; is the error value; The coordinates are The weight value of the point.

[0014] Furthermore, the weight value Use the following formula to express it: ; in, is the attenuation parameter of the spatial weight, is the set value; is the sensitivity parameter of the feature intensity weight, and is the set value; ; is the center value in the design standard template.

[0015] A spherical lattice shell structure node force analysis method based on refined simulation optimization of the present invention has the following beneficial effects: the high reflectivity of the glass curtain wall often leads to strong light interference during the imaging process, affecting the accuracy of feature extraction. The present invention adopts polarization imaging technology to obtain the reflection characteristics of the curtain wall surface, and uses a mathematical model to compensate the reflection intensity of the image, thereby significantly reducing the influence of mirror reflection, improving the image quality, and enabling the feature points to be stably extracted. The matching accuracy of the traditional computer vision method is low in the glass curtain wall environment, while the present invention adopts a feature extraction method based on gradient information and Gaussian filtering, and combines matching metric calculation to improve the stability and accuracy of point matching, ensuring that the subsequent error calculation is based on high-quality data. The optical reflection characteristics of the curtain wall surface are calculated by a mathematical model, and the reflection intensity of the image is compensated based on the model, ensuring that the system can still obtain high-quality feature images even under different lighting conditions. The traditional binocular stereo vision method is prone to failure in the glass curtain wall environment, while the present invention combines structured light projection, uses matching metrics to calculate the depth information of the points, and optimizes the calculation through the geometric relationship between the camera and the projector, ensuring that accurate depth data can be obtained even in a high-reflection environment. In the error calculation, a spatial attenuation weight based on Gaussian distribution is introduced to ensure that the error weights of key areas (such as curtain wall edges and joints) are higher, while the weights of flat areas with less impact on the error are lower, thereby optimizing the overall error evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a method flow of a spherical lattice shell structure node force analysis method based on refined simulation optimization provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0018] Example 1, reference Figure 1 : A glass curtain wall construction error detection method based on computer vision, the method comprising: Step 1: Obtain the original image of the glass curtain wall through a camera; use polarization imaging technology to obtain the surface reflection characteristics of the glass curtain wall; perform intensity reflection compensation on the original image based on the surface reflection characteristics to obtain a compensated image; The main component of the glass curtain wall is silicate material, which has a smooth surface and high reflectivity. When light is incident on the glass surface, partial reflection and partial transmission will occur. This reflection can be divided into two forms: one is specular reflection, that is, the light is reflected regularly according to the law of reflection; the other is diffuse reflection, that is, due to the microscopic unevenness of the glass surface, part of the light is reflected in an irregular direction. The impact of these two reflections on the computer vision system is completely different. Diffuse reflection is effective information that can be used by computer vision technology because it can truly reflect the physical form of the glass surface, while specular reflection is interference information that needs to be compensated and removed. Since the reflection characteristics of the glass curtain wall change with the incident angle and the refractive index of the material, when performing error detection, it is first necessary to accurately obtain the reflection characteristics of the glass curtain wall, and based on this, perform reflection compensation of the image to obtain more reliable visual information.

[0019] In this method, polarization imaging technology is introduced to effectively separate the specular reflection component and diffuse reflection component of the glass curtain wall. When light is reflected on the glass surface, the electric field vibration direction of part of the light wave will change, causing the reflected light to exhibit certain polarization characteristics. Using this physical phenomenon, images under different polarization directions can be obtained through a polarization camera or an additional polarization filter, and the reflectivity of the glass curtain wall under specific incident angles and phase angles can be calculated through a specific mathematical model. Polarization filters usually use a variable angle adjustment mode to obtain the reflection of the glass curtain wall at multiple polarization angles, thereby establishing a complete reflection characteristic model. This model can describe the reflection behavior of glass under different incident conditions and provide the necessary physical parameters for subsequent image compensation.

[0020] After polarization imaging obtains the surface reflection characteristics of the glass curtain wall, the computer vision system needs to perform intensity reflection compensation on the original image based on these characteristics to remove the interference of specular reflection, so that the image retains mainly the real information of the curtain wall surface. The core idea of ​​compensation is to separate the specular reflection component by establishing a mathematical model, and restore the diffuse reflection information of the curtain wall surface in a reasonable way. Since the intensity of specular reflection is closely related to the refractive index of the glass, the angle of incidence and the wavelength of light, the computer vision system needs to combine these parameters and use the physical compensation formula to correct each pixel in the image. Specifically, the real intensity of each pixel can be normalized by deducting the part affected by specular reflection and combining the reflectivity of the internal interface of the glass to restore its proper illumination information. In this way, the computer vision system can effectively suppress specular reflection, improve the contrast and clarity of the image, and provide more reliable basic data for subsequent feature extraction and error detection. Polarization imaging technology can separate the effective information of the curtain wall in a complex lighting environment, so that the computer vision system can still accurately perceive the real form of the curtain wall under a strong reflection background. In addition, traditional image de-reflection methods usually rely on multi-angle imaging or high dynamic range imaging, which not only increase the complexity of the system, but are also difficult to apply in glass curtain wall detection scenarios. The present invention adopts a method of combining polarization imaging with physical modeling, which only requires single-view imaging to achieve efficient reflection compensation, allowing the error detection system to work stably in ordinary construction site environments. In addition, this method avoids the uncertainty that may be introduced by data-driven models by utilizing the physical optical properties of glass curtain walls for image compensation, thereby improving the versatility and robustness of the system.

[0021] Step 2: Extract feature vectors from the compensated image, and calculate the matching metric between each point in the compensated image based on the feature vectors; First, after the reflection compensation, the glass curtain wall may still have local illumination changes and residual reflection effects. In order to ensure the stability of feature extraction, the system needs to enhance the edge features based on the gradient information of the image. Gradient calculation is one of the basic means of feature extraction in computer vision. It can effectively capture the edge information of the image, that is, the area where the pixel intensity changes dramatically. In the scene of the glass curtain wall, the edge information often corresponds to the structural features of the curtain wall, such as the seams of the glass panels, the fixed frame of the curtain wall, and the local slight deformation. Therefore, by calculating the gradient amplitude and direction of the image, the basic features of the glass curtain wall surface can be obtained to enhance its recognizability in the matching calculation. After obtaining the basic gradient features, in order to further enhance the robustness of the system, it is necessary to introduce Gaussian filtering for smoothing. The surface of the glass curtain wall may produce slight noise due to illumination, pollutants or processing errors. If these noises are directly involved in feature extraction, they may affect the matching accuracy. The role of the Gaussian filter is to process the image in scale space, so that the system can stably extract features at different scales and reduce the influence of local noise. Especially in the glass curtain wall scene, considering the different reflection characteristics of different materials (such as transparent glass, frosted glass, and coated glass), Gaussian filtering can play a certain normalization role, so that the surface features of curtain walls of different materials can be better processed in a unified manner. In addition, since the edge feature intensity of the glass curtain wall is greatly affected by the reflectivity, when calculating the gradient feature, it is necessary to normalize the feature value in combination with the reflection characteristic parameters to avoid the uneven feature extraction caused by the different reflectivities of different materials.

[0022] After completing feature extraction, the system needs to calculate the matching metric between each point in the image to establish a feature relationship at the pixel level. Matching metric is an important method used in computer vision to measure the similarity between two pixels or regions. In the glass curtain wall error detection task, the calculation of the matching metric directly determines the accuracy of the subsequent reconstruction correction, so its calculation method needs to be optimized in combination with the characteristics of the glass curtain wall material. The calculation of the matching metric is usually based on two core principles, one is the similarity of the feature vector, and the other is the similarity of the pixel intensity. The similarity of the feature vector can be calculated by the cosine similarity metric to ensure that the extracted feature vector can maintain a certain alignment in both direction and amplitude. The similarity of pixel intensity needs to be measured based on the Gaussian model to measure the impact of the brightness difference between the two points on the matching metric. In order to adapt to different types of glass curtain wall materials, the method of the present invention introduces different sensitivity parameters in the matching metric calculation. The transparency, surface roughness and coating condition of the glass curtain wall will affect the calculation of the matching metric, so in the calculation process, it is necessary to set the corresponding matching parameters according to different glass materials. For example, on a highly transparent glass curtain wall, the matching metric needs to pay more attention to the local texture information, while on a frosted glass or coated glass curtain wall, it needs to pay more attention to the overall structural information. Therefore, by adaptively adjusting the matching parameters, this method can ensure high-precision feature matching on different types of glass curtain walls.

[0023] Step 3: Perform 2D reconstruction correction based on the matching metrics between each point and the camera calibration parameters to obtain the 2D point coordinate correction value corresponding to each point; In the imaging process of computer vision, any object in the real world will be projected onto a two-dimensional image plane through the camera. This imaging process is affected by the internal and external parameters of the camera. The internal parameters of the camera include focal length, principal point coordinates, distortion coefficient, etc., while the external parameters involve the rotation and displacement of the camera. Since the construction error detection of glass curtain walls depends on accurate coordinate calculation, in order to ensure the detection accuracy, the system needs to compensate and correct these effects. Especially in the scene of glass curtain walls, there may be nonlinear distortions in the imaging process, such as lens distortion, projection deviation, and optical refraction. If coordinate correction is not performed, it may cause positioning deviations of certain key points in the error evaluation process, thereby affecting the reliability of error calculation. Therefore, the system first needs to use the camera calibration data to perform distortion correction on the original image, remove the geometric distortion caused by lens deformation, so that the points in the image can more accurately reflect the real projection position of the curtain wall. After completing the distortion correction, the system needs to combine the matching metric to optimize the coordinate correction of the points. Matching metrics are mathematical indicators used to measure the similarity between points in an image. They can help the system determine which points are stable and reliable and which points may drift due to lighting or reflection factors. In the scene of glass curtain wall construction error detection, due to the influence of the curtain wall surface material, installation accuracy and ambient lighting, the image acquired by the camera may have a certain degree of distortion and deviation. Therefore, when correcting the coordinates, it is necessary to focus on the points with low matching degrees and use the surrounding neighborhood points with high matching degrees for interpolation correction, so that the point coordinates of the entire image are more consistent, thereby improving the accuracy of error calculation. The calculation of matching metrics usually combines the similarity of feature vectors and the similarity of pixel intensities. Through these metrics, it is possible to effectively determine which points in the image can still maintain high stability after transformation and which points need further adjustment. In addition to using matching metrics to optimize coordinate correction, this method also combines the projection relationship of the camera to perform precise geometric transformation of the points. Since the structure of the glass curtain wall is usually relatively regular, and the construction error is generally manifested as local deformation or overall displacement, the projection transformation model can be used to adjust the coordinates to ensure that the distribution of the points in the image is more consistent with the actual geometric form of the curtain wall. In computer vision, projection transformation is a mathematical method to map a point from the original coordinate system to another coordinate system. It can adjust the coordinates of the points in the image based on the imaging model and calibration parameters of the camera so that its position is closer to the curtain wall structure in the real world. In the implementation of this method, the core of the projection transformation is to use the point pair information obtained by the matching metric calculation, and through the optimal transformation calculation, all points are aligned as much as possible in the transformed coordinate system, thereby improving the accuracy of error detection.

[0024] Step 4: Compare the 2D point coordinate correction values ​​with the design standard template, analyze the deformation and evaluate the error.

[0025] The design standard template of the glass curtain wall is a predefined ideal geometric model, which is usually accurate data generated according to the architectural design plan, including key information such as the standard size, installation angle, and joint spacing of each unit of the curtain wall. During the construction process, due to the influence of factors such as the external environment, material errors, and installation technology, the curtain wall actually installed often has a certain deviation from the theoretical design. Therefore, it is necessary to use an error detection system to determine whether the construction accuracy of the curtain wall meets the design requirements. In the method of the present invention, the basic principle of error calculation is to compare the corrected two-dimensional point coordinates with the corresponding coordinates of the standard template, calculate the difference between the two, and use a mathematical model to quantify and visualize the error, so that the construction personnel can intuitively understand the offset of the curtain wall and take necessary adjustment measures. In the process of calculating the error, the system first needs to establish an error measurement model, which is used to measure the degree of offset between the actual coordinates of each point and the design standard coordinates. The error of the glass curtain wall can be divided into multiple dimensions, such as horizontal translation error, vertical offset error, and local deformation error. In order to improve the accuracy of error calculation, this method introduces an error calculation model based on weight distribution. This model can assign different calculation weights according to the error contribution of different areas to ensure that the measurement results are more in line with the actual situation. For example, in the edge area of ​​the curtain wall, due to the large stress concentration effect during the installation process, the error may be relatively large. Therefore, when calculating the error, it is necessary to assign higher weights to these areas to ensure that the system can more accurately reflect the overall error distribution of the curtain wall. For the central area of ​​the curtain wall, due to the strong rigid support of the material, the error is usually small, so the system can assign lower weights to the error calculation of these areas to reduce the calculation complexity and improve the stability of the analysis.

[0026] After the error calculation is completed, the system needs to further analyze and visualize the error results so that the construction personnel can intuitively understand the error distribution and make construction adjustments accordingly. This method uses color coding to visually mark the errors, that is, different colors are used in the test results according to the size and direction of the errors, so that the construction personnel can quickly determine which areas have larger errors and which areas have smaller errors. For example, for areas with larger errors, red can be used for marking, and for areas with smaller errors, green can be used for marking. In this way, construction personnel can quickly identify the key error areas on the curtain wall surface and adjust the construction process according to the test results to improve the overall installation accuracy of the curtain wall. In addition, during the error analysis process, the system can also evaluate the overall error trend through statistical methods to determine the distribution law of construction errors. For example, if the error shows a certain linear change trend on the entire curtain wall surface, it may indicate that there is a systematic offset in the overall installation of the curtain wall, which may be caused by the error of the measurement reference point or the offset of the installation reference line. If the error shows a local irregular distribution, it may indicate that there is material deformation or uneven force on the supporting structure in the local area of ​​the curtain wall. Therefore, through statistical analysis of error distribution, the system can provide construction managers with more in-depth error assessment results, allowing them to take targeted construction optimization measures to ensure that the installation quality of the curtain wall meets the design standards.

[0027] Embodiment 2: Obtain the surface reflection characteristics of the glass curtain wall by the following formula: ; in, Indicates the incident angle of the glass curtain wall surface and phase angle Surface reflectivity under represents the refractive index of air; Indicates the refractive index of the glass curtain wall, ranging from 1.5 to 1.7; represents the incident angle of the light; represents the phase angle; Indicates the angle of the principal axis of polarized light.

[0028] Specifically, in the formula, Represents the glass curtain wall surface at a given incident angle and phase angle Reflectivity under certain conditions. It is used to describe the ratio of the intensity of reflected light to the intensity of incident light when light is incident on the surface of the glass curtain wall. Since glass is a typical transparent material, the reflection behavior of its surface depends not only on the incident angle of light, but also on the difference in refractive index between glass and air. Therefore, the formula includes the material refractive index and The calculation items represent the refractive index of air and glass respectively, where the refractive index of air is usually taken as , and the refractive index of the glass curtain wall The value is arrive The difference in refractive index will affect the reflection intensity of light on the glass surface, resulting in different reflection characteristics of different types of glass curtain walls. For example, ordinary float glass, low-emissivity coated glass or dimming glass all show different reflection behaviors. The molecular part of the formula Mainly used to describe the glass curtain wall surface at different phase angles The reflection characteristics under the action of It reflects the influence of the phase angle of light on the reflectivity. In the process of polarization imaging measurement of glass curtain wall, the phase angle It directly affects the intensity and polarization direction of the reflected light, so this term plays a role in characterizing the reflection characteristics of polarized light in the formula. When different values ​​are taken, the polarization direction of the light changes, causing the reflectivity of the glass surface to change accordingly. In addition, the denominator of the formula It is used for normalization processing, so that the calculation results are always kept within a reasonable physical range, avoiding numerical overflow or non-compliance with optical laws. In addition, the last term of the formula Further consideration is given to the angle of the polarization axis of light Impact on reflectivity. Since glass curtain walls are usually installed at different angles in buildings and the direction of incident light changes over time, it is necessary to introduce the main axis angle of polarized light. to ensure that the calculated reflectivity accurately matches the actual optical measurement. When , the value of this item is the largest, indicating that the polarization direction is completely consistent with the reflection behavior of the glass surface, making the reflectivity calculation more accurate. Deviation When the angle changes, the reflectivity will gradually decrease, which reflects the law of optical reflection change of the glass curtain wall under different incident conditions.

[0029] Embodiment 3: The original image is subjected to intensity reflection compensation based on the surface reflection characteristics by the following formula to obtain a compensated image: ; in, Indicates that the compensated image is at any point The strength of the place; The original image at any point The strength of the place; is the reflectivity of the internal interface of the glass curtain wall; Ambient light at any point The strength of the place.

[0030] Specifically, in this formula, Represents the compensated image intensity, that is, after reflection compensation, each pixel The real lighting information at the original image Affected by specular reflection, the brightness in some areas increases abnormally, making it difficult to distinguish the characteristic details of the curtain wall surface. Therefore, it is necessary to correct it through computer vision methods. The basic principle of correction is to use the surface reflectivity of the glass curtain wall The contribution of reflected light to the original image is calculated and subtracted from the image to restore the true illumination distribution on the curtain wall surface. It is calculated by the aforementioned optical model, which describes the light at different incident angles and phase angle The reflection intensity under the condition of the glass curtain wall determines the specular reflection contribution of the glass curtain wall to the ambient light. is an important variable, which represents the light intensity in the environment where the glass curtain wall is located. Since the glass curtain wall is usually in a complex outdoor environment, the ambient light source may include sunlight, sky light, reflected light from surrounding buildings, etc., so its impact is complex and difficult to predict. This method obtains the impact information of ambient light through polarization imaging technology and estimates it using a physical model, making the compensation calculation more accurate. In the formula, This item indicates that the glass curtain wall is at a certain pixel point At , the ambient light contribution due to specular reflection is the interference component that needs to be removed from the original image. Therefore, the numerator of the formula The image information remaining after deducting the reflected light is calculated. However, the reflection of light on the glass curtain wall does not only occur on the surface layer, but may also involve multiple reflections on the internal interface of the glass. This is because glass is a semi-transparent medium, and part of the incident light will penetrate the surface into the interior of the glass and reflect at the internal interface, further affecting the brightness of the final image. Therefore, the denominator of the formula It is used to normalize the overall illumination distribution to compensate for the brightness deviation caused by internal reflection of the glass. At the same time, in order to further reduce the impact of internal interface reflection of the glass, the formula also introduces the internal interface reflectivity of the glass Make corrections, including This item represents the correction of the effect of internal reflection of the glass on the overall brightness, so that the final compensated image can be closer to the real surface information of the glass curtain wall. This method not only removes reflections based on the optical properties of the glass curtain wall, but also combines the actual ambient lighting effects to make the compensation effect more accurate. Compared with traditional image processing methods, such as histogram equalization or adaptive filtering, which are only based on image statistical features, the physical modeling method of this method can more accurately separate the reflected light component and effectively restore the real visual information of the glass curtain wall. In addition, this method can also adapt to different types of glass curtain wall materials, such as transparent glass, low-e glass, coated glass, etc., because its compensation calculation is based on the physical parameters of refractive index and reflectivity, and these parameters can be adjusted according to the specific material, thereby ensuring that the compensation model can be applied to the detection needs of different curtain wall structures. In practical applications, this reflection compensation model can significantly improve the accuracy of error detection, so that the computer vision system can still accurately identify the structural characteristics of the glass curtain wall in complex lighting environments. For example, under strong sunlight, an uncompensated image may have large brightness unevenness due to excessive specular reflection, resulting in reduced accuracy of edge detection and feature matching. However, after being processed by the compensation formula, the brightness distribution of the image will be more balanced, allowing the system to more reliably extract effective information about the curtain wall surface, thereby improving the accuracy of error detection and construction quality assessment.

[0031] Embodiment 4: In step 2, the feature vector is extracted from the compensated image by the following formula: ; in, The scale is Gaussian filter; express The gradient of Represents any point The eigenvector at .

[0032] Specifically, in this formula, Represents any pixel The feature vector calculated at is used to describe the local structural information of the pixel on the surface of the glass curtain wall. In computer vision tasks, feature vectors are usually used to match and align corresponding points in different images, so their stability is crucial for subsequent matching calculations. In order to extract stable feature information, this method first calculates the compensated image The gradient of . The role of gradient calculation is to highlight the edge and texture information in the image and enhance the structural features of the curtain wall surface, so that the system can more accurately identify the boundaries and deformation areas of the curtain wall. Since the features of glass curtain walls are usually sparse, the effect of directly using grayscale values ​​as the matching basis is poor, and the gradient information can better describe the local changes in the image, thereby improving the effect of feature extraction. However, gradient calculation itself is easily affected by noise, especially in the glass curtain wall detection scenario. Due to the local illumination changes, pollutants or construction errors on the curtain wall surface, there may be slight texture interference, which may cause over-response or local anomalies in the gradient calculation results. Therefore, this method introduces a Gaussian filter. Smooth the gradient calculation results. Gaussian filtering is a commonly used image processing method that can reduce the interference of random noise while retaining the main features of the image, making the feature extraction process more robust. In this formula, the scale parameter of the Gaussian filter is Need to be adjusted according to the specific application scenario, usually smaller Values ​​are used to extract detailed features, while larger The value is used to extract global structural information. In the glass curtain wall error detection task, the system can adaptively adjust the to optimize the effect of feature extraction.

[0033] In addition, due to the reflectivity of the glass curtain wall This will affect the result of gradient calculation, so this method also introduces a normalization term in the feature extraction process. , which is used to compensate for the uneven gradient value caused by reflection. The reflectivity of the glass curtain wall varies greatly at different angles, especially in the case of large-angle incident light, the reflectivity of some areas may be close to 1, resulting in reduced contrast in these areas, thus affecting the accuracy of gradient calculation. Therefore, the role of this normalization term is to amplify the gradient response of the low-reflection area to ensure that the system can still extract stable feature vectors even under high-reflection conditions. In this way, the system can adapt to different types of glass curtain wall materials, including transparent glass, coated glass, and low-emissivity glass, and maintain the stability of feature extraction under different lighting environments. This feature extraction method combines the dual advantages of physical modeling and image processing, so that the feature extraction process of the glass curtain wall can not only reduce reflection interference, but also enhance effective information and ensure the stability of the feature vector under different environments. Compared with traditional feature extraction methods, such as SIFT (Scale Invariant Feature Transform) or SURF (Speeded Up Robust Features), this method is specifically optimized for the special optical properties of the glass curtain wall, avoiding the influence of high-smoothness surfaces on the stability of feature point matching. At the same time, since this method is based on image gradient calculation, it avoids the complex multi-scale feature extraction process, so the calculation amount is lower and can meet the real-time requirements in practical engineering applications.

[0034] Embodiment 5: In step 2, the matching metric between each point in the compensated image is calculated according to the feature vector by the following formula: in, Indicate point The eigenvector at ; Indicate point The eigenvector at ; Indicates that the compensated image is at point The strength of the place; Indicates that the compensated image is at point The strength of the place; Represents L1 norm operation; Represents the sensitivity parameter of image intensity matching; if the glass curtain wall is high-transparency flat glass, The value range is 8 to 12; if the glass curtain wall is frosted glass, The value range is 12 to 18; if the glass curtain wall is coated glass, The value range is 15 to 20; if the glass curtain wall is colored glass, The value range is 18 to 25; Indicate point With point The matching metric of ; by calculating the mean of the matching metrics between all points, the average matching metric is obtained .

[0035] Specifically, the first part of the formula is the cosine similarity calculation based on the feature vector: , the purpose of this part of the calculation is to measure the degree of similarity of the feature vectors of two points in the direction. Since the surface of the glass curtain wall is usually smooth, its local area may lack obvious texture information, so the direct matching method based on pixel intensity may not be able to accurately distinguish the similarity of different points, and the use of feature vectors can more stably extract the local structural information of the glass curtain wall surface. In this formula, and Respectively indicate points and The eigenvector at , and the denominator Normalization is performed so that the calculation result depends only on the direction of the feature vector and is not affected by the amplitude of the feature vector. This normalization method is called cosine similarity, which is widely used in feature matching and image retrieval tasks in the field of computer vision. Specifically, the range of cosine similarity is When the directions of the two eigenvectors are exactly the same, the value is 1, indicating that the local structures of the two points are highly similar; when the directions of the two vectors are completely opposite, the value is -1, indicating that the two do not match at all; when the two vectors are orthogonal, the value is 0, indicating that there is no correlation between the two. In the error detection task of glass curtain walls, eigenvectors are usually calculated based on gradient information and Gaussian filtering. Therefore, this matching method can maintain good stability under different lighting conditions and reduce mismatching in highly reflective environments.

[0036] However, relying solely on the similarity of feature vectors is not enough to ensure the accuracy of matching calculations, because the surface reflection characteristics of the glass curtain wall may cause the gradient information in some areas to be distorted, thus affecting the stability of the feature vectors. Therefore, in order to further enhance the robustness of matching, this method introduces a weighted correction factor based on image intensity differences in the matching calculation, namely: ; This item is used to measure the similarity of the light intensity of two points on the compensated image, and adjusts the matching confidence through exponential decay. Intuitively, this weighting factor can be understood as follows: if the difference in light intensity between two points on the compensated image is small, their matching metric value should be high, and if the difference in light intensity between the two points is large, the matching metric value should be low. The form of the exponential function ensures that the matching degree decreases nonlinearly with the increase of the light intensity difference, so that the system can more effectively ignore the changes in light intensity caused by local lighting changes or different glass curtain wall materials. For example, in some areas of the glass curtain wall, the local brightness may increase due to the influence of external ambient light, and the exponential function can automatically reduce the matching confidence of these areas, thereby reducing the matching error. Parameters It plays a role in sensitivity adjustment in this calculation, which determines the degree of response of the system to changes in light intensity. The optimal value of this parameter is different for different glass curtain wall materials. For example, for high-transparency glass, light propagation is mainly transmission, and the surface reflectivity is relatively low. Therefore, when matching calculations, the system needs to be more sensitive to smaller light intensity differences. The value of should be small (8 to 12). For frosted glass, due to its strong scattering characteristics, the local variation of light intensity is large, so the matching calculation requires a lower sensitivity to avoid excessive matching errors caused by changes in illumination. The value range should be between 12 and 18. Similarly, for coated glass, due to its complex surface structure, the reflection characteristics vary greatly with the incident angle, so The value of should be larger (15 to 20) to ensure the stability of the matching calculation. For colored glass, since its surface color may affect the perception of light intensity, the matching calculation needs to be more robust. The value range should be between 18 and 25. , this method can automatically optimize the matching calculation for glass curtain walls of different materials, thereby ensuring the stability and accuracy of the matching results. In addition, in order to evaluate the global characteristics of the entire matching calculation, this method also introduces the mean calculation of the matching metric, that is, by calculating the average matching metric between all points , the matching trend of the entire curtain wall surface can be obtained. The statistical analysis of the matching metric mean can be used to detect the overall construction error of the glass curtain wall. For example, when the matching metric mean is high, it indicates that the overall shape of the glass curtain wall is consistent with the design standard, while when the matching metric mean is low, it may mean that the curtain wall has large errors in some areas. In addition, the spatial distribution of the matching metric mean can also be used to identify the local deformation of the curtain wall surface. For example, if the matching metric of a certain area is significantly lower than the global mean, it indicates that the structural characteristics of the area deviate greatly from the standard template and may need local correction.

[0037] Embodiment 6: Step 3: Perform two-dimensional reconstruction correction based on the matching metric between each point and the camera calibration parameters to obtain the two-dimensional point coordinate correction value corresponding to each point: ;in, For any point The X-axis coordinate of For any point The Y-axis coordinate of Indicates the depth value; is the X-axis coordinate of the camera's principal point; Indicates the Y-axis coordinate of the camera's principal point; Indicates the position of the camera optical center; Indicates the location of the projector; Indicates the focal length of the camera; is the transpose symbol.

[0038] Specifically, the core variables of the formula are , which represents the corrected two-dimensional point coordinates, that is, the point position after matching measurement and projection geometry correction. Since the installation error of the glass curtain wall usually affects the overall shape of the curtain wall on a large scale, the error detection system needs to calculate the true position of each point as accurately as possible, so as to provide accurate coordinate information for subsequent error calculation. When the camera's imaging characteristics need to be considered, represents the optical center position of the camera, and is the focal length of the camera. It affects the camera's field of view and determines the perspective projection relationship in the image. In the error detection task of glass curtain walls, short-focal-length cameras are usually used to ensure that a larger range of curtain wall areas can be acquired while taking into account resolution. However, short-focal-length cameras are prone to cause more obvious perspective distortion, so when performing coordinate correction, it must be compensated by using the projection transformation method. In order to achieve this compensation, the following projection scaling factor is introduced into the formula: ; This part of the calculation describes the point The scaling transformation relationship in the camera coordinate system. Represents the depth value, represents the relative distance between the camera and the projector, and the denominator Indicate point The perspective scaling factor on the camera imaging plane. The purpose of this scaling factor is to adjust the position of the point in the image according to its projection depth, so that the coordinate correction can fully consider the spatial distribution of the curtain wall points. In the error detection task of the glass curtain wall, different points may have different projection scaling ratios due to different depths, so this scaling factor must be used to compensate for it to ensure the accuracy of the point calculation.

[0039] Next, the last term of the formula: ; This part is used to describe the point Direction vector in the camera coordinate system. Since the error detection of glass curtain walls depends on accurate point matching, and the perspective projection relationship of the camera may cause certain distortion of the points in the image, it is necessary to calculate the true projection direction of the point coordinates through the direction vector so that the corrected coordinates can more accurately reflect the actual position of the curtain wall. The calculation method of the direction vector is based on the optical projection model of the camera, where and Respectively indicate points Relative to the camera principal point The offset of Represents the focal length of the camera, ensuring the normalized calculation of the direction vector, so that the coordinate correction process will not be affected by projection distortion. The application of this formula in the error detection task of glass curtain wall construction enables the system to more accurately map the actual coordinates of the curtain wall points, thereby improving the accuracy of error calculation. Compared with the traditional method based on plane projection, this method combines the depth information obtained by matching calculation and uses projection transformation for coordinate correction, so that the system can adapt to different types of glass curtain wall materials and maintain high calculation stability under complex lighting conditions. In addition, by introducing camera calibration parameters, this method enables error detection to adapt to different camera devices, thereby improving the applicability of the system. In practical applications, this two-dimensional reconstruction correction method can significantly improve the accuracy of curtain wall error detection. For example, in ordinary construction measurement, due to the possible deviation of the camera's shooting angle, error calculation directly based on the original image may result in a large coordinate offset. After coordinate correction by this method, the system can accurately correct the point to the real coordinate system of the camera, thereby reducing the influence of perspective distortion on error detection. In addition, this method can adapt to different types of glass curtain wall materials, such as transparent glass, low-e glass, coated glass, etc., and can maintain high calculation stability under complex lighting conditions.

[0040] Example 7: Depth Value Calculated by the following formula: ; ; ; in, Represents the angle between the camera's line of sight and the optical axis; It represents the angle between the projector's line of sight and the projection axis; Indicates the baseline distance between the camera and the projector; Represents the wavelength of structured light.

[0041] Specifically, this method uses the principle of triangulation to establish a The projection geometry model is based on the fixed physical distance between the camera and the projector as the baseline, so that the fringe information generated by the structured light projection is directly corresponding to the depth of the actual object surface, thus solving the problem that the traditional stereo vision method is easily disturbed and difficult to obtain depth information on highly reflective surfaces such as glass curtain walls. In the formula, the depth is calculated by matching the geometric positions of the camera and the projector. The formula form is The physical meaning of this expression is to convert the actual physical baseline Combined with the sinusoidal function relationship between two angles, it constitutes a typical triangulation model.

[0042] Specifically, the projector projects a light pattern of a specific wavelength onto the glass curtain wall through structured light. Due to its special optical properties, the glass curtain wall will cause displacement and deformation of the projected pattern in different areas. After the preliminary matching measurement and feature extraction, this deformation information can reflect the offset of the pattern in the image. Using these offsets, the corresponding line of sight angle can be calculated, thereby deriving the depth of each point relative to the camera. Here, the angle is calculated by the camera imaging model, which reflects the The deviation from the camera's optical axis, which is calculated by the distance between the point and the camera's principal point and the focal length The size of this angle directly determines the scaling ratio of the point under perspective projection. It comes from structured light technology, which measures the average matching Structured light wavelength And the baseline distance The function relationship is inversely solved to obtain the angle between the projector's line of sight and the projection axis. In fact, the advantage of the structured light system in glass curtain wall construction error detection is that it uses a light source with a known wavelength and a fixed geometric structure to measure the depth information of the object surface through the distortion of the fringe pattern, which is particularly important for glass materials with high reflectivity and low texture.

[0043] The entire depth calculation process not only requires accurate measurement of the physical distance between the camera and the projector, but also relies on accurate calibration of the camera's internal parameters, such as focal length. and the principal point coordinates The accurate calibration of these parameters is the premise of building a correct projection model. Only on this basis can the pixel position in the two-dimensional image be mapped back to the actual object surface by using the perspective projection principle, thereby achieving accurate depth reconstruction. Through the sine function involved in the formula, it can be seen that when the angle between the camera line of sight and the optical axis is When the depth changes The angle between the projector's line of sight and the projection axis is also adjusted accordingly, which reflects the size difference between near and far objects and the effect of perspective offset in perspective projection. It plays a role in adjusting the sensitivity of depth calculation. The numerator of its sine function combines the wavelength of structured light with the matching metric, so that the depth information can remain stable under different lighting conditions and material reflection characteristics. In practical applications, glass curtain walls often have complex reflection phenomena and optical refraction problems due to their material characteristics, which makes traditional depth calculation methods susceptible to interference and produce large errors. The present invention, by introducing structured light technology and combining the precise calibration of cameras and projectors, can not only overcome the challenges brought by high reflectivity, but also maintain high measurement accuracy in complex environments. Using this triangulation method, the depth information of each pixel can be accurately obtained, and then combined with the subsequent two-dimensional coordinate correction step, high-precision measurement of the entire error detection system can be achieved. It is worth noting that in the actual measurement process, the installation angle and relative position of the camera and projector may deviate due to changes in construction site conditions. Therefore, the system usually needs to have self-correction capabilities in order to dynamically adjust these parameters based on real-time collected data. This is also a major innovation of the present invention. In addition, using the matching metric As calculation The depth calculation can adaptively reflect the subtle differences in the local structure of the glass curtain wall surface based on the multi-level and multi-parameter depth calculation method. When the curtain wall surface is locally deformed due to construction errors or material defects, the matching metric will show corresponding abnormal changes, which are directly transmitted to the depth calculation, so that the system can detect the depth deviation of the local area and further perform error analysis. It is through this multi-level and multi-parameter combined depth calculation method that the present invention realizes high-precision evaluation of deformation and installation errors in glass curtain wall construction error detection, ensuring the scientificity and reliability of the detection results.

[0044] Embodiment 8: In step 4, the following formula is used to compare the two-dimensional point coordinate correction value with the design standard template to analyze the deformation and evaluate the error: ; ; in, Indicates that the design standard template is at coordinates The standard value at the point of is the difference; is the error value; The coordinates are The weight value of the point.

[0045] Specifically, by calculating the actual measurement point Design standard points The difference , which reflects the deviation between the actual state and the ideal state of the curtain wall at each coordinate position, and then combined with the weight function and reflectivity compensation factor Further correct the error influence to get the overall error This method not only takes into account the local geometric deformation, but also weights the importance of different areas, making the error evaluation results more in line with the actual application needs. In the actual construction process, the glass curtain wall is affected by various factors such as installation process, material properties and ambient light, and often produces obvious deformation or offset in some local areas. If these errors are not quantified, it is difficult to judge whether the construction quality meets the design requirements. Through this formula, the deviation of each point can be Calculated precisely, is the actual measured coordinate obtained after image compensation, feature extraction and coordinate correction in the previous steps, and The ideal coordinates are pre-set according to the design standard template. In actual measurement, some points may be affected by optical distortion, changes in ambient light or reflection interference, which may cause noise or errors in the measured values. Therefore, it is necessary to identify the actual degree of deviation from the design by calculating the difference of each point. Next, the error calculation does not only stop at the difference itself, but also further reflects the real impact of the error by weighting the sum of the squares of the differences.

[0046] Weight function It plays a vital role here. It allocates each point according to its position, importance and possible error sensitivity in the curtain wall. It often gives a higher weight to the edge area or key joints of the curtain wall because these areas are usually subjected to greater forces or are more obvious in visual effects, while the weight in the center or symmetrical area may be lower to balance the distribution of the overall error. In addition, in order to compensate for the uneven effect of the reflective characteristics of the glass curtain wall surface on image acquisition, the reflectivity compensation factor is also introduced in the formula. This is mainly to amplify the areas where the measurement signal is attenuated due to high reflectivity, so that the final error value can more objectively reflect the actual construction error. and phase angle The surface reflectivity In the high-reflectivity area, the original signal may be overly suppressed. By introducing reflectivity compensation, the difference in these areas can be appropriately amplified during calculation to ensure that the error evaluation will not be distorted by optical characteristics. Finally, the overall error is obtained by taking the square root of the weighted sum of squares of the difference. This process actually combines all local deviations to form a global error index, which reflects both the severity of local deformation and the contribution of each area to the overall error. In glass curtain wall construction, this error index can be directly used to determine whether the installation needs to be adjusted or whether there are serious construction defects that require rework.

[0047] Example 9: Weight Value Use the following formula to express it: ; in, is the attenuation parameter of the spatial weight, is the set value; is the sensitivity parameter of the feature intensity weight, and is the set value; ; is the center value in the design standard template.

[0048] Specifically, the weight calculation formula used in this method consists of two main parts, which describe the importance of spatial position and the influence of feature gradient. The first term is a spatial weight term based on Gaussian decay: ; The physical meaning of this term is that according to the point Center point with design standard template The Euclidean distance between the two points is weighted in the form of Gaussian distribution, so that the points far from the center have a smaller weight, while the points close to the center have a larger weight. The parameter here is Controls the decay rate of the weight. The larger its value, the lower the sensitivity of the system to the spatial position. When it is smaller, the weight will decay rapidly, indicating that the system pays more attention to the central area. Since the construction error of the glass curtain wall may show certain distribution characteristics as a whole, for example, during the installation process, due to the influence of the supporting structure and the fixed frame, the error may be more obvious in the local area. Therefore, through this Gaussian weight term, the distribution of the error in space can be effectively reflected, and it is ensured that the error calculation can more accurately evaluate the overall deformation of the curtain wall. However, it is not enough to rely solely on the spatial position for weighting, because the error of the glass curtain wall depends not only on the coordinate offset, but also on the changes in surface features. For example, at the edge or joint of the curtain wall, the way the material is connected may cause the local deformation to be more obvious, so the influence of the characteristic gradient needs to be considered additionally. To this end, this method introduces a second weight calculation term based on the characteristic gradient: ; This item is used to measure the point The intensity of the feature change at , and weighted by exponential decay. Here, Indicate point The characteristic gradient at the position can describe the local changes of the glass curtain wall surface, and the parameter This controls the sensitivity of the feature weights. When is smaller, it means that the system is more sensitive to local feature changes, and when When the value is larger, the system responds less to feature changes. Since the feature changes on the curtain wall surface are often concentrated in the boundary area or joints, the introduction of this weight can effectively highlight the influence of these areas and reduce the error weight of the flat area, so that the system can pay more attention to key parts when calculating errors without being affected by the uniform surface. By combining these two weights, the final weight value is calculated. It can simultaneously reflect the positional relationship of points in space and the changes in their local features. Especially for the splicing area of ​​the glass curtain wall, since the error in this area has a greater impact on the stability of the overall structure, in this method, the weight of this area is often higher, so as to ensure that the error calculation can more accurately reflect the construction quality of the curtain wall. For the flat area in the center of the curtain wall, since its error has a smaller impact on the overall structure, the weight value will be relatively low, thereby reducing the influence of irrelevant areas in the error calculation process, making the system calculation more efficient and accurate. The advantage of this weight calculation method is that it not only considers spatial factors, but also combines the actual feature information of the glass curtain wall surface, so that the error calculation is more in line with the physical characteristics of the curtain wall. Compared with the traditional uniform weight distribution method, this method can more accurately identify the key areas of the error and improve the detection accuracy through a reasonable weighting method. In addition, since the weight calculation adopts Gaussian distribution and exponential decay function, the calculation process has high stability and can adapt to different types of glass curtain wall materials and construction environments. In practical applications, this method can effectively improve the accuracy of error detection, so that the system can not only remain stable in a highly reflective environment, but also automatically adjust the weight distribution according to different curtain wall structures to ensure the rationality of the error assessment results.

[0049] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. 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 invention.

Claims

1. A glass curtain wall construction error detection method based on computer vision, characterized in that: The method comprises: Step 1: Obtain the original image of the glass curtain wall through a camera; use polarization imaging technology to obtain the surface reflection characteristics of the glass curtain wall; perform intensity reflection compensation on the original image based on the surface reflection characteristics to obtain a compensated image; Step 2: Extract feature vectors from the compensated image, and calculate the matching metric between each point in the compensated image based on the feature vectors; Step 3: Perform 2D reconstruction correction based on the matching metrics between each point and the camera calibration parameters to obtain the 2D point coordinate correction value corresponding to each point; Step 4: Compare the 2D point coordinate correction values ​​with the design standard template, analyze the deformation and evaluate the error.

2. The glass curtain wall construction error detection method based on computer vision as claimed in claim 1, characterized in that: The surface reflection characteristics of the glass curtain wall are obtained through the following formula: ； in, Indicates the incident angle of the glass curtain wall surface and phase angle Surface reflectivity under represents the refractive index of air; Indicates the refractive index of the glass curtain wall, ranging from 1.5 to 1.7; represents the incident angle of the light; represents the phase angle; Indicates the angle of the principal axis of polarized light.

3. The glass curtain wall construction error detection method based on computer vision as claimed in claim 2, characterized in that: The following formula is used to perform intensity reflection compensation on the original image based on the surface reflection characteristics to obtain a compensated image: ； in, Indicates that the compensated image is at any point The strength of the place; The original image at any point The strength of the place; is the reflectivity of the internal interface of the glass curtain wall; Ambient light at any point The strength of the place.

4. The glass curtain wall construction error detection method based on computer vision as claimed in claim 3, characterized in that: In step 2, the feature vector is extracted from the compensated image using the following formula: ； in, The scale is Gaussian filter; express The gradient of Represents any point The eigenvector at .

5. The glass curtain wall construction error detection method based on computer vision as claimed in claim 4, characterized in that: In step 2, the matching metric between each point in the compensated image is calculated based on the feature vector using the following formula: ； in, Indicate point The eigenvector at ; Indicate point The eigenvector at ; Indicates that the compensated image is at point The strength of the place; Indicates that the compensated image is at point The strength of the place; Represents L1 norm operation; Represents the sensitivity parameter of image intensity matching; if the glass curtain wall is high-transparency flat glass, The value range is 8 to 12; if the glass curtain wall is frosted glass, The value range is 12 to 18; if the glass curtain wall is coated glass, The value range is 15 to 20; if the glass curtain wall is colored glass, The value range is 18 to 25; Indicate point With point The matching metric of The average matching metric is obtained by calculating the mean of the matching metrics between all points .

6. The glass curtain wall construction error detection method based on computer vision as claimed in claim 5, characterized in that: Step 3: According to the matching metric between each point and the camera calibration parameters, two-dimensional reconstruction correction is performed to obtain the expression of the two-dimensional point coordinate correction value corresponding to each point: ； in, For any point The X-axis coordinate of For any point The Y-axis coordinate of Indicates the depth value; is the X-axis coordinate of the camera's principal point; Indicates the Y-axis coordinate of the camera's principal point; Indicates the position of the camera optical center; Indicates the location of the projector; Indicates the focal length of the camera; is the transpose symbol.

7. The glass curtain wall construction error detection method based on computer vision as claimed in claim 6, characterized in that: Depth Value Calculated by the following formula: ； ； ； in, Represents the angle between the camera's line of sight and the optical axis; It represents the angle between the projector's line of sight and the projection axis; Indicates the baseline distance between the camera and the projector; Represents the wavelength of structured light.

8. The glass curtain wall construction error detection method based on computer vision as claimed in claim 7, characterized in that: In step 4, the following formula is used to compare the two-dimensional point coordinate correction value with the design standard template to analyze the deformation and evaluate the error: ； ； in, Indicates that the design standard template is at coordinates The standard value at the point of is the difference; is the error value; The coordinates are The weight value of the point.

9. The glass curtain wall construction error detection method based on computer vision as claimed in claim 8, characterized in that: Weight value Use the following formula to express it: ； in, is the attenuation parameter of the spatial weight, is the set value; Indicate point The characteristic gradient at is the sensitivity parameter of the feature intensity weight, and is the set value; ; is the center value in the design standard template.

Images