High-precision topographic surveying and mapping method and system based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of topographic surveying and mapping, in particular to a high-precision topographic surveying and mapping method and system based on an unmanned aerial vehicle. A high-precision topographic surveying and mapping method and system based on an unmanned aerial vehicle comprises a data acquisition unit, a data analysis unit and a data output unit, and the data acquisition unit is used for collecting ground data in real time; the data analysis unit is used for filtering and classifying the image data and the three-dimensional point cloud data collected by the data acquisition unit through point cloud, and establishing a three-dimensional model; and the data output unit corrects and compares the three-dimensional model obtained by the data analysis unit through a latest iterative algorithm and affine transformation. When three-dimensional reconstruction is carried out on the earth surface, the energy function is minimized through Poisson surface reconstruction, smooth transition of the surface can be achieved while original point cloud details are kept, the reconstructed surface is more natural and smoother, the sawtooth effect is reduced, and therefore a more accurate earth surface three-dimensional model result is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of topographic surveying and mapping, and specifically, to a high-precision topographic surveying and mapping method and system based on an unmanned aerial vehicle (UAV). Background Art

[0002] A high-precision topographic surveying and mapping system is a system integrating multiple advanced technologies, aiming to provide extremely accurate and detailed topographic data. Such a system is widely applied in multiple fields such as geological exploration, urban planning, engineering construction, environmental protection, etc., and can meet the highly accurate requirements for topographic information.

[0003] In traditional surveying and mapping systems, due to the difficulty in accurately registering the data collected from different perspectives, the model consistency is poor; and the curved surface of the 3D model may appear rough and cannot truly reflect the surface features. In view of this, a high-precision topographic surveying and mapping method and system based on an unmanned aerial vehicle are designed. Summary of the Invention

[0004] The purpose of the present invention is to provide a high-precision topographic surveying and mapping method and system based on an unmanned aerial vehicle, so as to solve the problems in the above background art that a large amount of time and manpower are required for on-site operations, and it is difficult to complete tasks for complex terrains.

[0005] To achieve the above purpose, on the one hand, the present invention aims to provide a high-precision topographic surveying and mapping system based on an unmanned aerial vehicle, including:

[0006] A data acquisition unit, which is used to collect ground data in real time;

[0007] A data analysis unit, which is used to obtain accurate point cloud data by point cloud classification registration and geometric transformation of the image data and 3D point cloud data collected by the data acquisition unit;

[0008] A data output unit, which establishes a 3D surface model of the ground through the Delaunay algorithm for the point cloud data obtained by the data analysis unit.

[0009] As a further improvement of this technical solution, the data collected by the data acquisition unit includes image data and 3D point cloud data;

[0010] Among them, a UAV is used to carry a camera to extract image data, mainly including RGB images; a UAV is used to carry a sensor to extract 3D point cloud data, mainly including the 3D coordinates of the surface spatial position.

[0011] As a further improvement of this technical solution, the data analysis unit is used to analyze and process the collected information, mainly including a point cloud classification module, a point cloud registration module, and a geometric transformation module;

[0012] Among them, the point cloud classification module filters and classifies the point cloud by introducing height information based on the three-dimensional point cloud data in the data acquisition unit. The point cloud registration module registers the point cloud data obtained by the point cloud classification module through an iterative algorithm. The geometric transformation module performs geometric transformation on the registered point cloud data of the point cloud registration module through an affine transformation algorithm.

[0013] As a further improvement of this technical solution, the specific steps for the point cloud classification module to implement point cloud data filtering and classification are as follows:

[0014] S1.1. Collect three-dimensional point cloud data from the data acquisition unit, convert all the point cloud data into a unified geographic coordinate, and remove the incorrect points;

[0015] S1.2. Use the local height statistics algorithm to filter the point cloud data;

[0016] S1.3. Use the geometric feature extraction model to extract features from the filtered point cloud data;

[0017] S1.4. Use the support vector machine algorithm to classify the point cloud data.

[0018] As a further improvement of this technical solution, in S1.2, the steps for the local height statistics algorithm to perform point cloud filtering are as follows:

[0019] S2.1. Determine the neighborhood of each point;

[0020] N r (p i ) = {p j |d(p j , p i ) ≤ r} (1)

[0021] Among them, N r (p i ) represents the neighborhood set of point p i ; d(p j , p i ) represents the Euclidean distance between point p j and point p i ; r represents the set neighborhood radius;

[0022] S2.2. Introduce height information to calculate the average height and standard deviation;

[0023]

[0024] Among them, represents the average height within the neighborhood of point p i ; |N r (p i) represents N r (p i ) the number of interior points; z j represents the height value of the j-th point in the neighborhood; σ(x i , y i ) represents at (x i , y i ), the point p i the standard deviation of the heights of the neighborhood points of p; x i represents the abscissa of the pixel point p i ; y i represents the ordinate of the pixel point p i ;

[0025] S2.3. Use the average height and standard deviation to distinguish the ground and non-ground point cloud data:

[0026]

[0027] where G(p i ) represents the filtering result.

[0028] As a further improvement of this technical solution, in S1.4, the support vector machine algorithm is used for point cloud classification, and the specific steps are as follows:

[0029] S3.1. Use a feature vector to represent the features of each point p i :

[0030] x i = [x, y, z, n x , n y , n z , κ i (5)

[0031] where,

[0032] p i = (x, y, z) (6)

[0033] n i = (n x , n y , n z ) (7)

[0034] n i represents the normal vector of the point p i ; k i represents the curvature of the point p i ; x represents the coordinate of the point on the X-axis; y represents the coordinate of the point on the Y-axis; z represents the coordinate of the point on the Z-axis; n x represents the component of the normal vector in the X-axis direction; n y represents the component of the normal vector in the Y-axis direction; nz Represents the component of the normal vector in the Z-axis direction;

[0035] S3.2. Classify the feature points using a support vector machine:

[0036] f(x) = w T x + b (8)

[0037] where w represents the weight vector; b represents the bias term;

[0038] S3.3. Minimize the objective function to find the optimal w and b:

[0039]

[0040] where C represents the regularization parameter; ξ i represents the slack variable; w represents the weight vector; and satisfies the following conditions:

[0041]

[0042] where y i represents the label of the i-th training sample; w T represents the transpose of w; x i represents the feature vector of the i-th training sample; n represents the total number of training samples; represents for all samples;

[0043] 3.4. Represent the classification type using a decision function:

[0044]

[0045] where, represents the predicted class label of the i-th sample.

[0046] As a further improvement of this technical solution, the iterative algorithm performs point cloud registration, and its specific steps are as follows:

[0047] S4.1. Use the initialized transformation parameters for each p i to find the nearest q j

[0048] P = {p1, p2,..., p N} (12)

[0049] Q = {q1, q2,..., q M} (13)

[0050] T(p i ) = Rp i + t 14)

[0051] where,

[0052] p i =(x i , y i , z i )(15)

[0053] q j =(x' j , y' j , z' j )(16)

[0054] P represents the source point set; Q represents the target point set; T(p i ) represents the position of point p i after transformation; N represents the number of newly acquired point clouds; M represents the number of reference point clouds; p N represents the Nth newly acquired point; q M represents the Mth reference point; R represents the rotation matrix; t represents the translation vector; p i represents the coordinates of the newly acquired point; q j represents the coordinates of the reference point; x i represents the X-axis coordinate of the newly acquired point i; y i represents the Y-axis coordinate of the newly acquired point i; z i represents the Z-axis coordinate of the newly acquired point i; x' j represents the X-axis coordinate of the reference point cloud j; y' j represents the Y-axis coordinate of the reference point cloud j; z' j represents the Z-axis coordinate of the reference point cloud j;

[0055] S4.2. Solve the optimal transformation by minimizing the distance between point pairs;

[0056]

[0057] i ) represents the position of point p i after transformation; q i represents the corresponding point in the target point set Q.

[0058] As a further improvement of this technical solution, the affine transformation algorithm corrects the point cloud data, and the steps are as follows:

[0059] S5.1. Construct an affine transformation:

[0060] p' i = Ap i + b (18)

[0061] where A represents the affine transformation matrix; b represents the translation vector;

[0062] S5.2. Error analysis is performed using the root mean square error:

[0063]

[0064] Among them, p' i represents the point after transformation; among them, q i represents the nearest neighbor point in the target point set corresponding to the transformed p i .

[0065] As a further improvement of this technical solution, the data output unit mainly includes a three-dimensional visualization module;

[0066] Among them, the three-dimensional visualization module establishes a three-dimensional surface model through the Delaunay algorithm for the point cloud data corrected by the data analysis unit.

[0067] As a further improvement of this technical solution, the steps for the Delaunay algorithm to establish a three-dimensional surface model are as follows:

[0068] S6.1. Remove outliers and noise points to reduce redundant points:

[0069]

[0070] Among them, G represents the set of classified ground points, N k (p i ) represents the set of points within the k-neighborhood of p i , μ i represents the mean distance; σ i represents the standard deviation; n represents the number of point clouds;

[0071]

[0072] Among them, G filtered represents the set after filtering; V k is the k-th voxel, and c k is the center point of this voxel;

[0073] S6.2. Use the Delaunay triangulation algorithm to generate a triangular mesh that satisfies the empty circle property:

[0074] T = DelaunayTriangulation(G filtered ) (22)

[0075] Among them, for those not in G filtered , the nearest neighbor method is used for estimation:

[0076]

[0077] S6.3. Perform surface reconstruction using the Poisson algorithm:

[0078]

[0079] Among them, represents the gradient at point pi; n i represents the normal vector of point p i . λ represents the regularization parameter; dV represents the volume element, and S(x, y, z) represents the smooth plane;

[0080] S6.4. Smooth the surface using Laplacian:

[0081]

[0082] Among them, represents the position of point pi after the (k + 1)-th iteration; represents the position of i at the k-th iteration; α represents the smoothing factor; N(i) represents the set of neighboring vertices of vertex i; represents the position of the j-th neighbor point in the neighborhood set N(i) at the k-th iteration; k represents the number of iterations; j represents the number of source points;

[0083] S6.5. Map the RGB image and color information onto the three-dimensional surface model using texture mapping:

[0084] C(p) = w tl C tl + w tr C tr + w bl C bl + w br C br (26)

[0085] Among them, C(p) represents the color value of point p; C tl represents the color value of the upper-left pixel in the I rgb image; C tr represents the color value of the upper-right pixel in the I rgb image; C bl represents the color value of the lower-left pixel in the I rgb image; C br represents the color value of the lower-right pixel in the I rgb image; W tl represents the weight of the upper-left pixel color value; W tr represents the weight of the upper-right pixel color value; W bl represents the weight of the lower-left pixel color value; W br represents the weight of the lower-right pixel color value.

[0086] On the other hand, the present invention provides a high-precision terrain mapping method based on an unmanned aerial vehicle, which is used for a high-precision terrain mapping system based on an unmanned aerial vehicle described in any one of the above, and includes the following steps:

[0087] S10.1. Determine the mapping range through the ground station software, and set the flight path, acquisition altitude, coverage range, image overlap degree, and data acquisition mode;

[0088] S10.2. Start the unmanned aerial vehicle, fly along the planned flight path, record the geographical coordinates of each sampling point through the data acquisition unit, capture the surface image, and obtain the three-dimensional spatial point cloud;

[0089] S10.3. Classify and register the point cloud data through the point cloud classification module and the point cloud registration module, and establish a three-dimensional surface model of the ground through the Delaunay algorithm using the point cloud data;

[0090] S10.4. Map the RGB image and color information onto the three-dimensional surface model by texture mapping;

[0091] S10.5. The data output unit establishes a three-dimensional surface model of the ground through the Delaunay algorithm and outputs a three-dimensional terrain file according to requirements.

[0092] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0093] In the high-precision terrain mapping method and system based on an unmanned aerial vehicle, when performing three-dimensional surface reconstruction, height is introduced to filter and classify the point cloud data, the classified point cloud data is registered and geometrically transformed, and the processed point cloud data can generate a triangular mesh that satisfies the empty circle property through the Delaunay algorithm, that is, the circumcircle of any triangle does not contain other points. This characteristic ensures the optimal distribution between triangles and avoids the generation of long and narrow triangles, thereby improving the quality of the three-dimensional model;

[0094] At the same time, by minimizing the energy function through Poisson surface reconstruction, smooth transition of the surface can be achieved while maintaining the details of the original point cloud, making the reconstructed surface more natural and smooth, reducing the jagged effect, and thus obtaining a more accurate three-dimensional surface model result. Description of the Drawings

[0095] Figure 1 It is the overall flow block diagram of the present invention;

[0096] The meanings of each label in the figure are as follows: 1. Data acquisition unit; 2. Data analysis unit; 21. Point cloud classification module; 22. Data registration module; 23. Geometric transformation module; 3. Data output unit; 31. Three-dimensional visualization module. Detailed Embodiments

[0097] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0098] Embodiment 1:

[0099] Please refer to Figure 1 As shown, a high-precision terrain mapping system based on an unmanned aerial vehicle is provided, including a data acquisition unit 1, and the data acquisition unit 1 is used to collect ground information in real time;

[0100] In this embodiment, the ground data collected by the data acquisition unit 1 includes image data and three-dimensional point cloud data;

[0101] Among them, an unmanned aerial vehicle is used to carry a camera to extract image data, mainly including RGB images; an unmanned aerial vehicle is used to carry a sensor to extract three-dimensional point cloud data, mainly including three-dimensional coordinates of the surface space position.

[0102] This high-precision terrain mapping system based on an unmanned aerial vehicle further includes a data analysis unit 2, and the data analysis unit 2 is used to obtain accurate point cloud data by performing point cloud classification registration and geometric transformation on the image data and three-dimensional point cloud data collected by the data acquisition unit 1;

[0103] In this embodiment, the specific steps for the point cloud classification module 21 to implement point cloud data filtering and classification are as follows:

[0104] S1.1. Collect three-dimensional point cloud data from the data acquisition unit 1, convert all the point cloud data into a unified geographic coordinate, and remove the wrong points;

[0105] S1.2. Use the local height statistics algorithm to perform filtering processing on the point cloud data;

[0106] S1.3. Use the geometric feature extraction model to extract features from the filtered point cloud data;

[0107] S1.4. Use the support vector machine algorithm to classify the point cloud data.

[0108] In this embodiment, in S1.2, the steps for the local height statistics algorithm to perform point cloud filtering processing are as follows:

[0109] S2.1. Determine the neighborhood of each point;

[0110] N r (p i ) = {p j|d(p j ,p i )≤r} (1)

[0111] where N r (p i ) represents the neighborhood set of point p i ; d(p j ,p i ) represents the Euclidean distance between point p j and point p i ; r represents the set neighborhood radius;

[0112] S2.2. Introduce height information to calculate the average height and standard deviation;

[0113]

[0114] where represents the average height within the neighborhood of point p i ; |N r (p i )| represents the number of points within N r (p i ); z j represents the height value of the j-th point within the neighborhood; σ(x i ,y i ) represents the standard deviation of the heights of the neighborhood points of point p i at (x i ,y i ); x i represents the abscissa of pixel point p i ; y i represents the ordinate of pixel point p i ;

[0115] S2.3. Use the average height and standard deviation to distinguish between ground and non-ground point cloud data:

[0116]

[0117] where G(p i ) represents the filtering result.

[0118] In this embodiment, in S1.4, the support vector machine algorithm is used for point cloud classification, and the specific steps are as follows:

[0119] S3.1. Represent the features of each point p i using a feature vector:

[0120] x i = [x, y, z, n x , n y , n z , κi (5)

[0121] Among them,

[0122] p i =(x, y, z) (6)

[0123] n i =(n x , n y , n z ) (7)

[0124] n i represents the normal vector of point p i ; k i represents the curvature of point p i ; x represents the coordinate of the point on the X-axis; y represents the coordinate of the point on the Y-axis; z represents the coordinate of the point on the Z-axis; n x represents the component of the normal vector in the X-axis direction; n y represents the component of the normal vector in the Y-axis direction; n z represents the component of the normal vector in the Z-axis direction;

[0125] S3.2. Use a support vector machine to classify the feature points:

[0126] f(x)=w T x + b (8)

[0127] Among them, w represents the weight vector; b represents the bias term;

[0128] S3.3. Minimize the objective function to find the optimal w and b:

[0129]

[0130] i Among them, C represents the regularization parameter; ξ

[0131]

[0132] Among them, y i represents the label of the i-th training sample; w T represents the transpose of w; x i represents the feature vector of the i-th training sample; n represents the total number of training samples; represents for all samples;

[0133] 3.4. Use a decision function to represent the classification type:

[0134]

[0135] Among them, Represents the class label predicted for the i-th sample.

[0136] In this embodiment, an iterative algorithm is used for point cloud registration, and its specific steps are as follows:

[0137] S4.1. Find the nearest qj for each pi using the initial transformation parameters

[0138] P = {p1, p2,..., p N} (12)

[0139] Q = {q1, q2,..., q M} (13)

[0140] T(p i ) = Rp i + t 14)

[0141] Where

[0142] p i = (x i , y i , z i ) (15)

[0143] q j = (x′ j , y′ j , z′ j ) (16)

[0144] P represents the source point set; Q represents the target point set; T(p i ) represents the position of point p i after transformation; N represents the number of newly acquired point clouds; M represents the number of reference point clouds; p N represents the N-th newly acquired point; q M represents the M-th reference point; R represents the rotation matrix; t represents the translation vector; pi represents the coordinates of the newly acquired point; qj represents the coordinates of the reference point; x i represents the X-axis coordinate of the newly acquired point i; y i represents the Y-axis coordinate of the newly acquired point i; z i represents the Z-axis coordinate of the newly acquired point i; x′ j represents the X-axis coordinate of the reference point cloud j; y′ j represents the Y-axis coordinate of the reference point cloud j; z′ j represents the Z-axis coordinate of the reference point cloud j;

[0145] S4.2. Solve for the optimal transformation by minimizing the distance between point pairs;

[0146]

[0147] Among them, t(p i ) represents the position of point p i after transformation; q i represents the corresponding point in the target point set Q.

[0148] In this embodiment, the affine transformation algorithm is used to correct the point cloud data, and the steps are as follows:

[0149] S5.1. Construct an affine transformation:

[0150] p′ i = Ap i + b (18)

[0151] Among them, A represents the affine transformation matrix; b represents the translation vector;

[0152] S5.2. Perform error analysis using the root mean square error:

[0153]

[0154] Among them, p′ i represents the point after transformation; among them, q i represents the nearest neighbor point in the target point set corresponding to the transformed p i .

[0155] In this embodiment, the high-precision terrain mapping system based on the unmanned aerial vehicle further includes a data output unit 3, and the data output unit 3 establishes a three-dimensional surface model through the Delaunay algorithm for the point cloud data obtained by the data analysis unit 2.

[0156] In this embodiment, the data output unit 3 mainly includes a three-dimensional visualization module 31;

[0157] Among them, the three-dimensional visualization module 31 establishes a three-dimensional surface model through the Delaunay algorithm for the point cloud data corrected by the data analysis unit 2.

[0158] In this embodiment, the Delaunay algorithm is used to establish a three-dimensional surface model, and the specific steps are as follows:

[0159] S6.1. Remove outliers and noise points and reduce redundant points:

[0160]

[0161] Among them, G represents the set of classified ground points, N k (p i ) represents the set of points within the k-neighborhood of p i , μ i represents the distance mean; σ i represents the standard deviation; n represents the number of point clouds;

[0162]

[0163] Among them, G filtered represents the set after filtering; V k is the k-th voxel, and c k is the center point of this voxel;

[0164] S6.2. Use the Delaunay triangulation algorithm to generate a triangular mesh that satisfies the empty circle property:

[0165] T = DelaunavTriangulation(G Filtered ) (22)

[0166] Among them, for those not in G filtered , the nearest neighbor method is used for estimation:

[0167]

[0168] S6.3. Use the Poisson algorithm for surface reconstruction:

[0169]

[0170] Among them, represents the gradient at point p i ; n i represents the normal vector of point p i ; λ represents the regularization parameter; dV represents the volume element, and S(x, y, z) represents the smooth plane;

[0171] S6.4. Use Laplacian to smooth the surface:

[0172]

[0173] Among them, represents the position of point pi after the (k + 1)-th iteration; represents the position of i at the k-th iteration; α represents the smoothing factor; N(i) represents the set of neighboring vertices of vertex i; represents the position of the j-th neighbor point in the neighborhood set N(i) at the k-th iteration; k represents the number of iterations; j represents the number of source points;

[0174] S6.5. Use texture mapping to map the RGB image and color information onto the 3D surface model of the ground:

[0175] C(p) = w tl C tl + w tr C tr + w bl C bl+w br C br (26)

[0176] Among them,

[0177] w tl = (1 - u)(1 - v);

[0178] w tr = u(1 - v);

[0179] w bl = (1 - u)v;

[0180] w br = uv;

[0181] u = X;

[0182] v = Y;

[0183] C(p) represents the color value of point p; C tl represents I rgb the color value of the upper-left pixel in the image; C tr represents I rgb the color value of the upper-right pixel in the image; C bl represents I rgb the color value of the lower-left pixel in the image; C br represents I rgb the color value of the lower-right pixel in the image; w tl represents the weight of the upper-left pixel color value; w tr represents the weight of the upper-right pixel color value; w bl represents the weight of the lower-left pixel color value; w br represents the weight of the lower-right pixel color value; u represents the u coordinate; v represents the V coordinate; X represents the Cartesian coordinate x-axis position; Y represents the Cartesian coordinate y-axis position.

[0184] Example 2:

[0185] The difference between Example 2 and Example 1 of the present invention is that this example introduces a high-precision terrain mapping method based on an unmanned aerial vehicle used in a high-precision terrain mapping system based on an unmanned aerial vehicle.

[0186] A high-precision terrain mapping method based on an unmanned aerial vehicle, for a high-precision terrain mapping system based on an unmanned aerial vehicle in any one of the above, includes the following steps:

[0187] S10.1. Determine the mapping range through the ground station software, set the flight path, acquisition altitude, coverage range, image overlap degree, and data acquisition mode (image, point cloud, or multi-mode);

[0188] S10.2. Start the drone and fly along the planned route. Record the geographical coordinates of each sampling point through the data acquisition unit 1, capture surface images and obtain three-dimensional spatial point clouds;

[0189] S10.3. Classify and register the point cloud data through the point cloud classification module 21 and the point cloud registration module 22, and use the point cloud data to establish a three-dimensional surface model through the Delaunay algorithm;

[0190] S10.4. Use texture mapping to map the RGB image and color information onto the three-dimensional surface model;

[0191] S10.5. The data output unit 3 establishes a three-dimensional surface model through the Delaunay algorithm and outputs a three-dimensional terrain file (in.obj or.las format) according to requirements.

[0192] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. A high-precision terrain mapping system based on drones, characterized in that: include A data collection unit (1), the data collection unit (1) is used to collect ground data in real time; A data analysis unit (2), the data analysis unit (2) is used to obtain accurate point cloud data by performing point cloud classification registration and geometric transformation on the image data and three-dimensional point cloud data collected by the data acquisition unit (1); A data output unit (3) is used to output the point cloud data obtained by the data analysis unit (2) through Delauna y The algorithm builds a three-dimensional model of the surface.

2. The high-precision terrain mapping system based on an unmanned aerial vehicle according to claim 1 is characterized in that: The ground data collected by the data collection unit (1) includes image data and three-dimensional point cloud data; The drone carries a camera to extract image data, mainly including RGB images; the drone carries a sensor to extract three-dimensional point cloud data, mainly including the three-dimensional coordinates of the surface space position.

3. The high-precision terrain mapping system based on an unmanned aerial vehicle according to claim 1, characterized in that: The data analysis unit (2) is used to analyze and process the collected information, and mainly includes a point cloud classification module (21), a point cloud registration module (22) and a geometric transformation module (23); The point cloud classification module (21) introduces height information to filter and classify the point cloud based on the three-dimensional point cloud data in the data acquisition unit (1); the point cloud registration module (22) registers the point cloud data obtained by the point cloud classification module (21) through an iterative algorithm; and the geometric transformation module (23) performs geometric transformation on the point cloud data registered by the point cloud registration module (22) through an affine transformation algorithm.

4. The high-precision terrain mapping system based on an unmanned aerial vehicle according to claim 3 is characterized in that: The point cloud classification module (21) implements the point cloud data filtering and classification in the following specific steps: S1.1, collecting three-dimensional point cloud data from the data acquisition unit (1), converting all point cloud data into unified geographic coordinates, and removing erroneous points; S1.2, using local height statistics algorithm to filter the point cloud data; S1.3, using a geometric feature extraction model to extract features from the filtered point cloud data; S1.

4. Use support vector machine algorithm to classify point cloud data.

5. The high-precision terrain mapping system based on unmanned aerial vehicle according to claim 4 is characterized in that: In S1.2, the steps of performing point cloud filtering processing by the local height statistics algorithm are as follows: S2.1, determine the neighborhood of each point; N r (p i )={p j |d(p j ,p i )≤r} (1) Among them, N r (p i ) represents point p i The neighborhood set of d(p j , p i ) represents point p j and point p i The Euclidean distance between them; r represents the set neighborhood radius; S2.2, introduce height information to calculate the average height and standard deviation; in, Represents point p i The average height in the neighborhood; |N r (p i )| represents N r (p i )The number of internal points; z j represents the height value of the jth point in the neighborhood; σ(x i ,y i ) means in (x i ,y i ), point p i The standard deviation of the height of the neighborhood points; x i Indicates the horizontal coordinate of pixel point p1; y i Represents pixel p i The vertical coordinate of S2.

3. Use average height and standard deviation to distinguish ground and non-ground point cloud data: Among them, G(p i ) represents the filtering result.

6. The high-precision terrain mapping system based on unmanned aerial vehicle according to claim 4 is characterized in that In S1.4, the support vector machine algorithm performs point cloud classification, and the specific steps are as follows: S3.

1. Use feature vector to represent each point p i Features: x i [x,y,z,n x ,n y ,n z κ i ] (5) in, p i =(x,y,z) (6) n i =(n x ,n y ,n z ) (7) n i Represents point p i The normal vector of i Represents point p i The curvature of the point; x represents the coordinate of the point on the X axis; y represents the coordinate of the point on the Y axis; z represents the coordinate of the point on the Z axis; n x Represents the component of the normal vector in the X-axis direction; n y Represents the component of the normal vector in the Y-axis direction; n z Represents the component of the normal vector in the Z-axis direction; S3.2, use support vector machine to classify feature points: F(x)=w T x+b (8) Among them, w represents the weight vector; b represents the bias term; S3.

3. Minimize the objective function and find the optimal w and b: Where C represents the regularization parameter; ξ i represents the slack variable; w represents the weight vector; satisfies the following conditions: Among them, y i represents the label of the i-th training sample; w T represents the transpose of w; x i represents the feature vector of the i-th training sample; n represents the total number of training samples; Vi represents all samples; 3.

4. Use decision function to represent classification type: in, Represents the category label predicted for the i-th sample.

7. The high-precision terrain mapping system based on an unmanned aerial vehicle according to claim 3 is characterized in that: The iterative algorithm performs point cloud registration, and its specific steps are: S4.1, using initialization transformation parameters for each p i Find the nearest neighbor q j P={p1,p2,…,p N } (12) Q={q1,q2,...,q M } (13) T(p i )=Rp i +t (14) in, p i =(x i ,y i ,z i ) (15) q j =(x′ j ,y′ j ,z′ j ) (16) P represents the source point set; Q represents the target point set; T(p i ) represents point p i The transformed position; N represents the number of newly acquired point clouds; M represents the number of reference point clouds; p N Indicates the Nth newly acquired point; q M represents the Mth reference point; R represents the rotation matrix; t represents the translation vector; p i Indicates the newly acquired point coordinates; q j Indicates the coordinates of the reference point; x i Indicates the X-axis coordinate of the newly acquired point i; y i Indicates the Y-axis coordinate of the newly acquired point i; z i Indicates the Z-axis coordinate of the newly acquired point i; x′ j Represents the X-axis coordinate of the reference point cloud j; y′ j Represents the Y-axis coordinate of the reference point cloud j; z′ j Represents the Z-axis coordinate of the reference point cloud j; S4.2, the optimal transformation is solved by minimizing the distance between point pairs; Among them, T(p i ) represents point p i The transformed position; q i represents the corresponding points in the target point set Q.

8. The high-precision terrain mapping system based on an unmanned aerial vehicle according to claim 9 is characterized in that: The affine transformation algorithm performs point cloud data correction, and the steps are as follows: S5.

1. Construct affine transformation: p′ i =Ap i +b (18) Where A represents the affine transformation matrix; b represents the translation vector; S5.

2. Error analysis using root mean square error: Among them, p′ i represents the point after transformation; where q i Represents the target point concentration and the transformed p i The corresponding nearest neighbor.

9. The high-precision terrain mapping system based on unmanned aerial vehicle according to claim 1, characterized in that: The data output unit (3) mainly comprises a three-dimensional visualization module (31); The three-dimensional visualization module (31) uses the Delaunay algorithm to construct a three-dimensional model of the surface using the point cloud data corrected by the data analysis unit (2); The Delaunay algorithm establishes a three-dimensional surface model, and its specific steps are as follows: S6.

1. Remove outliers and noise points and reduce redundant points: Among them, G represents the set of classified ground points, N k (p i ) indicates p i The point set in the k-neighborhood of i represents the distance mean; σ i represents the standard deviation; n represents the number of point clouds; Among them, G filtered represents the set after filtering; V k is the kth voxel, c k is the center point of the voxel; S6.

2. Use the Delaunay triangulation algorithm to generate a triangulated network that satisfies the empty circle property: T=DelaunayTriangulation(G filtered ) (22) Among them, for those not in G filtered In the paper, the nearest neighbor method is used to estimate: S6.

3. Surface reconstruction using Poisson algorithm: in, Represents point p i The gradient at n i Represents point p i The normal vector of ; λ represents the regularization parameter; dV represents the volume element, S(x, y, z) represents the smooth plane; S6.4, using Laplacian to smooth the surface: in, It means point p after the k+1th iteration i location; represents the position of i at the kth iteration; α represents the smoothing factor; N(i) represents the set of neighboring vertices of vertex i; Indicates the position of the jth neighbor point in the neighborhood set N(i) at the kth iteration; k represents the number of iterations; j represents the number of source points; S6.

5. Use texture mapping to map the RGB image and color information onto the three-dimensional surface model: C(p)=w tl C tl +w tr C tr +w bl C bl +w br C br (26) Where C(p) represents the color value of point p; C tl Indicates I rgb The color value of the upper left pixel in the image; C tr Indicates I rgb The color value of the upper right pixel in the image; C bl Indicates I rgb The color value of the lower left pixel in the image; C br Indicates I rgb The color value of the lower right pixel in the image; w tl Indicates the weight of the color value of the upper left pixel; w tr Indicates the weight of the color value of the upper right pixel; w bl Indicates the weight of the color value of the lower left pixel; w br Represents the weight of the color value of the lower right pixel.

10. A high-precision terrain mapping method based on an unmanned aerial vehicle, used in the high-precision terrain mapping system based on an unmanned aerial vehicle as claimed in any one of claims 1 to 9, characterized in that: The steps include: S10.

1. Determine the surveying range through the ground station software, set the flight path, acquisition altitude, coverage, image overlap and data acquisition mode; S10.2, start the UAV and fly it along the planned route to record the geographic coordinates of each sampling point through the data acquisition unit (1), capture the surface image and obtain the three-dimensional space point cloud; S10.3, classifying and registering the point cloud data through the point cloud classification module (21) and the point cloud registration module (22), and using the point cloud data to establish a three-dimensional surface model through the Delaunay algorithm; S10.4, using texture mapping to map the RGB image and color information onto the three-dimensional model of the surface; S10.

5. The data output unit (3) establishes a three-dimensional surface model using the Delaunay algorithm and outputs a three-dimensional terrain file as required.

Images