Method for constructing three-dimensional light field model of electric power tower by using unmanned aerial vehicle image

Abstract

The invention relates to the technical field of electric power towers, and discloses an electric power tower three-dimensional light field model construction method using unmanned aerial vehicle images, and the method comprises the steps: obtaining an electric power tower image, processing the electric power tower image, and obtaining internal and external parameters of a camera; sampling along the light direction of the camera in a three-dimensional space to obtain three-dimensional coordinates of corresponding sampling points, and performing position coding on the three-dimensional coordinates of the sampling points and unit vectors in the light direction of the camera to obtain feature vectors of the sampling points; inputting the feature vectors of the sampling points into a neural network to obtain RGB color values and densities of the sampling points, performing neural network training, and obtaining RGB color values and densities of any points in a three-dimensional space from the trained neural network; and extracting a three-dimensional scene from the trained neural network by using a moving cube algorithm to obtain a three-dimensional light field model of the power tower. According to the method, the accuracy of the constructed electric power tower image three-dimensional light field model can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric pole towers, and in particular relates to a method for constructing a three-dimensional light field model of an electric pole tower using unmanned aerial vehicle images. Background Art

[0002] In the field of 3D light field model construction of power poles and towers, machine vision and 3D light field technology are receiving more and more attention and research. Although traditional 3D reconstruction technologies, such as LiDAR and traditional multi-view stereo geometry (Multi-ViewStereo, MVS), have been applied in 3D reconstruction, defect management and patrol management, they have some limitations. For example, LiDAR cannot obtain the color texture of objects, while traditional MVS reconstructs models through feature extraction and matching, stereo matching, parallax smoothing and filling, surface reconstruction and 3D reconstruction, texture mapping and rendering, but due to parallax discontinuity, occlusion problems and computational complexity, problems such as holes and texture aliasing may occur during the reconstruction process. Therefore, there is an urgent need for a new and superior method to restore the spatial light field and realize the automated realistic modeling of power poles and towers. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for constructing a three-dimensional light field model of a power tower using drone images, which can improve the accuracy of the constructed three-dimensional light field model of the power tower image.

[0004] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0005] In a first aspect, the present invention provides a method for constructing a three-dimensional light field model of a power tower using drone images, comprising the following steps:

[0006] Acquire a power pole tower image, perform feature matching and motion recovery structure analysis on the power pole tower image, and obtain camera internal and external parameters, wherein the camera internal and external parameters include position and camera light direction;

[0007] Sampling is performed along the direction of the camera light in three-dimensional space to obtain the three-dimensional coordinates of the corresponding sampling points , and the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Performing position encoding to obtain a feature vector of the sampling point; the feature vector is the vector after position encoding;

[0008] Inputting the feature vector of the sampling point into the neural network, obtaining the RGB color value and density of the sampling point, using volume rendering to obtain a rendered image from the original camera perspective, comparing it with the real power tower image to calculate the loss function, performing neural network training, and obtaining the RGB color value and density of any point in the three-dimensional space from the trained neural network;

[0009] The marching cubes algorithm is used to extract the three-dimensional scene from the trained neural network to obtain the three-dimensional light field model of the power tower.

[0010] Furthermore, the method for obtaining the power pole tower image is to use the power pole tower image obtained by shooting with a drone. The above three-dimensional space is a three-dimensional space established by the coordinate system used by the drone.

[0011] Furthermore, the camera internal and external parameters also include focal length and pixel size.

[0012] It should be noted that: the method of performing feature matching and Structure from Motion (SFM) analysis on power tower images to obtain camera internal and external parameters is an existing technology.

[0013] In combination with the first aspect, further, sampling is performed along the direction of the camera light in the three-dimensional space to obtain the three-dimensional coordinates of the corresponding sampling points. The method is as follows: train the first network and the second network at the same time, use the layered sampling method to obtain N sampling points for each camera light and input them into the first network, and obtain the RGB color value and density of each sampling point output by the first network; according to the density of each sampling point output by the first network, use the inverse transformation sampling method to newly generate M sampling points, input the newly generated M sampling points together with the previous N sampling points into the second network, and obtain the RGB color value and density of each sampling point output by the second network; wherein the sampling points input into the first network are less than the sampling points input into the second network, that is, N <M。

[0014] In combination with the first aspect, further, the three-dimensional coordinates of the sampling point and the unit vector of the camera light direction The specific method for position encoding and obtaining the feature vector of the sampling point is:

[0015] The position encoding equation is:

[0016]

[0017] in, are the three-dimensional coordinates acting on the sampling points Each component and the unit vector of the camera light direction The three components of For the dimension, for Parameters used to represent the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction .

[0018] In combination with the first aspect, further, the architecture of the neural network is: the neural network includes sequentially connected a first fully connected layer and a second fully connected layer, is an even number, and a ReLU activation function is used between each first fully connected layer; the neural network also includes a skip connection, which is used to convert the feature vector after position encoding Input to The first fully connected layer uses the activation layer; the second fully connected layer uses the sigmoid activation function; the feature vector after position encoding Input to the first fully connected layer, and then pass through The feature vector after the first fully connected layer and the positional encoding after the jump connection After passing through the first fully connected layer, the output density , the output density Feature vector after adding the camera light direction encoding Input to the second fully connected layer, and output RGB color value after passing through the second fully connected layer.

[0019] Preferably, The value is 8; the first fully connected layer has 256 channels, and the second fully connected layer has 128 channels.

[0020] In combination with the first aspect, further, the network architecture of the first network and the second network is the same as the architecture of the neural network.

[0021] In combination with the first aspect, further, the loss function of the neural network is as follows:

[0022]

[0023] in, is the loss function, is the set of rays in each training batch in the step of layering rays on the power tower image, is every ray in the set, , , They are the RGB color values ​​of the light output in the power tower image, the first network, and the second network respectively.

[0024] In combination with the first aspect, further, the volume rendering formula is as follows:

[0025] ;

[0026] in, Indicates color, represents all the parameters in the neural network, , Representative is the starting point and the direction is of light; Represents the input light The density of the output of the neural network; Represents input light The color value output by the neural network; The unit vector representing the direction of the camera light; , Represent the far plane and the near plane respectively; is the position of the sampling point along the direction of the camera light, the minimum is , is the position of the sampling point of the light on the nearest plane; Indicates that the light arrive The transmittance within a distance, i.e. the probability that a ray of light can propagate without hitting any other particles, is defined as follows:

[0027] .

[0028] In a second aspect, the present invention provides a device for constructing a three-dimensional light field model of a power tower using drone images, comprising:

[0029] A camera internal and external parameter acquisition module is configured to acquire an image of a power pole tower, perform feature matching and motion recovery structure analysis on the image of the power pole tower, and obtain camera internal and external parameters, wherein the camera internal and external parameters include a position and a camera light direction;

[0030] The feature vector extraction module is configured to sample along the direction of the camera light in three-dimensional space and obtain the three-dimensional coordinates of the corresponding sampling points. , and the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Perform position encoding to obtain the feature vector of the sampling point;

[0031] The RGB color value and density acquisition module is configured to input the feature vector of the sampling point into the neural network, obtain the RGB color value and density of the sampling point, use volume rendering to obtain the rendered image of the original camera perspective, compare it with the real power pole tower image to calculate the loss function, perform neural network training, and obtain the RGB color value and density of any point in the three-dimensional space from the trained neural network;

[0032] The module for obtaining the three-dimensional light field model of the power tower is configured to extract the three-dimensional scene from the trained neural network using the marching cube algorithm to obtain the three-dimensional light field model of the power tower.

[0033] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images.

[0034] In a fourth aspect, the present invention provides a computer device, comprising:

[0035] Memory for storing computer programs;

[0036] The processor is used to execute the computer program to implement the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images.

[0037] In a fifth aspect, the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention can realize rapid three-dimensional modeling of power poles and towers using only ordinary photos, with higher modeling accuracy and lower cost, and solves the problems that three-dimensional reconstruction requires professional equipment, relies on a large amount of observation data, has slow modeling speed and low modeling accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the process of constructing a three-dimensional light field model of a power tower according to the present invention;

[0041] Figure 2 It is a partial schematic diagram of the neural network in the present invention. DETAILED DESCRIPTION

[0042] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0043] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0044] Example 1

[0045] like Figure 1 As shown, this embodiment proposes a method for constructing a three-dimensional light field model of a power tower using drone images, comprising the following steps:

[0046] (1) Obtain the power pole tower image and analyze the power pole tower image to obtain the camera intrinsic parameters and pose parameters; the camera intrinsic parameters include focal length, position and pixel size; the camera pose parameters are the camera extrinsic parameters, including rotation matrix and translation vector. The camera light direction is expressed by the rotation matrix and translation vector.

[0047] (2) Sampling along the direction of the camera light in three-dimensional space to obtain the three-dimensional coordinates of the corresponding sampling points , the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Perform position encoding to obtain the feature vector of the sampling point. Specifically, the sampling point position obtained from the sampling of the camera light direction is encoded, and the coordinates are mapped to a higher-dimensional space, so that the neural network composed of a multilayer perceptron (MLP) can represent higher-frequency functions, making the geometry and texture of the object surface more realistic.

[0048] (3) Inputting the feature vector of the sampling point into the neural network, obtaining the RGB color value and density of the sampling point, using volume rendering to obtain a rendered image from the original camera perspective, comparing it with the real power pole tower image to calculate the loss function, performing neural network training, and obtaining the RGB color value and density of any point in the three-dimensional space from the trained neural network.

[0049] (4) The marching cube algorithm is used to extract the three-dimensional scene from the trained neural network and obtain the three-dimensional light field model of the power tower.

[0050] in, Figure 1 middle In Represents a neural network, which is a function with input and output. Represents all the parameters in this neural network; It represents the density of sampling points, which can reflect the density of light; RGB represents the color value of light.

[0051] In a specific implementation of this embodiment, step (2) performs sampling along the direction of the camera light in three-dimensional space.

[0052] In a specific implementation of this embodiment, step (2) adopts a coarse-to-fine sampling technique to sample along the direction of the camera light in three-dimensional space, densely sample points with large color contribution (i.e., high density), and sparsely sample points with small contribution (i.e., low density) to reduce unnecessary sampling times and fully adapt to the frequency of scene signals. The samples are proportionally distributed according to the expected impact of the obtained light samples on the final rendering, and are used for subsequent optimization of the first network (also known as the coarse network) and the second network (also known as the fine network) to improve rendering efficiency and time efficiency. The specific sampling method is: train a first network and a second network at the same time, use a layered sampling method to obtain 64 sampling points for each camera light and input them into the first network, and obtain the RGB color values ​​and density of the 64 sampling points output by the first network; according to the density of each sampling point output by the first network, use the inverse transformation sampling method to newly generate 128 to 256 sampling points, and input the newly generated 128 to 256 sampling points together with the previous 64 sampling points into the second network to obtain the RGB color value and density of each sampling point output by the second network. By inputting the above sampling method into two networks, one coarse and one fine, respectively, the RGB color value and density of each sampling point can be obtained more accurately, so that the description of the scene can be more accurate.

[0053] Stratified sampling first performs the sampling on the camera ray by (near) and The interval range formed by (far) is divided into equal parts, and then a sampling point is obtained by uniformly sampling in each small interval, with a total of N sampling points:

[0054]

[0055] in, , .

[0056] For the light sampling part, the number of randomly sampled rays is 2048, the chunksize value (block size) during training is 2048, and the chunksize value during verification is 65536. The number of sampling points for each ray on the coarse network and the fine network is 64.

[0057] In a specific implementation of this embodiment, step (2) converts the three-dimensional coordinates obtained by sampling the light into Performing position encoding, mapping it to a higher-dimensional space through a high-frequency function and then passing it to the neural network can better fit data containing high-frequency changes, so that the images generated by learning will not be too blurry;

[0058] The position encoding equation used is:

[0059]

[0060] in, are the three-dimensional coordinates acting on the sampling points Each component and the unit vector of the camera light direction The three components of For the dimension, for Parameters used to represent the three-dimensional coordinates of the sampling points and the unit vector of the viewing direction Dimensions The choice of is related to the complexity of the scene and the computing power of the hardware used, and also determines the maximum frequency that the neural network can learn. In terms of , for the unit vector of the camera light direction In terms of .

[0061] like Figure 2 As shown, the position-encoded feature vector The three-dimensional coordinates are input into the neural network for neural network training. The specific process of neural network training is as follows: The feature vector after position encoding , through 8 first fully connected layers (using ReLU activation function), each with 256 channels and including a skip connection that connects The input is connected to the fifth fully connected layer. After passing through the 8 first fully connected layers, the density of each sampling point is output and a 256-dimensional feature vector. This 256-dimensional feature vector is consistent with the input They are concatenated and processed by an additional second fully connected layer with 128 channels (using sigmoid activation function) to output RGB color.

[0062] In a neural network, the loss function of a neural network is as follows:

[0063]

[0064] in, is the loss function, is the set of rays in each training batch in the step of layering rays on the power tower image, is every ray in the set, , , They are the RGB color values ​​of the light output in the power tower image, the first network, and the second network respectively.

[0065] Figure 2 In the figure, 60 and 24 represent the dimensions of the feature vector, and 60 is the three-dimensional coordinate. The dimension after position encoding, 24 is the unit vector of the camera light direction The dimension after position encoding is represents the density of the location, that is, the density of the sampling points, Represents three-dimensional coordinates The feature vector after position encoding is The unit vector representing the direction of the camera light The feature vector after position encoding.

[0066] In a specific implementation of this embodiment, volume rendering is used in step (3) to obtain a rendered image from the original camera's perspective. The principle of volume rendering technology is to sample in a three-dimensional data set, and then use the focal length and pixel size of the camera to trace rays for each sampling point, and finally generate a two-dimensional image. In this process, the value of each sampling point represents the color (corresponding to the RGB color value) and the transparency (corresponding to the density). Since each pixel value of the two-dimensional image can be regarded as the cumulative superposition of all points on a ray emitted by the camera, the color of each pixel can be obtained by integrating the color value based on the density, thereby rendering and generating a two-dimensional image from this perspective.

[0067] The above sampling results are accumulated and calculated, assuming that the distance from the camera to the near plane is , the distance to the far plane is , then for the sampling ray For example, the final desired color calculation formula is as follows:

[0068]

[0069] in, Indicates color, represents all the parameters in the neural network, , Representative is the starting point and the direction is of light; Represents the input light The density of the output of the neural network; Represents input light The color value output by the neural network; The unit vector representing the direction of the camera light; , Represent the far plane and the near plane respectively; is the position of the sampling point along the direction of the camera light, the minimum is , the maximum is , is the position of the sampling point of the light on the nearest plane, is the position of the sampling point of the light on the farthest plane; Indicates that the light arrive The transmittance within a distance, i.e. the probability that a ray of light can propagate without hitting any other particles, is defined as follows:

[0070] .

[0071] Example 2

[0072] Based on the same inventive concept as Example 1, this example introduces a device for constructing a three-dimensional light field model of a power tower using drone images, comprising:

[0073] The camera internal and external parameter acquisition module is configured to acquire the power pole tower image, perform feature matching and motion recovery structure analysis on the power pole tower image, and obtain the camera internal and external parameters, which include the position and camera light direction;

[0074] The feature vector extraction module is configured to sample along the direction of the camera light in three-dimensional space and obtain the three-dimensional coordinates of the corresponding sampling points. , and the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Perform position encoding to obtain the feature vector of the sampling point;

[0075] The RGB color value and density acquisition module is configured to input the feature vector of the sampling point into the neural network, obtain the RGB color value and density of the sampling point, use volume rendering to obtain the rendered image of the original camera perspective, compare it with the real power pole tower image to calculate the loss function, perform neural network training, and obtain the RGB color value and density of any point in the three-dimensional space from the trained neural network;

[0076] The module for obtaining the three-dimensional light field model of the power tower is configured to extract the three-dimensional scene from the trained neural network using the marching cube algorithm to obtain the three-dimensional light field model of the power tower.

[0077] Example 3

[0078] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images are implemented.

[0079] Example 4

[0080] Based on the same inventive concept as other embodiments, this embodiment introduces a computer device, including: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images.

[0081] Example 5

[0082] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for constructing a three-dimensional light field model of a power tower using drone images.

[0083] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0085] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0087] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, which all fall within the protection of the present invention.

Claims

1. A method for constructing a three-dimensional light field model of a power tower using drone images, characterized in that: include: Acquire a power pole tower image, perform feature matching and motion recovery structure analysis on the power pole tower image, and obtain camera internal and external parameters, wherein the camera internal and external parameters include position and camera light direction; Sampling is performed along the direction of the camera light in three-dimensional space to obtain the three-dimensional coordinates of the corresponding sampling points , the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Perform position encoding to obtain the feature vector of the sampling point; Input the feature vector of the sampling point into the neural network to obtain the RGB color value and density of the sampling point, use volume rendering to obtain the rendered image from the original camera perspective, compare it with the real power tower image to calculate the loss function, perform neural network training, and obtain the RGB color value and density of any point in the three-dimensional space from the trained neural network; The marching cubes algorithm is used to extract the three-dimensional scene from the trained neural network to obtain the three-dimensional light field model of the power tower.

2. The method for constructing a three-dimensional light field model of a power tower using drone images according to claim 1, characterized in that: Sampling is performed along the direction of the camera light in three-dimensional space to obtain the three-dimensional coordinates of the corresponding sampling points The method is as follows: training the first network and the second network at the same time, using the layered sampling method to obtain N sampling points for each camera light and inputting them into the first network, and obtaining the RGB color value and density of each sampling point output by the first network; according to the density of each sampling point output by the first network, using the inverse transformation sampling method, newly generating M sampling points, inputting the newly generated M sampling points together with the previous N sampling points into the second network, and obtaining the RGB color value and density of each sampling point output by the second network; The sampling points input to the first network are less than the sampling points input to the second network, that is, N <M。 3. The method for constructing a three-dimensional light field model of a power tower using drone images according to claim 1, characterized in that: The three-dimensional coordinates of the sampling points and the unit vector of the camera light direction The specific method for position encoding and obtaining the feature vector of the sampling point is: The position encoding equation is: ； in, are the three-dimensional coordinates acting on the sampling points Each component and the unit vector of the camera light direction The three components of For the dimension, for Parameters used to represent the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction .

4. The method for constructing a three-dimensional light field model of a power tower using drone images according to claim 1, characterized in that: The architecture of the neural network is: the neural network includes a first fully connected layer and a second fully connected layer, is an even number; the neural network also includes a skip connection, which is used to convert the feature vector after position encoding Input to The first fully connected layer is activated; the feature vector after position encoding is Input to the first fully connected layer, and then pass through The feature vector after the first fully connected layer and the positional encoding after the jump connection After passing through the first fully connected layer, the output density , the output density Feature vector after adding view direction encoding Input to the second fully connected layer, and output RGB color value after passing through the second fully connected layer.

5. The method for constructing a three-dimensional light field model of a power tower using drone images according to claim 1, characterized in that: The loss function of the neural network is as follows: ； in, is the loss function, is the set of rays in each training batch in the step of layering rays on the power tower image, is every ray in the set, , , They are the RGB color values ​​of the light output in the power tower image, the first network, and the second network respectively.

6. The method for constructing a three-dimensional light field model of a power tower using drone images according to claim 1, characterized in that: The volume rendering formula is as follows: ； in, Indicates color, represents all the parameters in the neural network, , Representative is the starting point and the direction is of light; Represents the input light The density of the output of the neural network; Represents input light The color value output by the neural network; The unit vector representing the direction of the camera light; , Represent the far plane and the near plane respectively; is the position of the sampling point along the direction of the camera light, the minimum is , is the position of the sampling point of the light on the nearest plane; Indicates that the light arrive The transmittance within a distance, i.e. the probability that a ray of light can propagate without hitting any other particles, is defined as follows: 。 7. A device for constructing a three-dimensional light field model of a power tower using drone images, characterized in that: include: A camera internal and external parameter acquisition module is configured to acquire an image of a power pole tower, perform feature matching and motion recovery structure analysis on the image of the power pole tower, and obtain camera internal and external parameters, wherein the camera internal and external parameters include a position and a camera light direction; The feature vector extraction module is configured to sample along the direction of the camera light in three-dimensional space and obtain the three-dimensional coordinates of the corresponding sampling points. , and the three-dimensional coordinates of the sampling points and the unit vector of the camera light direction Perform position encoding to obtain the feature vector of the sampling point; The RGB color value and density acquisition module is configured to input the feature vector of the sampling point into the neural network, obtain the RGB color value and density of the sampling point, use volume rendering to obtain the rendered image of the original camera perspective, compare it with the real power pole tower image to calculate the loss function, perform neural network training, and obtain the RGB color value and density of any point in the three-dimensional space from the trained neural network; The module for obtaining the three-dimensional light field model of the power tower is configured to extract the three-dimensional scene from the trained neural network using the marching cube algorithm to obtain the three-dimensional light field model of the power tower.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a three-dimensional light field model of a power tower using drone images as described in any one of claims 1 to 6 are implemented.

9. A computer device, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the method for constructing a three-dimensional light field model of a power tower using drone images according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a three-dimensional light field model of a power tower using drone images described in any one of claims 1 to 6 are implemented.

Images