Photovoltaic bracket three-dimensional model construction method and system based on aerial images

Abstract

The present invention relates to the technical field of three-dimensional model construction, and specifically to a method and system for constructing a three-dimensional model of a photovoltaic bracket based on aerial images. Taking into account the influence of reflection noise, a reflection suppression factor is first determined to separate the reflection noise source and retain the diffuse reflection base texture; then, the geometric significance of each pixel point is comprehensively determined in combination with the edge direction consistency determined by the rigidity and periodicity characteristics of the photovoltaic bracket; further, in combination with the characteristics that the arrangement of the photovoltaic bracket will make the edge chain present parallelism and continuity, the reference continuity of the significant edge chain and the topological correlation between the significant edge chain and the adjacent significant edge chain are determined; finally, based on the geometric significance of each pixel point and the reference continuity and topological correlation of the edge chain in which it is located, the edge points of the bracket main beam are screened out; thereby, a more accurate three-dimensional model of the photovoltaic bracket is constructed based on the bracket main beam edge points.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional model construction, and in particular to a method and system for constructing a three-dimensional model of a photovoltaic bracket based on aerial images. Background Art

[0002] The construction of a three-dimensional model of photovoltaic brackets is crucial in the planning and maintenance of photovoltaic power stations. Existing technologies typically use aerial photography to construct three-dimensional models of photovoltaic brackets. However, aerial images are easily affected by reflections from the photovoltaic panels, resulting in a large number of exposed pixels in the image. This makes it difficult to distinguish between reflection noise and actual structural edges during subsequent feature extraction and model reconstruction, seriously interfering with modeling accuracy. Furthermore, for weakly textured areas within photovoltaic brackets and panels, existing methods for directly modeling three-dimensionally based on aerial images lack effective utilization of the bracket's rigidity and periodicity, making it difficult to accurately capture the bracket's structural features. Consequently, the constructed three-dimensional models fail to meet the precision requirements of engineering applications. Consequently, the accuracy of photovoltaic bracket three-dimensional models constructed using existing technologies is low. Summary of the Invention

[0003] In order to solve the technical problem of low accuracy of the three-dimensional model of photovoltaic brackets constructed by the existing technology, the purpose of this application is to provide a method and system for constructing a three-dimensional model of photovoltaic brackets based on aerial images. The technical solutions adopted are as follows:

[0004] The first aspect of the present application provides a method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images, comprising:

[0005] Acquire an aerial image of a photovoltaic bracket; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic bracket; divide the aerial image of the photovoltaic bracket into at least two local image blocks;

[0006] In each local image block, the geometric saliency of each pixel is determined based on the polarization gradient modulus of each pixel, the overall deviation of the polarization gradient modulus, and the overall similarity of the normal vector direction corresponding to each pixel in the photovoltaic support point cloud;

[0007] Based on the extension trend of the polarization gradient direction of the pixel points, significant edge chains are determined; the corresponding reference continuity is determined based on the length, curvature and geometric significance distribution of the significant edge chains; and the corresponding topological correlation is determined based on the overall deviation between each significant edge chain and its adjacent edge chains.

[0008] According to the geometric significance, the reference continuity and the topological correlation, the comprehensive structural confidence of each pixel point is determined; the edge points of the bracket main beam are screened out according to the comprehensive structural confidence; and the three-dimensional model of the photovoltaic bracket is constructed according to the edge points of the bracket main beam.

[0009] Furthermore, the process of determining the polarization gradient modulus and polarization gradient direction of each pixel point in the photovoltaic bracket aerial image includes:

[0010] The polarization image of the spectral bracket is collected by an aerial drone; the two polarization components of the aerial image of the photovoltaic bracket are obtained by using the Stokes vector based on the polarization images of all photovoltaic brackets; the polarization gradient modulus of each pixel is determined based on the maximum value of the gradient modulus corresponding to the two polarization components.

[0011] The polarization component corresponding to the polarization gradient modulus corresponding to each pixel point is used as the reference polarization component; the gradient direction of each pixel point under the corresponding reference polarization component is used as the polarization gradient direction of each pixel point.

[0012] Furthermore, the process of obtaining the geometric saliency includes:

[0013] In each local image block, a corresponding reflection suppression factor is determined according to the deviation of the polarization gradient modulus of each pixel compared to the overall polarization gradient modulus of the local image block;

[0014] A dense point cloud is generated using photogrammetry software based on multi-viewpoint aerial images of the photovoltaic bracket to obtain the photovoltaic bracket point cloud; the corresponding edge direction consistency is determined based on the similarity between the normal vector direction corresponding to each pixel point in the photovoltaic bracket point cloud and the normal vector direction of the entire pixel point in the local image block;

[0015] The geometric saliency of each pixel is determined according to the product of the reflection suppression factor, the edge direction consistency and the polarization gradient modulus of each pixel.

[0016] Furthermore, the process of obtaining the edge direction consistency includes:

[0017] Obtain the point cloud data point corresponding to the location of each pixel point in the photovoltaic support point cloud; form a corresponding neighboring point cloud set based on all point cloud data points within a preset neighborhood range of the point cloud data point corresponding to each pixel point; perform SVD decomposition on the covariance matrix constructed by the neighboring point cloud set, and use the normal vector corresponding to the minimum eigenvalue obtained after the decomposition as the reference normal vector for each pixel point;

[0018] The direction corresponding to the vector obtained by vector accumulation operation of the reference normal vectors of all pixels in each local image block is used as the corresponding main direction of the bracket; the angle between the main direction of the bracket and the polarization gradient direction of each pixel point is negatively correlated and mapped to determine the edge direction consistency of each pixel point.

[0019] Furthermore, the process of obtaining the reflection inhibition factor includes:

[0020] A polarization gradient modulus distribution coordinate system is constructed, with the direction in which the polarization gradient modulus lengths are arranged from small to large as the horizontal axis and the direction in which the number of pixels is arranged from small to large as the vertical axis. A polarization gradient modulus distribution curve is constructed based on the polarization gradient modulus lengths of all pixels in each local image block in combination with the polarization gradient modulus distribution coordinate system. Half of the mean between the maximum polarization gradient modulus length and the minimum polarization gradient modulus length in the polarization gradient modulus distribution curve is used as a peak screening value.

[0021] Among all extreme points on the polarization gradient modulus distribution curve whose polarization gradient modulus is greater than the peak screening value, the maximum point with the largest value in the polarization gradient modulus distribution curve is taken as the main peak point; the polarization gradient modulus corresponding to the minimum point closest to the main peak point is taken as the first main peak boundary value; the polarization gradient modulus corresponding to the minimum point closest to the main peak point is taken as the second main peak boundary value; and all pixel points whose corresponding polarization gradient modulus is greater than the first main peak boundary value and less than the second main peak boundary value are taken as main peak pixel points;

[0022] In each local image block, the sum of the positive correlation mapping value of the standard deviation of the polarization gradient modulus corresponding to all main peak pixels and the mean of the polarization gradient modulus of all main peak pixels is used as the gradient failure threshold of each local image block;

[0023] The difference between the polarization gradient modulus of each pixel in each local image block and the gradient failure threshold is negatively correlated to determine the reflection suppression factor of each pixel.

[0024] Furthermore, the process of obtaining the significant edge chain includes:

[0025] The next pixel point of each pixel point in the corresponding polarization gradient direction is used as the direction determination point of each pixel point;

[0026] Take each pixel as the target point in turn;

[0027] The target point is used as the starting search point. When the angle between the polarization gradient direction of the target point and the polarization gradient direction of the corresponding direction determination point is less than or equal to a preset angle threshold, the corresponding direction determination point is used as the edge chain point of the target point. The edge chain point is used as a new search point, and iteration is continued until a preset iteration stop condition is reached. The target point is sequentially connected with all edge chain points obtained during the iteration process to determine the corresponding significant edge chain.

[0028] The preset iteration stop condition includes: the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the corresponding direction determination point is greater than a preset angle threshold, or the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the target point is greater than a preset deviation threshold;

[0029] Changing the target point obtains all salient edge chains in the saliency image.

[0030] Furthermore, the reference continuity acquisition process includes:

[0031] The minimum geometric saliency of all pixels on each salient edge chain is used as the reference weak saliency;

[0032] The reference continuity of each salient edge chain is determined according to the product of the negative correlation mapping value of the average curvature of each salient edge chain, the number of pixels and the reference weak saliency.

[0033] Furthermore, the process of obtaining the topological association degree includes:

[0034] The unit vector corresponding to the polarization gradient direction of each pixel is used as the gradient vector; the gradient vectors of all pixels on each significant edge chain are summed up to determine the edge direction vector of each significant edge chain;

[0035] Obtain all adjacent edge chains for each significant edge chain; the Euclidean distance between the centroid of each significant edge chain and the centroid of each corresponding adjacent edge chain is less than a preset distance threshold;

[0036] Normalize the angle between the edge direction vector of each significant edge chain and the edge direction vector of each corresponding adjacent edge chain to determine the corresponding adjacent direction deviation;

[0037] Normalizing the Euclidean distance between the centroid of each significant edge chain and the centroid of each corresponding adjacent edge chain to determine a corresponding adjacent reference distance; and using the difference between the adjacent reference distance and a preset installation spacing as the corresponding adjacent distance deviation;

[0038] Determining a corresponding continuity deviation value according to the difference between the reference continuity of each significant edge chain and the reference continuity of each corresponding adjacent edge chain;

[0039] A negative correlation mapping is performed on the product of the adjacent direction deviation, the adjacent distance deviation and the continuity deviation value to determine the topological association between each significant edge chain and each corresponding adjacent edge chain.

[0040] Furthermore, the process of obtaining the comprehensive structure confidence includes:

[0041] Determine the corresponding proximity regularity based on the mean of the topological correlation between the significant edge chain where each pixel is located and all corresponding adjacent edge chains;

[0042] Normalizing the product of the reference continuity of the significant edge chain where each pixel point is located and the proximity regularity to determine the edge chain regularity of each pixel point;

[0043] The comprehensive structural confidence of each pixel is determined according to the product of the edge chain regularity and the geometric significance of each pixel.

[0044] In a second aspect, the present application provides a system for constructing a three-dimensional model of a photovoltaic bracket based on aerial images, the system comprising:

[0045] An image acquisition and preprocessing module is used to acquire polarization images, aerial images, and point clouds of photovoltaic supports using an aerial photography drone; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic support based on the polarization image of the photovoltaic support; and divide the aerial image of the photovoltaic support into at least two local image blocks;

[0046] The first determination module is used to determine, in each local image block, a corresponding reflection suppression factor based on the deviation of the polarization gradient modulus of each pixel point compared to the overall polarization gradient modulus of the local image block; determine the corresponding edge direction consistency based on the similarity of the normal vector direction corresponding to each pixel point in the photovoltaic support point cloud compared to the normal vector direction of the overall pixel point of the local image block; determine the corresponding geometric saliency based on the polarization gradient modulus of each pixel point, the reflection suppression factor and the edge direction consistency;

[0047] The second determination module is used to determine all significant edge chains based on the extension trend of the polarization gradient direction corresponding to the pixel points in the aerial image of the photovoltaic bracket; determine the corresponding reference continuity based on the length and curvature of each significant edge chain and the geometric significance distribution of the corresponding pixel points; and determine the corresponding topological correlation based on the polarization gradient direction deviation, relative distance and reference continuity deviation between each significant edge chain and the corresponding adjacent edge chain;

[0048] The photovoltaic bracket three-dimensional model construction module is used to determine the comprehensive structural confidence of each pixel point based on the geometric significance of each pixel point, the reference continuity of the significant edge chain in which it is located, and the overall size of the topological correlation between it and the adjacent edge chains; screen out the edge points of the bracket main beam according to the said comprehensive structural confidence; and construct the photovoltaic bracket three-dimensional model based on the said bracket main beam edge points.

[0049] In a third aspect, the present application provides a computer device comprising a memory and a processor. The memory is configured to store computer program code, and the processor is configured to call and execute the computer program code from the memory to perform the method of the first aspect or any embodiment of the first aspect of the present application.

[0050] In a fourth aspect, the present application provides a computer program product, comprising a computer program code. When the computer program code is executed, the method of the first aspect or any embodiment of the first aspect of the present application is performed.

[0051] In a fifth aspect, the present application provides a computer-readable storage medium, which stores computer program code. When the computer program code is executed, it performs the method of the first aspect of the present application or any embodiment of the first aspect.

[0052] This application has the following beneficial effects:

[0053] This application takes into account the influence of reflection noise and the characteristic that metal reflection will generate pseudo edges with high gradient values ​​in non-edge areas. First, the reflection suppression factor is determined to separate the reflection noise source and retain the diffuse reflection base texture; then, the geometric significance of each pixel point is comprehensively determined based on the edge direction consistency determined by the rigidity and periodicity characteristics of the photovoltaic bracket to highlight the structural characteristics of the photovoltaic bracket; further, the arrangement of the photovoltaic bracket will make the edge chain present parallelism and continuity, and the reference continuity of the significant edge chain and the topological correlation between adjacent significant edge chains in parallel are analyzed; finally, based on the bracket structural characteristics expressed by the geometric significance of each pixel point and the photovoltaic bracket arrangement edge characteristics expressed by the reference continuity and topological correlation of the edge chain, the edge points of the bracket main beam are screened out; thereby, a more accurate three-dimensional model of the photovoltaic bracket is constructed based on the edge points of the bracket main beam, so that the obtained three-dimensional model of the photovoltaic bracket strictly conforms to the rigidity, parallelism and periodicity characteristics of the bracket. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 A flowchart of a method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images provided by one embodiment of the present invention;

[0056] Figure 2 A structural diagram of a photovoltaic bracket three-dimensional model construction system based on aerial images provided by one embodiment of the present invention;

[0057] Figure 3 The present invention provides a schematic diagram of a computer device structure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following is a detailed description of a method and system for constructing a three-dimensional model of a photovoltaic bracket based on aerial images proposed by the present invention, in combination with the accompanying drawings and preferred embodiments, its specific implementation method, structure, characteristics and effects are as follows. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment, and the specific features, structures or characteristics in one or more embodiments may be combined in any suitable form. In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0059] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0060] The following describes in detail a method and system for constructing a three-dimensional model of a photovoltaic bracket based on aerial images provided by the present invention with reference to the accompanying drawings.

[0061] This application embodiment provides a method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images. Figure 1 , which shows a flow chart of a method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images provided by one embodiment of the present invention, the method comprising:

[0062] Step S101: Acquire an aerial image of a photovoltaic bracket; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic bracket; and divide the aerial image of the photovoltaic bracket into at least two local image blocks.

[0063] In one specific implementation of an embodiment of the present invention, an aerial drone is used to capture the target area to be modeled to obtain an aerial image of the photovoltaic bracket. The multi-viewpoint aerial images of the photovoltaic bracket in the target area are then used to generate a dense point cloud using photogrammetry software (such as ContextCapture or Pix4D), thereby obtaining the desired photovoltaic bracket point cloud for the target area. Simultaneously, during the drone's flight, polarization filters at four angles (0°, 45°, 90°, and 135°) are used to capture polarized images of the photovoltaic bracket corresponding to the target location. These images are then used to determine the polarization component, thereby determining the polarization gradient modulus and polarization gradient direction at each pixel, in conjunction with existing technologies.

[0064] Preferably, in some possible implementations of the embodiments of the present invention, the process of determining the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic bracket according to the polarization image of the photovoltaic bracket includes:

[0065] Based on the polarization images of all photovoltaic brackets, the two polarization components of the aerial image of the photovoltaic bracket are obtained through the Stokes vector. Specifically, the polarization components include polarization component S1 and polarization component S2. Based on the pixel value of each pixel position at the polarization component, the gradient modulus and gradient direction corresponding to the current pixel in polarization component S1 and polarization component S2 are obtained. The polarization component S1 is usually calculated from the difference between the polarization directions of 0° and 90°, reflecting the difference in linear polarization between the horizontal and vertical directions, reflecting the surface directionality. The polarization component S1 is usually calculated from the difference between the polarization directions of 45° and 135°, supplementing the surface anisotropy information. The gradient modulus at each pixel position obtained through this operation can suppress sky specular reflection (such as glare on the photovoltaic panel surface) to a certain extent, preserving the bracket structure.

[0066] Considering that each pixel in the aerial image of the photovoltaic support corresponds to two gradient moduli and gradient directions in two polarization components, further screening is performed to determine the polarization gradient modulus of each pixel based on the maximum gradient modulus corresponding to each pixel in the two polarization components. The polarization component corresponding to the polarization gradient modulus of each pixel is used as the reference polarization component, and the gradient direction of each pixel in the corresponding reference polarization component is used as the polarization gradient direction of each pixel. In other words, the polarization gradient modulus and gradient direction corresponding to the most obvious gradient modulus and the gradient direction corresponding to this gradient modulus are used to determine the polarization gradient modulus and polarization gradient direction corresponding to each pixel, thereby minimizing the influence of specular reflection and preserving the support structure.

[0067] The aerial image of the photovoltaic bracket is further divided into blocks to determine each local image block, and each local image block is processed to better capture the local texture features of mirror reflections such as glare on the surface of the photovoltaic panel, so as to analyze the impact of overexposure on each local image block, so as to make the representation of the geometric significance of each pixel more accurate. In a specific implementation method of an embodiment of the present invention, the aerial image of the photovoltaic bracket is evenly divided and the size of each divided local image block is set to 21×21. The size of the local image block can be adjusted according to the specific implementation environment, and the pixel points that appear at the corner positions during the division process and cannot form a local image block are processed as a whole as a local image block, which will not be further elaborated here.

[0068] Step S102: In each local image block, the geometric saliency of each pixel is determined based on the polarization gradient modulus of each pixel, the overall deviation of the polarization gradient modulus, and the overall similarity of the normal vector direction corresponding to each pixel in the photovoltaic support point cloud.

[0069] For local image blocks affected by photovoltaic brackets, the true edge gradient amplitude of the photovoltaic bracket is usually in the medium and high range, while metal reflections will form pseudo edges with high gradient values ​​in non-edge areas (such as bright spots of mirror reflection). Therefore, if it is necessary to distinguish between the true structure edge and the reflective noise area, it is necessary to set a threshold between the true edge gradient amplitude of the photovoltaic bracket and the pseudo edge with high gradient value for analysis. Then, based on the set threshold, the texture performance of the pixel point is determined according to the size of the polarization gradient modulus of the pixel point relative to the threshold, that is, whether the pixel point has lost texture information due to overexposure. Therefore, in each local image block, the corresponding reflection suppression factor is determined according to the deviation of the polarization gradient modulus of each pixel point compared to the overall polarization gradient modulus of the local image block.

[0070] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the reflection suppression factor includes:

[0071] A polarization gradient modulus distribution coordinate system is constructed, with the direction of polarization gradient modulus arranged from small to large as the horizontal axis and the direction of pixel number arranged from small to large as the vertical axis. A polarization gradient modulus distribution curve is constructed based on the polarization gradient modulus of all pixels in each local image block in combination with the polarization gradient modulus distribution coordinate system. Half of the mean between the maximum polarization gradient modulus and the minimum polarization gradient modulus in the polarization gradient modulus distribution curve is used as the peak screening value. Among all extreme points on the polarization gradient modulus distribution curve where the polarization gradient modulus is greater than the peak screening value, the maximum point with the largest value in the polarization gradient modulus distribution curve is used as the main peak point. The polarization gradient modulus corresponding to the minimum point closest to the main peak point is used as the first main peak boundary value. The polarization gradient modulus corresponding to the minimum point closest to the main peak point is used as the second main peak boundary value. All pixels whose corresponding polarization gradient modulus is greater than the first main peak boundary value and less than the second main peak boundary value are used as the main peak pixels. It should be noted that when there is no minimum point before the main peak point, the polarization gradient modulus corresponding to the peak screening value is used as the first main peak boundary value; when there is no minimum point after the main peak point, the maximum polarization gradient modulus is used as the second main peak boundary value, which can be adjusted according to the specific implementation environment.

[0072] The rigid structure of the photovoltaic bracket (such as a metal beam) forms a continuous and stable edge in the image, so its gradient modulus is concentrated in a medium or high range. Therefore, among all the peaks, the height of the peak corresponding to the polarization gradient modulus interval corresponding to the edge formed by the rigid structure of the photovoltaic bracket is higher than the height of the peak formed by other edges. Therefore, the peak corresponding to the main peak point is the peak corresponding to the continuous and stable edge formed by the rigid structure of the photovoltaic bracket (such as a metal beam) in the image. According to this feature, all the main peak pixels belonging to the main peak are determined in combination with the valley points corresponding to the minimum points on both sides of the main peak point, so as to further analyze the main peak pixels. The gradient amplitude of the reflective pseudo-edge is bimodal, and the corresponding low peak is the normal gradient modulus interval of the base texture under the influence of diffuse reflection, and the high peak is the abnormal gradient modulus interval of overexposure and loss of texture information under the influence of mirror reflection. Therefore, by further analyzing the numerical distribution of the polarization gradient modulus of the main peak pixel point, the gradient failure threshold that can screen out the abnormal gradient modulus interval of overexposure and loss of texture information under the influence of mirror reflection can be determined. It should be noted that the purpose of setting the peak screening value is to prevent the occurrence of pixel points with low polarization gradient modulus length from interfering with the acquisition of the main peak point, which will not be further elaborated here.

[0073] Therefore, further in each local image block, the sum of the positive correlation mapping value of the standard deviation of the polarization gradient modulus corresponding to all main peak pixels and the mean of the polarization gradient modulus of all main peak pixels is used as the gradient failure threshold of each local image block. In a specific implementation of an embodiment of the present invention, the method of performing positive correlation mapping on the standard deviation of the polarization gradient modulus corresponding to all main peak pixels is adopted: multiplying the standard deviation of the polarization gradient modulus corresponding to all main peak pixels by 2.5 to obtain the value after positive correlation mapping, so that based on the principle of normal distribution, after obtaining the gradient failure threshold by adding the standard deviation of the polarization gradient modulus corresponding to all main peak pixels and the mean of the polarization gradient modulus, the gradient failure threshold is made to include more than 95% of the edges of the main peak pixels as much as possible, thereby judging whether the current local image block has lost local texture information due to overexposure.

[0074] The difference between the polarization gradient modulus of each pixel point in each local image block and the gradient failure threshold is negatively correlated and mapped to determine the reflection suppression factor of each pixel point. For each local image block, if the polarization gradient modulus of the pixel point is larger than the corresponding gradient failure threshold, it means that the pixel point is more likely to be a pixel point that is overexposed and loses texture information under the influence of mirror reflection; therefore, for each pixel point, the larger the reflection suppression factor calculated, the less likely the pixel point is to lose texture information due to overexposure, and the more credible the corresponding texture information. In a specific implementation of an embodiment of the present invention, a method for negatively correlating the difference between the polarization gradient modulus of each pixel point in each local image block and the gradient failure threshold is adopted: after linearly normalizing the difference between the polarization gradient modulus of each pixel point in each local image block and the gradient failure threshold, the value 1 is subtracted from the normalized value to obtain the negative correlation mapping result, that is, the reflection suppression factor.

[0075] There are some weak texture areas in photovoltaic brackets and photovoltaic panels. Therefore, it is necessary to use the rigidity and periodicity characteristics of the photovoltaic bracket to determine whether each pixel point belongs to the edge structure of the photovoltaic bracket. Based on the photovoltaic bracket point cloud, the normal vector of the point cloud data point corresponding to each pixel point can be determined. Due to the rigidity and periodicity characteristics of the photovoltaic bracket, the directions of the normal vectors corresponding to the pixels belonging to the photovoltaic bracket in each local image block are generally similar. Therefore, the corresponding edge direction consistency is further determined based on the similarity between the normal vector direction corresponding to each pixel point in the photovoltaic bracket point cloud and the normal vector direction of the entire pixel point in the local image block.

[0076] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining edge direction consistency includes:

[0077] Obtain the point cloud data point corresponding to the location of each pixel in the photovoltaic bracket point cloud; form a corresponding neighbor point cloud set based on all the point cloud data points within the preset neighborhood range of the point cloud data point corresponding to each pixel; perform SVD decomposition on the covariance matrix constructed by the neighbor point cloud set, and use the normal vector corresponding to the minimum eigenvalue obtained after the decomposition as the reference normal vector of each pixel. It should be noted that the SVD decomposition of the covariance matrix constructed by the neighbor point cloud set into the process of determining the normal vector by principal component analysis is a technical means well known to those skilled in the art and will not be further defined or elaborated here. According to the characteristic that the directions of the normal vectors corresponding to the pixels belonging to the photovoltaic bracket are generally similar, the reference normal vectors of all pixels in each local image block are vector-accumulated to obtain the direction corresponding to the vector, thereby determining the main direction of the bracket corresponding to the entire local image block, and characterizing the rigid structure reference direction of the photovoltaic bracket; therefore, the angle between the main direction of the bracket and the polarization gradient direction of each pixel is further negatively correlated to determine the edge direction consistency of each pixel. The greater the deviation of the polarization gradient direction of the pixel point from the main direction of the bracket, the less the pixel point conforms to the rigid structure reference direction of the photovoltaic bracket, the less likely it is to belong to the edge structure of the photovoltaic bracket, and the smaller the geometric significance should be; that is, the edge direction consistency and geometric significance are positively correlated.

[0078] In one specific implementation of an embodiment of the present invention, a method for performing negative correlation mapping on the angle between the main direction of the support and the polarization gradient direction of each pixel is performed by raising the negative number of one-quarter of the angle between the main direction of the support and the polarization gradient direction of each pixel to the power of an exponential function with a natural constant as the base. The output of the exponential function is the value after the negative correlation mapping. Those skilled in the art may use other basic mathematical methods to implement this method, which is not limited or elaborated herein.

[0079] For each pixel point, the smaller its corresponding reflection suppression factor is, the greater the impact of mirror reflection noise is, and the lower the corresponding edge texture confidence is; the greater the edge direction consistency is, the higher the fit between the edge direction of the pixel point and the bracket rigid set is, and the more likely it is to belong to the edge of the photovoltaic bracket; the larger the polarization gradient modulus of the pixel point is, the more significant the corresponding edge texture is. Since this application needs to determine the edge of the photovoltaic bracket more accurately, it is necessary to introduce the reflection suppression factor and edge direction consistency on the basis of the polarization gradient modulus of each pixel point to correct the significant features of the edge texture represented by the polarization gradient modulus, so that the geometric significance obtained for each pixel point can represent the significance of the corresponding pixel point on the edge of the photovoltaic bracket, thereby improving the accuracy of the subsequent construction of the three-dimensional model of the photovoltaic bracket.

[0080] In a specific implementation of an embodiment of the present invention, the geometric significance of each pixel point is determined based on the product of the reflection suppression factor, the edge direction consistency and the polarization gradient modulus of each pixel point, so that the greater the geometric significance, the more prominent the edge information of the corresponding pixel point and the higher the possibility that the corresponding edge texture belongs to the edge of the photovoltaic bracket.

[0081] Step S103: Determine the significant edge chain based on the extension trend of the polarization gradient direction of the pixel point; determine the corresponding reference continuity based on the length, curvature and geometric significance distribution of the significant edge chain; determine the corresponding topological correlation based on the overall deviation between each significant edge chain and the adjacent edge chain.

[0082] In order to further determine the edge characteristics of the photovoltaic bracket, it is necessary to screen out all significant edge chains that may be the edges of the photovoltaic bracket based on the gradient direction extension characteristics of the photovoltaic bracket; according to the rigid structure of the edge of the photovoltaic bracket, the edge direction of each edge chain is close to the same and conforms to the linear structure as a whole; therefore, all significant edge chains are determined based on the extension trend of the polarization gradient direction corresponding to the pixel points in the aerial image of the photovoltaic bracket.

[0083] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining a significant edge chain includes:

[0084] The next pixel point in the corresponding polarization gradient direction of each pixel point is used as the direction determination point of each pixel point; each pixel point is used as the target point in turn; the target point is used as the starting search point, and when the angle between the polarization gradient direction of the target point and the polarization gradient direction of the corresponding direction determination point is less than or equal to the preset angle threshold, the corresponding direction determination point is used as the edge chain point of the target point; the edge chain point is used as the new search point, and iteration continues until the iteration stops when the preset iteration stop condition is reached; the target point is connected to all edge chain points obtained during the iteration process in sequence to determine the corresponding significant edge chain. The preset iteration stop condition includes: the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the corresponding direction determination point is greater than the preset angle threshold, or the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the target point is greater than the preset deviation threshold. In a specific implementation of an embodiment of the present invention, the preset angle threshold is set to 5 degrees, and the preset deviation threshold is set to 10 degrees, which can be adjusted according to the specific implementation environment.

[0085] Based on the linear structural characteristics of the photovoltaic support edge, the polarization gradient directions of each pixel on the edge are identical or similar. Combined with the continuity characteristics of the edge, the polarization gradient direction of the target pixel is iterated sequentially, so that the resulting salient edge chain always appears as a continuous linear structure with a nearly consistent polarization gradient direction, which conforms to the characteristics of the photovoltaic support edge. Finally, the target point is changed to obtain all salient edge chains in the saliency image. That is, the above iterative process is performed for each pixel to determine all possible salient edge chains. Based on the obtained salient edge chains, the conformity of the photovoltaic support edge is further analyzed.

[0086] The edges of photovoltaic supports are generally characterized by low curvature and high continuity. The process of determining significant edge chains uses the low curvature feature, i.e., the feature that approaches a straight line. Therefore, each significant edge chain is further analyzed in terms of continuity. Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining reference continuity includes:

[0087] The minimum geometric saliency of all pixels on each salient edge chain is used as the reference weak saliency;

[0088] For a significant edge chain belonging to the edge of a photovoltaic bracket, all pixels on it should have a large geometric significance. In order to prevent the influence of some pixels with large geometric significance that appear accidentally on some noise edges on the continuity calculation process, the minimum value of the geometric significance of all pixels on each significant edge chain is selected as the reference weak significance for the calculation of reference continuity to ensure global consistency; therefore, for each significant edge chain, the smaller the reference weak significance, the higher the possibility that it belongs to the edge of a photovoltaic bracket; since the edge of a photovoltaic bracket usually completely conforms to the straight line feature, its corresponding curvature is usually small and close to 0, so for each significant edge chain, the smaller its average curvature, the more likely it is to belong to the edge of a photovoltaic bracket; in terms of continuity, the more pixels in the significant edge chain, the more continuous the overall chain is, and the more likely it is to belong to the edge of a photovoltaic bracket.

[0089] Therefore, further, the reference continuity of each significant edge chain is determined based on the product of the negative correlation mapping value of the average curvature of each significant edge chain, the number of pixels, and the reference weak saliency; so that the greater the reference continuity, the more likely the corresponding significant edge chain is to belong to the edge of the photovoltaic support. In a specific implementation of an embodiment of the present invention, the method of performing negative correlation mapping on the average curvature of each significant edge chain is: the opposite number of the average curvature of each significant edge chain is used as the power of an exponential function with a natural constant as the base, and the output result of the exponential function is the value after negative correlation mapping. Those skilled in the art may choose to implement it by other basic mathematical methods, which will not be limited or elaborated here.

[0090] Furthermore, based on the fact that photovoltaic brackets are usually installed in parallel with a fixed spacing, adjacent edge chains should satisfy parallelism and equidistantness. Therefore, based on the regularity characteristics of parallelism, the corresponding topological correlation is determined according to the polarization gradient direction deviation, relative distance and reference continuity deviation between each significant edge chain and the corresponding adjacent edge chain, so that the greater the topological correlation, the more the two adjacent significant edge chains conform to the characteristics of the photovoltaic bracket edge.

[0091] Preferably, in some possible implementations of the embodiments of the present invention, the process of acquiring the topological association degree includes:

[0092] The unit vector corresponding to the polarization gradient direction of each pixel point is used as the gradient vector; the gradient vectors of all pixel points on each significant edge chain are vector accumulated to determine the edge direction vector of each significant edge chain; all adjacent edge chains of each significant edge chain are obtained; the Euclidean distance between the center of mass of each significant edge chain and the center of mass of each corresponding adjacent edge chain is less than a preset distance threshold; the angle between the edge direction vector of each significant edge chain and the edge direction vector of each corresponding adjacent edge chain is normalized to determine the corresponding adjacent direction deviation; the Euclidean distance between the center of mass of each significant edge chain and the center of mass of each corresponding adjacent edge chain is normalized to determine the corresponding adjacent reference distance; the difference between the adjacent reference distance and the preset installation spacing is used as the corresponding adjacent distance deviation.

[0093] In a specific implementation method of an embodiment of the present invention, the preset installation spacing needs to be determined based on the size of the fixed spacing corresponding to the installation of the photovoltaic bracket set in the specific implementation environment, mapped to the aerial image of the photovoltaic bracket, and can be set according to the specific implementation environment. The preset distance threshold is set to five times the preset installation spacing and can be adjusted according to the specific implementation environment.

[0094] In the parallelism dimension, the extension direction characteristics of each significant edge chain can be determined by accumulating the gradient vectors of all pixels on the significant edge chain. For two adjacent significant edge chains, the closer their extension directions are to being consistent, that is, the smaller the angle between the corresponding two edge direction vectors, the closer the two significant edge chains are to being parallel. Therefore, the smaller the deviation in the adjacent direction, the more the corresponding significant edge chain and the adjacent edge chain conform to the edge parallelism characteristic, and the higher the probability that the corresponding edge chain belongs to the edge of the photovoltaic support. In the equidistance dimension, the closer the distance between the significant edge chain and the corresponding adjacent edge chain is to the preset installation spacing, the more it conforms to the equidistance characteristic. Therefore, the smaller the deviation in the adjacent distance, the higher the probability that the corresponding edge chain belongs to the edge of the photovoltaic support.

[0095] The corresponding continuity deviation value is determined based on the difference between the reference continuity of each significant edge chain and the reference continuity of each corresponding adjacent edge chain. Similarly, the reference continuity characterizes the rigid structural characteristics of the photovoltaic support edge, so the significant edge chain corresponding to the photovoltaic support edge usually exhibits similar reference continuity values. Therefore, the smaller the deviation of the reference continuity between the significant edge chain and the adjacent edge chain, that is, the smaller the continuity deviation value, the higher the possibility that the corresponding edge belongs to the photovoltaic support edge. Therefore, the topological correlation degree is further determined by combining the adjacent direction deviation degree, the adjacent distance deviation degree and the continuity deviation value, so that the greater the topological correlation degree, the higher the possibility that the corresponding significant edge chain belongs to the photovoltaic support edge relative to each adjacent edge chain.

[0096] In a specific implementation of an embodiment of the present invention, a negative correlation mapping is performed on the product of the neighboring direction deviation, the neighboring distance deviation, and the continuity deviation value to determine the topological association between each significant edge chain and each corresponding neighboring edge chain. The method for performing negative correlation mapping on the product of the neighboring direction deviation, the neighboring distance deviation, and the continuity deviation value is as follows: linearly normalizing the product of the neighboring direction deviation, the neighboring distance deviation, and the continuity deviation value, and then subtracting the normalized value from 1 to obtain the negative correlation mapping result, i.e., the topological association between each significant edge chain and each corresponding neighboring edge chain.

[0097] Step S104: Determine the comprehensive structural confidence of each pixel point based on geometric saliency, reference continuity, and topological correlation; select the edge points of the bracket main beam based on the comprehensive structural confidence; and construct a three-dimensional model of the photovoltaic bracket based on the edge points of the bracket main beam.

[0098] At this point, the geometric saliency of each pixel point in the aerial image of the photovoltaic bracket and the reference continuity and topological correlation corresponding to each salient edge chain, which represent the possibility of belonging to the edge of the photovoltaic bracket, are determined. The above features are further integrated to determine the comprehensive structural confidence that each pixel point belongs to the photovoltaic bracket structure.

[0099] Preferably, in some possible implementations of the embodiments of the present invention, the process of obtaining the comprehensive structure confidence includes:

[0100] The corresponding proximity regularity is determined based on the mean of the topological correlation between the significant edge chain where each pixel is located and all the corresponding adjacent edge chains. For each significant edge chain, the greater the corresponding proximity regularity, the greater the overall topological correlation between it and the corresponding adjacent edge chains, which means that the corresponding significant edge chain is more consistent with the equidistance and parallelism characteristics of the photovoltaic support, and the higher the possibility that it belongs to the edge corresponding to the photovoltaic support structure. The reference continuity characterizes the possibility that each significant edge chain belongs to the edge corresponding to the photovoltaic support structure in terms of high continuity and low curvature. Therefore, the product of the reference continuity and proximity regularity of the significant edge chain where each pixel is located is further normalized to determine the edge chain regularity of each pixel. The larger the edge chain regularity, the more likely the corresponding significant edge chain is to belong to the edge corresponding to the photovoltaic support structure, that is, the more likely the corresponding pixel is to belong to the pixel on the photovoltaic support structure. In a specific implementation of an embodiment of the present invention, the method for normalizing the product of the reference continuity and the proximity regularity of the significant edge chain where each pixel point is located adopts linear normalization. The normalization method can be adjusted according to the specific implementation environment and will not be further limited or elaborated here.

[0101] On the basis of the edge chain regularity, it is also necessary to consider that the greater the geometric significance of the pixel point, the more prominent the edge information of the corresponding pixel point and the higher the possibility that the corresponding edge texture belongs to the edge of the photovoltaic support. Therefore, the comprehensive structural confidence of each pixel point is further determined based on the product between the edge chain regularity and the geometric significance of each pixel point, so that the greater the comprehensive structural confidence, the higher the possibility that the corresponding pixel point belongs to the edge of the photovoltaic support, that is, the pixel point on the photovoltaic support structure. Further, based on the comprehensive structural confidence, the edge points of the support main beam that meet the photovoltaic support structure are screened out. In a specific implementation method of an embodiment of the present invention, the pixel points whose corresponding comprehensive structural confidence is greater than a preset confidence threshold are used as the edge points of the support main beam; the preset confidence threshold is set to 0.7, which can be adjusted according to the specific implementation environment.

[0102] After determining the edge points of the support main beam, the three-dimensional model of the photovoltaic support is further constructed based on the edge points of the support main beam. In a specific implementation of an embodiment of the present invention, all the edge points of the support main beam whose Euclidean distance is less than a preset connection threshold are connected to obtain all the edge chains of the support main beam, wherein the preset connection threshold is set to 4 pixels and can be adjusted automatically. By connecting the edge points of the support main beam with a short distance, it is possible to avoid accidental edge breaks, so that the obtained support main beam edge chain is more complete. The obtained support main beam edge chain is further used as a strong constraint to guide the grid generation algorithm to fit the support structure to construct a three-dimensional model of the photovoltaic support. In a specific implementation of an embodiment of the present invention, the grid generation algorithm adopts Poisson reconstruction. It should be noted that Poisson reconstruction is a technical means well known to those skilled in the art and will not be further defined or elaborated here.

[0103] In summary, a method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images takes into account the influence of reflection noise and the characteristic that metal reflection will generate pseudo edges with high gradient values ​​in non-edge areas. First, the reflection suppression factor is determined to separate the reflection noise source and retain the diffuse reflection base texture; then, the geometric significance of each pixel point is comprehensively determined based on the edge direction consistency determined by the rigidity and periodicity characteristics of the photovoltaic bracket to highlight the structural characteristics of the photovoltaic bracket; further, the arrangement of the photovoltaic bracket will make the edge chain present parallelism and continuity, and the reference continuity of the significant edge chain and the topological correlation between adjacent significant edge chains in parallel are analyzed; finally, based on the bracket structural characteristics represented by the geometric significance of each pixel point and the photovoltaic bracket arrangement edge characteristics represented by the reference continuity and topological correlation of the edge chain, the edge points of the bracket main beam are screened out; thereby, a more accurate three-dimensional model of the photovoltaic bracket is constructed based on the edge points of the bracket main beam, so that the obtained three-dimensional model of the photovoltaic bracket strictly conforms to the rigidity, parallelism and periodicity characteristics of the bracket.

[0104] This application also provides a photovoltaic bracket three-dimensional model construction system based on aerial images, please refer to Figure 2 , which shows a structural diagram of a photovoltaic bracket three-dimensional model construction system based on aerial images provided by an embodiment of the present invention. The system includes: an image acquisition and preprocessing module 201, a first determination module 202, a second determination module 203 and a photovoltaic bracket three-dimensional model construction module 204.

[0105] The image acquisition and preprocessing module 201 is used to obtain an aerial image of the photovoltaic support; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic support; and divide the aerial image of the photovoltaic support into at least two local image blocks;

[0106] The first determination module 202 is configured to determine the geometric saliency of each pixel in each local image block based on the polarization gradient modulus of each pixel, the overall deviation of the polarization gradient modulus, and the overall similarity of the normal vector direction corresponding to each pixel in the photovoltaic support point cloud;

[0107] The second determination module 203 is configured to determine significant edge chains based on the extension trend of the polarization gradient direction of the pixel points; determine the corresponding reference continuity based on the length, curvature and geometric significance distribution of the significant edge chains; and determine the corresponding topological correlation based on the overall deviation between each significant edge chain and its adjacent edge chains.

[0108] The photovoltaic bracket three-dimensional model construction module 204 is used to determine the comprehensive structural confidence of each pixel point based on geometric significance, reference continuity and topological correlation; filter out the edge points of the bracket main beam according to the comprehensive structural confidence; and construct the photovoltaic bracket three-dimensional model based on the edge points of the bracket main beam.

[0109] It should be noted that the system provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the above embodiment provides a photovoltaic bracket three-dimensional model construction system based on aerial images and a photovoltaic bracket three-dimensional model construction method based on aerial images. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0110] The present application also provides a computer device. Figure 3 , which shows a schematic diagram of the structure of a computer device provided by an embodiment of the present invention, the computer device includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and running on the processor 302, wherein when the processor 302 executes the computer program 303, the computer device can execute any of the above-mentioned methods for constructing a three-dimensional model of a photovoltaic bracket based on aerial images.

[0111] An embodiment of the present application also provides a computer program product. When the computer program product is run on a computer device, the computer device can execute any of the above-mentioned methods for constructing a three-dimensional model of a photovoltaic bracket based on aerial images.

[0112] An embodiment of the present application also provides a computer-readable storage medium, which stores computer program code. When the computer program code runs on a computer device, the computer device can execute any of the above-mentioned methods for constructing a three-dimensional model of a photovoltaic bracket based on aerial images.

[0113] In the embodiments provided in the present application, it should be understood that the provided computer devices, computer program products and computer-readable storage media are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the methods provided above and will not be repeated here.

[0114] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0115] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images, characterized in that: The method comprises: Acquire an aerial image of a photovoltaic bracket; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic bracket; divide the aerial image of the photovoltaic bracket into at least two local image blocks; In each local image block, the geometric saliency of each pixel is determined based on the polarization gradient modulus of each pixel, the overall deviation of the polarization gradient modulus, and the overall similarity of the normal vector direction corresponding to each pixel in the photovoltaic support point cloud; Based on the extension trend of the polarization gradient direction of the pixel points, significant edge chains are determined; the corresponding reference continuity is determined based on the length, curvature and geometric significance distribution of the significant edge chains; and the corresponding topological correlation is determined based on the overall deviation between each significant edge chain and its adjacent edge chains. According to the geometric significance, the reference continuity and the topological correlation, the comprehensive structural confidence of each pixel point is determined; the edge points of the bracket main beam are screened out according to the comprehensive structural confidence; and the three-dimensional model of the photovoltaic bracket is constructed according to the edge points of the bracket main beam.

2. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 1, characterized in that: The process of determining the polarization gradient modulus and polarization gradient direction of each pixel point in the photovoltaic bracket aerial image includes: The polarization image of the spectral bracket is collected by an aerial drone; the two polarization components of the aerial image of the photovoltaic bracket are obtained by using the Stokes vector based on the polarization images of all photovoltaic brackets; the polarization gradient modulus of each pixel is determined based on the maximum value of the gradient modulus corresponding to the two polarization components. The polarization component corresponding to the polarization gradient modulus corresponding to each pixel point is used as the reference polarization component; the gradient direction of each pixel point under the corresponding reference polarization component is used as the polarization gradient direction of each pixel point.

3. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 1, characterized in that: The process of obtaining the geometric saliency includes: In each local image block, a corresponding reflection suppression factor is determined according to the deviation of the polarization gradient modulus of each pixel compared to the overall polarization gradient modulus of the local image block; A dense point cloud is generated using photogrammetry software based on multi-viewpoint aerial images of the photovoltaic bracket to obtain the photovoltaic bracket point cloud; the corresponding edge direction consistency is determined based on the similarity between the normal vector direction corresponding to each pixel point in the photovoltaic bracket point cloud and the normal vector direction of the entire pixel point in the local image block; The geometric saliency of each pixel is determined according to the product of the reflection suppression factor, the edge direction consistency and the polarization gradient modulus of each pixel.

4. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 3, characterized in that: The process of obtaining the edge direction consistency includes: Obtain the point cloud data point corresponding to the location of each pixel point in the photovoltaic support point cloud; form a corresponding neighboring point cloud set based on all point cloud data points within a preset neighborhood range of the point cloud data point corresponding to each pixel point; perform SVD decomposition on the covariance matrix constructed by the neighboring point cloud set, and use the normal vector corresponding to the minimum eigenvalue obtained after the decomposition as the reference normal vector for each pixel point; The direction corresponding to the vector obtained by vector accumulation operation of the reference normal vectors of all pixels in each local image block is used as the corresponding main direction of the bracket; the angle between the main direction of the bracket and the polarization gradient direction of each pixel point is negatively correlated and mapped to determine the edge direction consistency of each pixel point.

5. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 3, characterized in that: The process of obtaining the reflex inhibition factor includes: A polarization gradient modulus distribution coordinate system is constructed, with the direction in which the polarization gradient modulus lengths are arranged from small to large as the horizontal axis and the direction in which the number of pixels is arranged from small to large as the vertical axis. A polarization gradient modulus distribution curve is constructed based on the polarization gradient modulus lengths of all pixels in each local image block in combination with the polarization gradient modulus distribution coordinate system. Half of the mean between the maximum polarization gradient modulus length and the minimum polarization gradient modulus length in the polarization gradient modulus distribution curve is used as a peak screening value. Among all extreme points on the polarization gradient modulus distribution curve whose polarization gradient modulus is greater than the peak screening value, the maximum point with the largest value in the polarization gradient modulus distribution curve is taken as the main peak point; the polarization gradient modulus corresponding to the minimum point closest to the main peak point is taken as the first main peak boundary value; the polarization gradient modulus corresponding to the minimum point closest to the main peak point is taken as the second main peak boundary value; and all pixel points whose corresponding polarization gradient modulus is greater than the first main peak boundary value and less than the second main peak boundary value are taken as main peak pixel points; In each local image block, the sum of the positive correlation mapping value of the standard deviation of the polarization gradient modulus corresponding to all main peak pixels and the mean of the polarization gradient modulus of all main peak pixels is used as the gradient failure threshold of each local image block; The difference between the polarization gradient modulus of each pixel in each local image block and the gradient failure threshold is negatively correlated to determine the reflection suppression factor of each pixel.

6. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 1, characterized in that: The process of obtaining the significant edge chain includes: The next pixel point of each pixel point in the corresponding polarization gradient direction is used as the direction determination point of each pixel point; Take each pixel as the target point in turn; The target point is used as the starting search point. When the angle between the polarization gradient direction of the target point and the polarization gradient direction of the corresponding direction determination point is less than or equal to a preset angle threshold, the corresponding direction determination point is used as the edge chain point of the target point. The edge chain point is used as a new search point, and iteration is continued until a preset iteration stop condition is reached. The target point is sequentially connected with all edge chain points obtained during the iteration process to determine the corresponding significant edge chain. The preset iteration stop condition includes: the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the corresponding direction determination point is greater than a preset angle threshold, or the angle between the polarization gradient direction of the new search point and the polarization gradient direction of the target point is greater than a preset deviation threshold; Changing the target point obtains all salient edge chains in the saliency image.

7. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 4, characterized in that: The process of obtaining the reference continuity includes: The minimum geometric saliency of all pixels on each salient edge chain is used as the reference weak saliency; The reference continuity of each salient edge chain is determined according to the product of the negative correlation mapping value of the average curvature of each salient edge chain, the number of pixels and the reference weak saliency.

8. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 1, characterized in that: The process of obtaining the topological association degree includes: The unit vector corresponding to the polarization gradient direction of each pixel is used as the gradient vector; the gradient vectors of all pixels on each significant edge chain are summed up to determine the edge direction vector of each significant edge chain; Obtain all adjacent edge chains for each significant edge chain; the Euclidean distance between the centroid of each significant edge chain and the centroid of each corresponding adjacent edge chain is less than a preset distance threshold; Normalize the angle between the edge direction vector of each significant edge chain and the edge direction vector of each corresponding adjacent edge chain to determine the corresponding adjacent direction deviation; Normalizing the Euclidean distance between the centroid of each significant edge chain and the centroid of each corresponding adjacent edge chain to determine a corresponding adjacent reference distance; and using the difference between the adjacent reference distance and a preset installation spacing as the corresponding adjacent distance deviation; Determining a corresponding continuity deviation value according to the difference between the reference continuity of each significant edge chain and the reference continuity of each corresponding adjacent edge chain; A negative correlation mapping is performed on the product of the adjacent direction deviation, the adjacent distance deviation and the continuity deviation value to determine the topological association between each significant edge chain and each corresponding adjacent edge chain.

9. The method for constructing a three-dimensional model of a photovoltaic bracket based on aerial images according to claim 1, characterized in that: The process of obtaining the comprehensive structure confidence includes: Determine the corresponding proximity regularity based on the mean of the topological correlation between the significant edge chain where each pixel is located and all corresponding adjacent edge chains; Normalizing the product of the reference continuity of the significant edge chain where each pixel point is located and the proximity regularity to determine the edge chain regularity of each pixel point; The comprehensive structural confidence of each pixel is determined according to the product of the edge chain regularity and the geometric significance of each pixel.

10. A photovoltaic bracket three-dimensional model construction system based on aerial images, characterized in that: The system comprises: An image acquisition and preprocessing module is used to obtain an aerial image of the photovoltaic bracket; determine the polarization gradient modulus and polarization gradient direction of each pixel in the aerial image of the photovoltaic bracket; and divide the aerial image of the photovoltaic bracket into at least two local image blocks; The first determination module is used to determine the geometric saliency of each pixel in each local image block based on the polarization gradient modulus of each pixel, the overall deviation of the polarization gradient modulus, and the overall similarity of the normal vector direction corresponding to each pixel in the photovoltaic support point cloud; The second determination module is configured to determine significant edge chains based on the extension trend of the polarization gradient direction of the pixel points; determine the corresponding reference continuity based on the length, curvature and geometric significance distribution of the significant edge chains; and determine the corresponding topological correlation based on the overall deviation between each significant edge chain and its adjacent edge chains; The photovoltaic bracket three-dimensional model construction module is used to determine the comprehensive structural confidence of each pixel point based on the geometric significance, the reference continuity and the topological correlation; screen out the edge points of the bracket main beam according to the comprehensive structural confidence; and construct the photovoltaic bracket three-dimensional model based on the bracket main beam edge points.

Images