Method for calculating engineering quantity through aerial photography three-dimensional modeling of unmanned aerial vehicle

Abstract

The invention discloses an unmanned aerial vehicle aerial photography three-dimensional modeling engineering quantity calculation method, which comprises the steps of performing three-dimensional modeling on a to-be-modeled area according to multiple frames of images of the to-be-modeled area acquired by an unmanned aerial vehicle and flight data of the unmanned aerial vehicle, and dividing three-dimensional point cloud data obtained by modeling into multiple categories, determining classification evaluation values of the categories according to the minimum principal component direction projection values of all the point clouds in the categories and the dispersion degree values of the categories, and determining re-classification coefficients between the adjacent categories according to the classification evaluation values and the similarity between the adjacent categories, thereby screening the categories to be re-classified for re-classification. And obtaining the edge of the three-dimensional model according to the reclassification result, thereby determining the size data of the three-dimensional model, and carrying out engineering quantity calculation. According to the invention, the edge of the obtained three-dimensional model is accurate, and the accuracy and robustness of engineering quantity calculation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional modeling, and more specifically, to a method for calculating engineering quantities by using unmanned aerial vehicle (UAV) aerial three-dimensional modeling. Background Art

[0002] During construction, it is necessary to calculate the engineering quantity of the currently constructed projects in order to control the construction progress.

[0003] The image data of the constructed project is collected by drones for 3D modeling. The volume of the constructed project is estimated based on the 3D model. The volume estimation result can be used to calculate the engineering quantity. Due to the large interference factors when the drone collects images, the volume estimation result of the constructed project may be inaccurate. The edge is an important basis for volume estimation. In order to improve the accuracy of volume estimation, the edge of the 3D model needs to be accurately extracted when performing volume estimation.

[0004] The existing edge extraction method is: calculate the normal plane of the three-dimensional model, project the point cloud data of the three-dimensional model onto the normal plane, obtain the projected two-dimensional image, use the image edge detection method to realize edge detection in the two-dimensional image, obtain the edge of the two-dimensional image, map the edge of the two-dimensional image back to the three-dimensional model, and obtain the edge of the three-dimensional model.

[0005] When calculating the normal plane of a 3D model, it is necessary to calculate the normal vector of each point cloud data and cluster the normal vectors of all point cloud data to obtain the corresponding normal plane. However, due to the unstable sparseness and density of point cloud data during the acquisition process, the clustering effect of the normal vector may be poor, which will affect the accuracy of the normal plane, resulting in inaccurate edges of the acquired 3D model, and further inaccurate volume estimation results and engineering quantity calculation results. Summary of the invention

[0006] In order to solve the technical problem that the normal vector clustering effect is poor, the accuracy of the normal plane is affected, and the engineering quantity calculation result is inaccurate, the present invention provides a method for calculating the engineering quantity by using drone aerial photography three-dimensional modeling, comprising:

[0007] Performing three-dimensional modeling of the area to be modeled based on multiple frames of images of the area to be modeled collected by the drone and the flight data of the drone to obtain a three-dimensional model of the area to be modeled, wherein the three-dimensional model is three-dimensional point cloud data; dividing the point cloud into multiple categories based on the normal vector and three-dimensional coordinates of each point cloud in the three-dimensional point cloud data;

[0008] According to the variance of the minimum principal component directional projection values ​​of all point clouds in each category and the dispersion value of each category, the classification evaluation value of each category is determined to evaluate the classification effect of each category; the dispersion value is negatively correlated with the minimum principal component directional projection value of the point cloud in the category, and positively correlated with the Euclidean distance between the point cloud and the nearest boundary point;

[0009] Taking any one category as the target category, determining a reclassification coefficient between the target category and each adjacent category of the target category according to the classification evaluation value of the target category, the classification evaluation value of each adjacent category of the target category, and the similarity between the target category and each adjacent category, wherein the reclassification coefficient is positively correlated with the similarity;

[0010] According to the size of the reclassification coefficient, the categories to be reclassified are screened for reclassification; according to the result of the reclassification, the edge of the three-dimensional model is obtained, the dimension data of the three-dimensional model is obtained according to the edge, and the engineering quantity is calculated according to the dimension data.

[0011] Preferably, the point cloud is divided into multiple categories according to the normal vector and three-dimensional coordinates of each point cloud in the three-dimensional point cloud data, including:

[0012] The three-dimensional coordinates and normal vectors of the point cloud constitute the feature values ​​of the point cloud, the feature values ​​of all the point clouds in the three-dimensional point cloud data are used as the first data samples, and the first data samples are clustered to obtain multiple categories.

[0013] Preferably, the method for obtaining the minimum principal component direction projection value is:

[0014] Perform principal component analysis on the eigenvalues ​​of all point clouds in each category to obtain the minimum principal component direction of each category; obtain the projection value of the eigenvalue of each point cloud in each category on the minimum principal component direction of the category as the minimum principal component direction projection value of each point cloud.

[0015] Preferably, the dispersion degree value satisfies the expression:

[0016] ;

[0017] in, Indicates The dispersion value of each category; Indicates of the categories The minimum principal component direction projection value of the point cloud; Indicates The mean of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates of the categories Point cloud and The distance between the categories The Euclidean distance between the nearest boundary points of a point cloud; Indicates The number of point clouds in each category; is a S-type function: .

[0018] Preferably, the classification evaluation value satisfies the expression:

[0019] ;

[0020] In the formula, Indicates The classification evaluation value of each category; Indicates The variance of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates The dispersion value of each category; is a hyperparameter; is an exponential function with a natural constant as its base.

[0021] Preferably, the method for obtaining the nearest boundary point is:

[0022] For each point cloud in each category, the Euclidean distance between the point cloud and all boundary points in the category is calculated, and the boundary point with the smallest Euclidean distance to the point cloud is taken as the boundary point closest to the point cloud.

[0023] Preferably, the reclassification coefficient satisfies the expression:

[0024] ;

[0025] In the formula, Indicates the target category and the The reclassification coefficient between adjacent categories; Represents the classification evaluation value of the target category; The target category The classification evaluation values ​​of adjacent categories; The eigenvector representing the minimum principal component direction of the target category is The cosine similarity between the feature vectors in the direction of the minimum principal component of adjacent categories, represents the absolute value, Indicates the target category and the The similarity between adjacent categories; represents the minimum function; is an exponential function with a natural constant as its base.

[0026] Preferably, the screening of the categories to be reclassified according to the size of the reclassification coefficient includes:

[0027] In response to a reclassification coefficient between a target category and one of its adjacent categories being greater than a preset classification threshold, the target category and the adjacent category are respectively regarded as a category to be reclassified; and all the categories to be reclassified are obtained by judging the reclassification coefficient between each category and each of its adjacent categories.

[0028] Preferably, the reclassification comprises:

[0029] All the categories to be reclassified are divided into multiple groups, each category in each group has at least one adjacent category in the group, and each category in each group is not adjacent to all categories in other groups; for each group, the characteristic values ​​of all point clouds in all the categories to be reclassified in the group are used as the second data samples of the group, the second data samples are clustered into multiple categories, and the categories to be reclassified are re-obtained according to the reclassification coefficients between each category and its adjacent categories obtained by clustering, and reclassification is performed again until there are no categories to be reclassified or the number of reclassifications reaches a preset threshold.

[0030] Preferably, obtaining the edge of the three-dimensional model according to the reclassification result includes:

[0031] Calculate the normal plane corresponding to each category, project the point cloud data of each category onto the normal plane of the category to obtain the projected two-dimensional image, use the image edge detection method to implement edge detection in the two-dimensional image to obtain the edge of the two-dimensional image, map the edge of the two-dimensional image back to the three-dimensional model to obtain the edge of the three-dimensional model.

[0032] The beneficial effects of the present invention are:

[0033] The present invention evaluates the classification effect of each category according to the minimum principal component direction projection value of all point clouds in each category obtained by dividing the three-dimensional point cloud data and the dispersion value of the category. The use of the minimum principal component direction can improve the robustness of the evaluation result; the present invention determines the reclassification coefficient between adjacent categories according to the classification evaluation value of adjacent categories and the similarity between adjacent categories, can realize the screening of categories that need to be reclassified in low classification effects, reclassify the categories that need to be reclassified, improve the accuracy of the classification results, thereby improving the accuracy of the edge of the three-dimensional model, can obtain accurate dimensional data, and improve the robustness and accuracy of engineering quantity calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The above and other purposes, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below through the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0035] Figure 1 It is a flow chart schematically showing a method for calculating engineering quantity by three-dimensional modeling of drone aerial photography in the present invention;

[0036] Figure 2 It is a flowchart schematically showing step S1 of calculating engineering quantity of three-dimensional modeling of drone aerial photography according to the present invention;

[0037] Figure 3 It is a flowchart schematically showing step S2 of calculating engineering quantity of three-dimensional modeling of aerial photography by unmanned aerial vehicle according to the present invention;

[0038] Figure 4 The present invention schematically shows a schematic diagram of obtaining boundary points in calculating engineering quantities by three-dimensional modeling of aerial photography by unmanned aerial vehicles. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0040] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0041] The embodiment of the present invention discloses a method for calculating engineering quantities by using a drone to perform three-dimensional modeling. Figure 1 , including steps S1 to S3:

[0042] S1. Perform three-dimensional modeling of the area to be modeled based on multiple frames of images of the area to be modeled collected by the drone and the flight data of the drone.

[0043] See the flowchart of step S1. Figure 2 , including steps S101 to S103, specifically:

[0044] S101, using a drone to collect multiple frames of images of the area to be modeled and flight data of the drone.

[0045] The UAV is equipped with multiple cameras, a GPS positioning system, and a posture sensor. When the UAV flies according to the set flight path, the multiple cameras continuously collect multiple frames of images of the area to be modeled, and the GPS positioning system and posture sensor are used to obtain the real-time flight data of the UAV, including flight altitude and flight attitude.

[0046] S102, preprocessing the image of the area to be modeled.

[0047] Each frame image of the area to be modeled is preprocessed, and the preprocessing includes but is not limited to denoising and lighting correction of the image. It should be noted that the embodiment of the present invention does not limit the denoising algorithm and lighting correction algorithm used, and the implementer can set them according to the actual implementation situation. For example, the denoising algorithm uses mean filtering, median filtering, Gaussian filtering, etc., and the lighting correction algorithm uses histogram equalization, limited contrast adaptive histogram equalization, etc.

[0048] S103, performing three-dimensional modeling according to the pre-processed image of the area to be modeled and the flight data of the UAV to obtain a three-dimensional model of the area to be modeled.

[0049] The three-dimensional modeling of the area to be modeled is performed based on the preprocessed multi-frame images of the area to be modeled and the flight data of the drone to obtain a three-dimensional model of the area to be modeled. It should be noted that the present invention does not limit the method of three-dimensional modeling, and the implementer can set it according to the actual implementation situation, such as performing three-dimensional modeling by calculating the projection matrix, performing three-dimensional modeling by binocular depth estimation, etc. Performing three-dimensional modeling by calculating the projection matrix and performing three-dimensional modeling by binocular depth estimation are prior arts and will not be described in detail here.

[0050] The three-dimensional model of the area to be modeled is three-dimensional point cloud data, and each point cloud contains the horizontal and vertical coordinates and elevation data of the point cloud. The normal vector of each point cloud in the three-dimensional point cloud data is obtained by using a method based on local surface fitting, and the normal vector of each point cloud is three-dimensional data. It should be noted that the method based on local surface fitting to obtain the normal vector is a prior art and will not be repeated here.

[0051] S2. Divide the three-dimensional point cloud data into multiple categories, determine the classification evaluation value of each category, which is used to evaluate the classification effect of each category, determine the reclassification coefficient between each category and each adjacent category according to the classification evaluation value, and select the categories to be reclassified according to the size of the reclassification coefficient for reclassification to obtain the classification result of the three-dimensional point cloud data.

[0052] See the flowchart of step S2. Figure 3 , including steps S201 to S204, specifically:

[0053] S201, dividing the three-dimensional point cloud data into multiple categories.

[0054] Specifically, for each point cloud in the three-dimensional point cloud data, the three-dimensional coordinates of the point cloud and the three-dimensional data in the normal vector are spliced ​​into a six-dimensional data as the eigenvalue of the point cloud. The eigenvalues ​​of all point clouds in the three-dimensional point cloud data are used as the first data sample, and the first data sample is subjected to mean shift clustering to obtain multiple categories. It should be noted that the clustering radius of mean shift clustering in the embodiment of the present invention is 0.1. In other embodiments, the implementer can set the clustering radius of mean shift clustering according to the actual implementation situation. The embodiment of the present invention only takes mean shift clustering as an example for illustration. In other embodiments, the implementer can set the classification method for data categories according to the actual implementation situation, such as the density-based spatial clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN).

[0055] S202 : Determine a classification evaluation value for each category according to the variance of the minimum principal component directional projection values ​​of all point clouds in each category and the dispersion value of each category.

[0056] It should be noted that, since the normal vector of the surface model changes relatively uniformly, if the normal vector is used as the classification basis for finding the normal plane, the surface with the same curvature will be divided into multiple normal planes, that is, the point cloud originally belonging to the same normal plane will be mistakenly divided into multiple categories. In addition, during the clustering process, it is difficult to balance the weights of the normal vector and the three-dimensional coordinate data of the point cloud, which may cause boundary segmentation errors, and then cause the point clouds belonging to multiple normal planes to be mistakenly divided into the same category. Therefore, some categories need to be reclassified. The present invention performs principal component analysis (PCA) on each category, and evaluates the classification effect of each category based on the results of the principal component analysis, so as to subsequently screen the categories that need to be reclassified.

[0057] Specifically, principal component analysis is performed on the eigenvalues ​​of all point clouds in each category to obtain the minimum principal component direction of each category and the projection value of the eigenvalue of each point cloud in each category on the minimum principal component direction of the category as the minimum principal component direction projection value of each point cloud.

[0058] It should be noted that after the eigenvalues ​​of all point clouds in each category are reduced in dimension by PCA, the variance of the projection value in the direction of the minimum component is the smallest relative to the variance of the projection values ​​in other directions. If the point cloud data in the category is relatively smooth or the normal vector is relatively consistent, the variance of the projection value in the direction of the minimum principal component should be close to 0, and is not affected by the size of the normal vector and the size of the normal plane. Since the minimum principal component direction is calculated based on the data distribution, it has strong robustness. Therefore, the present invention evaluates the classification effect of each category according to the variance of the projection values ​​of the minimum principal component direction of all point clouds in each category, and obtains the classification evaluation value of each category.

[0059] Specifically, the classification evaluation value of each category satisfies the expression:

[0060] ;

[0061] In the formula, Indicates The classification evaluation value of each category; Indicates The variance of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates The dispersion value of each category; is a hyperparameter used to adjust the variance of the minimum principal component direction projection value of all point clouds in the category and dispersion value The classification weight between , the empirical value is 0.7; is an exponential function with a natural constant as the base, used for as well as Negative correlation mapping.

[0062] When The variance of the minimum principal component directional projection values ​​of all point clouds in the categories The smaller the time, the The more concentrated the distribution of the minimum principal component directional projection values ​​of the point cloud in the first category, The better the clustering effect of each category, the The larger the classification evaluation value of a category, the greater the The larger the value, the The more dispersed the minimum principal component directional projection values ​​of the point clouds in the first category are, the more likely it is that the normal vectors or point clouds with large differences will be clustered into one category, resulting in The worse the clustering effect of each category, the The smaller the classification evaluation value of each class is, the smaller the classification evaluation value of each class is.

[0063] because It is calculated by The variance of the projection values ​​of the eigenvalues ​​of all point clouds in the direction of the smallest principal component is obtained. When there are multiple poor clustering effects in a category, it will not be possible to Therefore, the embodiment of the present invention combines the The dispersion value of each category is used to obtain the The classification evaluation value of each category.

[0064] Specifically, the dispersion degree value satisfies the expression:

[0065] ;

[0066] in, Indicates The dispersion value of each category; Indicates of the categories The minimum principal component direction projection value of the point cloud; Indicates The mean of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates of the categories Point cloud and The distance between the categories The Euclidean distance between the nearest boundary points of a point cloud; Indicates The number of point clouds in each category; is a S-type function: When the dispersion value The bigger, the The projection value of the point cloud of the first category in the direction of the smallest principal component has a large local deviation. The worse the clustering effect of each category, the The smaller the classification evaluation value of each category, the smaller the dispersion value. The smaller the size, the better the clustering effect. The larger the classification evaluation value of a category.

[0067] The method for obtaining the boundary points is as follows: For each point cloud in a category, make a tangent plane perpendicular to the normal vector based on the normal vector of the point cloud, project the point cloud in the neighborhood of the point cloud onto the tangent plane, and connect each projected point cloud on the tangent plane with the point cloud to form a series of vectors. Sort all vectors counterclockwise, and form an angle between any two adjacent vectors. Get the maximum angle, and when the maximum angle is greater than the preset angle threshold, use the point cloud as a boundary point. It should be noted that the angle threshold is set by the implementer based on the actual implementation situation and is not specifically limited. For example, . Figure 4 Get a schematic diagram for the boundary points, Figure 4 Point Cloud , , , , , Point Cloud Point cloud within the neighborhood, Figure 4 1, 2, 3, 4, 5, and 6 represent the angles, and angle 5 is the largest angle. , so the point cloud is a boundary point.

[0068] The distance The method for obtaining the nearest boundary point of a point cloud is: of the categories Point cloud and The Euclidean distance between all boundary points in the first category will be The boundary point with the smallest Euclidean distance of the point cloud is taken as the distance The nearest boundary point of the point cloud.

[0069] S203. Determine a reclassification coefficient between the target category and each adjacent category of the target category according to the classification evaluation value of the target category, the classification evaluation value of each adjacent category of the target category, and the similarity between the target category and each adjacent category.

[0070] It should be noted that when the classification effect of the point cloud data within the category is not good, the classification evaluation value of the category is smaller, and when the classification effect of the point cloud data within the category is better, the classification evaluation value of the category is larger, but when a normal plane is divided into multiple categories, the classification evaluation value of the category is also larger. For categories with smaller classification evaluation values ​​and categories with larger classification evaluation values ​​due to a normal plane being divided into multiple categories, reclassification is required. Therefore, the present invention combines the classification evaluation values ​​of adjacent categories to determine whether each category needs to be reclassified.

[0071] Specifically, for each category, all boundary points of the category are obtained, any boundary point in the category is used as the target boundary point, the Euclidean distance between the target boundary point and all boundary points of each other category is calculated, and the category corresponding to the boundary point with the smallest Euclidean distance to the target boundary point is used as the adjacent category of the category. Similarly, all adjacent categories of the category are obtained by calculating the Euclidean distance between all boundary points in the category and all boundaries of each other category.

[0072] It should be noted that if the classification evaluation values ​​of two adjacent categories are both large and the similarity between the two categories is low, the classification effect of the two categories is good and no reclassification is required; if the classification evaluation values ​​of two adjacent categories are both large and the similarity between the two categories is high, it means that the point cloud belonging to the same normal plane is divided into multiple categories, and the two categories need to be reclassified at this time; if the classification evaluation values ​​of two adjacent categories are low, it means that the classification effect is poor, and at this time, regardless of whether the two categories are similar, they should be reclassified. Since the projection values ​​of all point clouds of each category in the direction of the minimum principal component can constitute a feature vector, the present invention uses the absolute value of the cosine similarity between the feature vectors of two categories to represent the similarity of the two categories, and combines the classification evaluation values ​​of the two adjacent categories and the similarity of the two categories to obtain the reclassification coefficients of the two categories, which are used to evaluate whether the two categories need to be reclassified.

[0073] Specifically, take any category as the target category and obtain the reclassification coefficient between the target category and each adjacent category of the target category:

[0074] ;

[0075] In the formula, Indicates the target category and the The reclassification coefficient between adjacent categories; Represents the classification evaluation value of the target category; The target category The classification evaluation values ​​of adjacent categories; The eigenvector representing the minimum principal component direction of the target category is The cosine similarity between the feature vectors in the direction of the minimum principal component of adjacent categories, Represents the absolute value. Since the value range of cosine similarity is [-1, 1], when the cosine similarity is negative, it indicates a negative correlation between the two eigenvectors. Since the eigenvector represents a direction, regardless of positive or negative, the embodiment of the present invention takes the absolute value of the cosine similarity between the two eigenvectors to represent the similarity between the two categories. represents the minimum function; is an exponential function with a natural constant as its base.

[0076] In the formula, Represents the classification evaluation value in the target category With The classification evaluation value of adjacent categories The minimum value is selected between the two. The larger the minimum value is, the more concentrated the point cloud in the target category is. The point clouds in the adjacent categories are more concentrated. If the target category is The greater the similarity between the target category and the adjacent category, the The absolute value of the cosine similarity between the eigenvectors of the smallest principal component direction of adjacent categories The larger the value, the more likely the target category is to be The adjacent categories may be on the same normal plane. In this case, it is necessary to compare the target category with the first The adjacent categories are reclassified and merged. If the target category is The smaller the similarity between adjacent categories, the closer the target category is to the first It is impossible for two adjacent categories to be on the same normal plane. In this case, there is no need to compare the target category with the first The smaller the minimum value is, the closer the target category is to the adjacent category. The worse the classification effect of a category in the adjacent categories, the worse the classification effect of the target category is. The similarity of adjacent categories needs to be reclassified.

[0077] S204, selecting the categories to be reclassified according to the size of the reclassification coefficient and reclassifying them.

[0078] In response to the reclassification coefficient between the target category and one of its adjacent categories being greater than a preset classification threshold, the target category and the adjacent category are each regarded as a category to be reclassified. By judging the reclassification coefficient between each category and each of its adjacent categories, all the categories to be reclassified are obtained. The classification threshold is set by the implementer according to the actual implementation situation and is not specifically limited, for example, 0.2.

[0079] All the categories to be reclassified are divided into multiple groups, each category in each group has at least one adjacent category in the group, and each category in each group is not adjacent to any category in other groups. For each group, the feature values ​​of all point clouds in all the categories to be reclassified in the group are used as the second data samples of the group, and the second data samples are reclassified using the methods in steps S201 to S204, and the process is repeated until there are no categories to be reclassified or the number of repetitions reaches a preset threshold. All the categories finally obtained constitute the classification results of the three-dimensional point cloud data.

[0080] The quantity threshold is set by the implementers based on the actual implementation situation, for example, 10 times.

[0081] S3. Obtain the edge of the three-dimensional model according to the classification result of the three-dimensional point cloud data, obtain the dimension data of the three-dimensional model according to the edge, and perform engineering quantity calculation according to the dimension data.

[0082] In one embodiment, the normal plane corresponding to each category is calculated based on the classification results of the three-dimensional point cloud data, the point cloud data of each category is projected onto the normal plane of the category to obtain a projected two-dimensional image, and an image edge detection method is used to implement edge detection in the two-dimensional image to obtain the edge of the two-dimensional image. The edge of the two-dimensional image is mapped back to the three-dimensional model to obtain the edge of the three-dimensional model.

[0083] It should be noted that the calculation of normal plane is an existing technology and will not be described in detail here. The present invention does not limit the image edge detection method, and the implementer can set it according to the actual implementation situation, such as Sobel operator, Canny edge detection algorithm, etc.

[0084] In another embodiment, the boundary points of each category in the classification result of the three-dimensional point cloud data are used as edges as the edges of the three-dimensional model.

[0085] The dimensional data of the 3D model is obtained according to the edge of the 3D model, the surface area and volume are calculated according to the dimensional data, and the engineering quantity is calculated according to the calculation results.

[0086] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0087] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for calculating engineering quantities by using drone aerial photography three-dimensional modeling, characterized in that: include: Performing three-dimensional modeling of the area to be modeled based on multiple frames of images of the area to be modeled collected by the drone and the flight data of the drone, to obtain a three-dimensional model of the area to be modeled, wherein the three-dimensional model is three-dimensional point cloud data; Divide the point cloud into multiple categories according to the normal vector and three-dimensional coordinates of each point cloud in the three-dimensional point cloud data; According to the variance of the minimum principal component directional projection values ​​of all point clouds in each category and the dispersion value of each category, the classification evaluation value of each category is determined to evaluate the classification effect of each category; The dispersion value is negatively correlated with the minimum principal component directional projection value of the point cloud in the category, and positively correlated with the Euclidean distance between the point cloud and the nearest boundary point; Taking any one category as the target category, determining a reclassification coefficient between the target category and each adjacent category of the target category according to the classification evaluation value of the target category, the classification evaluation value of each adjacent category of the target category, and the similarity between the target category and each adjacent category, wherein the reclassification coefficient is positively correlated with the similarity; According to the size of the reclassification coefficient, the categories to be reclassified are selected for reclassification; The edge of the three-dimensional model is obtained according to the result of the reclassification, the dimension data of the three-dimensional model is obtained according to the edge, and the engineering quantity is calculated according to the dimension data.

2. The method for calculating engineering quantities by three-dimensional modeling of drone aerial photography according to claim 1 is characterized in that: The point cloud is divided into multiple categories according to the normal vector and three-dimensional coordinates of each point cloud in the three-dimensional point cloud data, including: The three-dimensional coordinates and normal vectors of the point cloud constitute the feature values ​​of the point cloud, the feature values ​​of all the point clouds in the three-dimensional point cloud data are used as the first data samples, and the first data samples are clustered to obtain multiple categories.

3. The method for calculating engineering quantity by three-dimensional modeling of drone aerial photography according to claim 1 is characterized in that: The method for obtaining the minimum principal component direction projection value is: Perform principal component analysis on the eigenvalues ​​of all point clouds in each category to obtain the minimum principal component direction of each category; obtain the projection value of the eigenvalue of each point cloud in each category on the minimum principal component direction of the category as the minimum principal component direction projection value of each point cloud.

4. The method for calculating engineering quantity by three-dimensional modeling of drone aerial photography according to claim 1, characterized in that: The dispersion value satisfies the expression: ； in, Indicates The dispersion value of each category; Indicates of the categories The minimum principal component direction projection value of the point cloud; Indicates The mean of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates of the categories Point cloud and The distance between the categories The Euclidean distance between the nearest boundary points of a point cloud; Indicates The number of point clouds in each category; is a S-type function: .

5. The method for calculating engineering quantities by three-dimensional modeling based on drone aerial photography according to claim 1 or 4, characterized in that: The classification evaluation value satisfies the expression: ； In the formula, Indicates The classification evaluation value of each category; Indicates The variance of the minimum principal component directional projection values ​​of all point clouds in the category; Indicates The dispersion value of each category; is a hyperparameter; is an exponential function with a natural constant as its base.

6. The method for calculating engineering quantities by three-dimensional modeling based on drone aerial photography according to claim 1 or 4, characterized in that: The method for obtaining the nearest boundary point is: For each point cloud in each category, the Euclidean distance between the point cloud and all boundary points in the category is calculated, and the boundary point with the smallest Euclidean distance to the point cloud is taken as the boundary point closest to the point cloud.

7. The method for calculating engineering quantity by three-dimensional modeling of drone aerial photography according to claim 1, characterized in that: The reclassification coefficient satisfies the expression: ； In the formula, Indicates the target category and the The reclassification coefficient between adjacent categories; Represents the classification evaluation value of the target category; The target category The classification evaluation values ​​of adjacent categories; The eigenvector representing the minimum principal component direction of the target category is The cosine similarity between the feature vectors in the direction of the minimum principal component of adjacent categories, represents the absolute value, Indicates the target category and the The similarity between adjacent categories; represents the minimum function; is an exponential function with a natural constant as its base.

8. The method for calculating engineering quantity by three-dimensional modeling of drone aerial photography according to claim 1, characterized in that: The method of screening the categories to be reclassified according to the size of the reclassification coefficient includes: In response to a reclassification coefficient between a target category and one of its adjacent categories being greater than a preset classification threshold, the target category and the adjacent category are respectively regarded as a category to be reclassified; and all the categories to be reclassified are obtained by judging the reclassification coefficient between each category and each of its adjacent categories.

9. The method for calculating engineering quantity by three-dimensional modeling based on drone aerial photography according to claim 1 or 8, characterized in that: The reclassification includes: All the categories to be reclassified are divided into multiple groups, each category in each group has at least one adjacent category in the group, and each category in each group is not adjacent to all categories in other groups; for each group, the characteristic values ​​of all point clouds in all the categories to be reclassified in the group are used as the second data samples of the group, the second data samples are clustered into multiple categories, and the categories to be reclassified are re-obtained according to the reclassification coefficients between each category and its adjacent categories obtained by clustering, and reclassification is performed again until there are no categories to be reclassified or the number of reclassifications reaches a preset threshold.

10. The method for calculating engineering quantity by three-dimensional modeling of drone aerial photography according to claim 1, characterized in that: The step of obtaining the edge of the three-dimensional model according to the reclassification result includes: Calculate the normal plane corresponding to each category, project the point cloud data of each category onto the normal plane of the category to obtain the projected two-dimensional image, use the image edge detection method to implement edge detection in the two-dimensional image to obtain the edge of the two-dimensional image, map the edge of the two-dimensional image back to the three-dimensional model to obtain the edge of the three-dimensional model.

Images