A method for improving the precision of unmanned aerial vehicle oblique photogrammetry

Abstract

This invention relates to the field of photogrammetry, specifically to a method for improving the accuracy of oblique photogrammetry using a UAV. The method includes: acquiring the UAV's hardware information, environmental data, and operational data; constructing a hyperparameter space and performing gridding processing; initializing the grid density based on the environmental data and hardware information to obtain the gridded hyperparameter space; performing iterative hyperparameter search within the gridded hyperparameter space; locally adjusting the grid density during the search process and calculating target coefficients at each iteration number based on the search process; thereby determining the optimal hyperparameter matrix using the target coefficients; and controlling the UAV using the optimal hyperparameter matrix to acquire and upload image data under the optimal hyperparameter matrix. This invention effectively improves the measurement accuracy of oblique photogrammetry using UAVs.

Description

Technical Field

[0001] This invention relates to the field of photogrammetry, and more specifically to a method for improving the accuracy of oblique photogrammetry using unmanned aerial vehicles (UAVs). Background Technology

[0002] Oblique photogrammetry is widely used in urban modeling, topographic mapping, and smart city applications because it can efficiently acquire multi-view images of ground features and construct high-precision 3D geographic models. However, in practical applications, traditional oblique photogrammetry methods are prone to problems such as incomplete image coverage, image blurring, and matching difficulties in complex environments due to the influence of terrain undulations, changes in weather conditions, and limitations of UAV hardware performance, leading to a decrease in the accuracy of 3D reconstruction models.

[0003] Existing technologies typically employ a fixed flight altitude and camera tilt angle shooting mode, lacking the ability to adaptively adjust to dynamic environmental changes and system hardware constraints, making it difficult to balance image quality and measurement accuracy in different scenarios. Furthermore, while some solutions introduce parameter optimization mechanisms, they often rely on global traversal searches or static preset parameter libraries, resulting in high computational overhead, poor real-time performance, and inefficient, blind parameter adjustments. Especially in variable field environments, how to combine UAV hardware configuration, real-time environmental conditions, and operating parameters to construct an environmentally adaptable dynamic parameter optimization mechanism has become a key issue in improving the accuracy of oblique photogrammetry. Summary of the Invention

[0004] This invention provides a method for improving the accuracy of oblique photogrammetry using unmanned aerial vehicles (UAVs) to address existing problems.

[0005] The present invention provides a method for improving the accuracy of oblique photogrammetry using unmanned aerial vehicles (UAVs), which employs the following technical solution:

[0006] One embodiment of the present invention provides a method for improving the accuracy of oblique photogrammetry measurements using a UAV, the method comprising the following steps:

[0007] Acquire hardware information, environmental data, and operational data of the drone;

[0008] A hyperparameter space is constructed and meshed. The mesh density is initialized based on environmental data and hardware information to obtain the meshed hyperparameter space.

[0009] Iterative hyperparameter search is performed in a gridded hyperparameter space. During the search process, the grid density is locally adjusted and the target coefficients are calculated for each iteration based on the search process. The optimal hyperparameter matrix is ​​then determined using the target coefficients.

[0010] The drone is controlled using the optimal hyperparameter matrix, thereby acquiring and uploading image data under the optimal hyperparameter matrix.

[0011] Optionally, the specific methods for constructing the hyperparameter space and performing meshing processing, and initializing the mesh density based on environmental data and hardware information to obtain the meshed hyperparameter space, include:

[0012] Construct a hyperparameter space and determine its size based on the UAV's environmental data and hardware information;

[0013] The hyperparameter space is meshed, and the mesh density is initialized using environmental data and hardware information.

[0014] Optionally, the specific methods for constructing the hyperparameter space and determining its size based on the UAV's environmental data and hardware information include:

[0015] Create a hyperparameter space, which corresponds to a spatial matrix. The elements of the spatial matrix include the flight altitude of the UAV and the tilt angles of the four tilt cameras. Each dimension of the hyperparameter space corresponds to an element in the spatial matrix.

[0016] The spatial matrix of the hyperparameter space is initialized based on the hardware information of the UAV.

[0017] Optionally, the specific methods for meshing the hyperparameter space and initializing the mesh density using environmental data and hardware information include:

[0018] The flight altitude step size and tilt angle step size are preset respectively. The corresponding dimensions in the hyperparameter space are divided at equal intervals by the flight altitude step size and tilt angle step size to form an initial grid point set. The initial grid point set contains several initial grid points, and each grid point corresponds to a matrix, denoted as the hyperparameter matrix. The dimension of the hyperparameter matrix is ​​the same as the dimension of the hyperparameter space.

[0019] The grid coefficient is calculated based on the environmental data and hardware information of the UAV. The spacing of each dimension in the hyperparameter space is dynamically adjusted based on the grid coefficient to obtain the gridded hyperparameter space at the current moment.

[0020] Optionally, the specific method for calculating the grid coefficient based on the UAV's environmental data and hardware information includes:

[0021] Acquire atmospheric pressure, wind speed, and wind direction data from environmental data, as well as the drone's battery power data, and perform standardization processing. Preset the prediction length, and use the ARIMA model to predict the atmospheric pressure, wind speed, and battery power data respectively, obtaining the predicted values ​​of atmospheric pressure, wind speed, and battery power data after the current time. Obtain several data points within the neighborhood radius centered on the current time, as the neighborhood data points for the current time. Calculate the grid coefficient for the current time based on the changes and fluctuations of the neighborhood data points corresponding to the atmospheric pressure, wind speed, and battery power data at the current time.

[0022] Optionally, the method for dynamically adjusting the spacing of each dimension in the hyperparameter space based on the grid coefficient to obtain the gridded hyperparameter space at the current moment includes:

[0023] Will As step size coefficients, these coefficients are multiplied by the flight altitude step size and the tilt angle step size, respectively, to obtain the gridded hyperparameter space at the current moment; where... Indicates the current time The grid coefficient below; This represents an exponential function with the natural constant as its base.

[0024] Optionally, the iterative hyperparameter search in the gridded hyperparameter space, with local adjustment of the grid density during the search process, includes the following specific methods:

[0025] Each hyperparameter combination in the gridded hyperparameter space is treated as a grid point, and these points are uniformly distributed throughout the gridded hyperparameter space. A longhorn beetle uses grid points as its position in each iteration within a gridded hyperparameter space; the step size of the beetle in each iteration is set to... There are grid points, among which The preset number of longhorn beetles, The preset step size parameter is used; when the beetle searches in the gridded hyperparameter space, it obtains the matrix formed by each hyperparameter at the corresponding grid point, which is denoted as the hyperparameter matrix at the corresponding iteration number.

[0026] In the digital simulation model, the hyperparameter matrix at the corresponding iteration number is input into the UAV's control module, thereby adjusting the parameters of the components corresponding to each element in the hyperparameter matrix. Based on the digital elevation model (DEM) and camera parameters, geometric projection simulation is performed to calculate the theoretical overlap, which is denoted as the theoretical overlap at the corresponding iteration number. The absolute value of the difference between the theoretical overlap and the baseline overlap is obtained, and the quality coefficient at the corresponding hyperparameter matrix is ​​calculated based on the absolute value of the difference. The absolute value of the difference is negatively correlated with the quality coefficient.

[0027] Based on the changes in the corresponding mass coefficients of each longhorn beetle during position iteration in the gridded hyperparameter space, the grid point density of the gridded hyperparameter space is locally adjusted.

[0028] Optionally, the method for locally adjusting the grid point density of the gridded hyperparameter space based on the changes in the corresponding mass coefficients of each longhorn beetle during position iteration in the gridded hyperparameter space includes:

[0029] Obtain the sequence of mass coefficients corresponding to all iterations before the current time for any beetle, and use it as the mass coefficient sequence of the beetle at the current time; calculate the search coefficient of the beetle at the corresponding iteration number based on the changing trend of the element values ​​in the mass coefficient sequence; preset the search radius, and use the search coefficient to multiply the flight altitude step and tilt angle step between grids within the search radius to obtain the local adjustment result of the gridded hyperparameter space.

[0030] Optionally, the specific method for calculating the target coefficients at each iteration number based on the search process, and then using the target coefficients to determine the optimal hyperparameter matrix, includes:

[0031] Obtain the mass coefficients of all longhorn beetles in the gridded hyperparameter space, and calculate the target coefficients for each iteration number by combining the position distribution of the longhorn beetles in the gridded hyperparameter space.

[0032] A two-dimensional rectangular coordinate system is constructed, with the iteration number as the horizontal axis and the target coefficient as the vertical axis. A scatter plot of the target coefficients corresponding to all iteration numbers is obtained in the two-dimensional rectangular coordinate system. The elbow method is used to determine the inflection point in the scatter plot, and the iteration number corresponding to the inflection point is recorded as the target iteration number. The hyperparameter matrix of the beetle corresponding to the maximum value among all search coefficients under the target iteration number is obtained as the optimal hyperparameter matrix.

[0033] Optionally, the specific method for controlling the UAV using the optimal hyperparameter matrix to acquire and upload image data under the optimal hyperparameter matrix includes:

[0034] The optimal hyperparameter matrix is ​​input into the flight control system of the UAV, thereby parsing the optimal hyperparameter matrix and extracting the various control parameters contained therein; the optimal hyperparameter matrix includes a flight altitude parameter and the tilt angle parameters of each of the four tilt cameras; the flight control system generates corresponding command signals based on the matrix and sends them to the altitude adjustment module of the UAV and the attitude adjustment module of each camera gimbal respectively;

[0035] Once the drone's flight status and the attitude of each camera meet the preset requirements, the main control unit issues a synchronous shooting command and uses the AES-256 encryption algorithm to encrypt and encapsulate the image file, which is then uploaded to the ground station server or cloud storage platform through the communication module on the drone.

[0036] The beneficial effects of the technical solution of this invention are as follows: Based on environmental data and the hardware capabilities of the UAV, the grid coefficients are calculated, the density of the gridded hyperparameter space is initialized, and environmentally adaptive grid partitioning is achieved. The beetle whisker algorithm is introduced for iterative searching in the dynamic grid space, and the optimal hyperparameters are efficiently located through local density adjustment and target coefficient optimization. The optimal parameters are fed back to the UAV control system in real time to drive high-precision image acquisition. This solves the problem of insufficient photographic accuracy caused by terrain changes; that is, the dynamic initialization of the grid driven by environmental data automatically thins the grid density in harsh environments, avoiding invalid calculations. By combining with the target coefficients, the overlap error of the output image data is controlled within a reasonable range, thereby improving the accuracy of 3D reconstruction and significantly enhancing the robustness and practicality of UAV oblique photogrammetry in complex terrain. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the steps of a method for improving the accuracy of oblique photogrammetry using an unmanned aerial vehicle (UAV) according to the present invention.

[0039] Figure 2 A schematic diagram of an oblique photography viewpoint of a drone provided in one embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of a two-dimensional hyperparameter space after meshing, as provided in one embodiment of the present invention.

[0041] Figure 4 This is a structural block diagram of a system for improving the accuracy of oblique photogrammetry using unmanned aerial vehicles (UAVs) according to the present invention. Detailed Implementation

[0042] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method for improving the accuracy of UAV oblique photogrammetry according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0044] The following description, in conjunction with the accompanying drawings, details a specific scheme for improving the accuracy of UAV oblique photogrammetry provided by the present invention.

[0045] Please see Figure 1 The diagram illustrates a flowchart of a method for improving the accuracy of UAV oblique photogrammetry according to an embodiment of the present invention, the method comprising the following steps:

[0046] Step S001: Obtain the drone's hardware information, environmental data, and operational data.

[0047] It should be noted that, due to the influence of terrain changes during the oblique photography process, there may be situations where the oblique photography image information is missing or the image quality is low. This may result in insufficient accuracy of the model after 3D geographic reconstruction using the relevant data and images from the oblique photography process. Therefore, in order to ensure the adaptability of oblique photography to terrain changes and improve the accuracy of oblique photography measurement, this embodiment of the invention selects to collect relevant data during the oblique photography process of the UAV to facilitate subsequent data analysis.

[0048] Specifically, in order to implement the method for improving the accuracy of UAV oblique photogrammetry proposed in this embodiment, it is first necessary to collect the UAV's hardware information, environmental data, and operational data. The specific process is as follows:

[0049] First, a drone is equipped with one orthophoto camera and four oblique cameras, and images are collected simultaneously from five different perspectives formed by one vertical and four oblique angles.

[0050] like Figure 2The diagram shows the oblique photography perspective of a drone. It illustrates the different camera perspectives of the drone as it flies in its flight direction. The orthophoto camera corresponds to the downward view, while the four oblique cameras form the forward, left, right, and rear views respectively through their respective tilt angles and positions.

[0051] Then, acquire the drone's hardware information and collect the drone's environmental and operational data in real time.

[0052] As an optional embodiment, the specific method for acquiring the hardware information of the UAV and collecting the environmental and operational data of the UAV in real time includes:

[0053] The hardware information includes: inertial measurement unit (IMU) parameters and the drone's battery level data.

[0054] The environmental data includes: atmospheric pressure data collected by a barometer, wind speed and wind direction data collected by an anemometer, and ambient light intensity data collected by a light intensity sensor.

[0055] The working data includes: real-time flight altitude data obtained through the GPS and IMU fusion positioning system, flight speed data obtained through the flight control system, flight attitude data (including pitch angle, roll angle and yaw angle) obtained through the attitude sensor, positioning accuracy data obtained through the GPS module, and gimbal angle data of each camera obtained through the gimbal control system.

[0056] Finally, environmental and operational data are preprocessed and standardized.

[0057] As an optional embodiment, the preprocessing of environmental data and work data includes the following specific methods: performing time synchronization processing on the collected environmental data and work data, using GPS time as the reference time, and precisely aligning the environmental data and work data according to the timestamp; and performing data cleaning on the environmental data and work data to remove outliers and noise data.

[0058] Thus, the hardware information, environmental data, and operational data of the drone have been obtained through the above methods.

[0059] Step S002: Construct the hyperparameter space and perform meshing. Initialize the mesh density based on environmental data and hardware information to obtain the meshed hyperparameter space.

[0060] It should be noted that, typically, to improve the accuracy of UAV oblique photography, adjustments are made during the UAV oblique photography process. Specific adjustment parameters usually include the UAV's flight altitude and the oblique angles corresponding to the four oblique cameras. In this embodiment of the invention, to facilitate subsequent adjustments to the UAV's flight altitude and the oblique camera angles, thereby further improving the accuracy of oblique photography measurements, a hyperparameter space formed by the flight altitude and oblique angles is constructed. This allows specific parameter values ​​to be obtained within the hyperparameter space during photography adjustments, avoiding the unrestricted range of hyperparameter values, which would hinder rapid and effective adjustments to the flight altitude and oblique angles and create significant computational burden.

[0061] Specifically, as a preferred embodiment, the steps of constructing the hyperparameter space and performing meshing processing, initializing the mesh density based on environmental data and hardware information to obtain the meshed hyperparameter space, include the following:

[0062] Step S201: Construct the hyperparameter space and determine its size based on the UAV's environmental data and hardware information.

[0063] It should be noted that, in order to ensure the reasonableness of the size of the hyperparameter space and avoid problems caused by the space being too large or too small when constructing the hyperparameter space, this embodiment of the invention chooses to use the current environmental conditions of the UAV and the limitations at the hardware level to initialize the constructed hyperparameter space and adjust the size of the hyperparameter space.

[0064] As a preferred embodiment, the specific method for constructing the hyperparameter space and determining its size based on the UAV's environmental data and hardware information is as follows:

[0065] First, a hyperparameter space is created, which corresponds to a spatial matrix. The elements in the spatial matrix include the UAV's flight altitude and the tilt angles of the four tilt cameras. Each dimension of the hyperparameter space corresponds to an element in the spatial matrix.

[0066] Then, the spatial matrix of the hyperparameter space is initialized based on the hardware information of the UAV.

[0067] As an optional embodiment, the method for initializing the spatial matrix of the hyperparameter space based on the hardware information of the UAV includes: obtaining the value range of the tilt angle corresponding to each tilt camera of the UAV, and recording the interval formed by the maximum and minimum values ​​of the tilt angle value range as the tilt interval of the tilt camera; presetting the maximum and minimum values ​​of the UAV's flight altitude, and recording the interval formed by the maximum and minimum values ​​of the flight altitude as the flight altitude interval of the UAV; and using the matrix formed by the flight altitude interval of the UAV and the tilt intervals of all tilt cameras as the spatial matrix corresponding to the initial hyperparameter space of the UAV.

[0068] It should be noted that, based on experience, the maximum and minimum flight altitude of the UAV are preset to be 200m and 100m, respectively. In addition, during oblique photography, the flight altitude of the UAV is not a fixed value, but is determined according to the specific mission requirements. Therefore, the flight altitude range can be adjusted according to the actual situation. This embodiment of the invention does not impose specific limitations.

[0069] Step S202: The hyperparameter space is meshed, and the mesh density is initialized using environmental data and hardware information.

[0070] It should be noted that, after determining the size of the hyperparameter space, this embodiment of the invention, in order to facilitate subsequent hyperparameter search and selection, combines a grid search method to perform gridding processing on the hyperparameter space, thereby discretizing the values ​​of hyperparameters in the hyperparameter space and improving search efficiency during subsequent hyperparameter value searches; for ease of intuitive understanding and representation, in such... Figure 3 The diagram shows a two-dimensional hyperparameter space after being meshed. In this diagram, the horizontal and vertical axes each correspond to one dimension. In the two-dimensional space of the horizontal and vertical axes, a mesh of size is formed. The hyperparameter space is divided into several grids by dashed lines in the figure. The intersections between the dashed lines in the grid are the combinations of hyperparameters formed in this two-dimensional space. In addition, considering that the density of the high-dimensional matrix or grid formed by the hyperparameter values ​​will affect the search efficiency and accuracy when the hyperparameter space is discretized, this embodiment of the invention further utilizes relevant information of the UAV to initialize the density or resolution of the hyperparameter values ​​in the hyperparameter space, thereby ensuring a balance between efficiency and accuracy.

[0071] As a preferred embodiment, the specific method for meshing the hyperparameter space and initializing the mesh density using environmental data and hardware information includes:

[0072] First, the flight altitude step size and tilt angle step size are preset respectively. The corresponding dimensions in the hyperparameter space are divided at equal intervals by the flight altitude step size and tilt angle step size to form an initial grid point set. The initial grid point set contains several initial grid points, and each grid point corresponds to a matrix, denoted as the hyperparameter matrix. The dimension of the hyperparameter matrix is ​​the same as the dimension of the hyperparameter space.

[0073] It should be noted that, based on experience, the preset flight altitude step size is 5m and the tilt angle step size is 3°, which can be adjusted according to the actual situation. This embodiment of the invention does not impose specific limitations.

[0074] Then, based on the environmental data and hardware information of the UAV, the grid coefficient is calculated, and the spacing of each dimension in the hyperparameter space is dynamically adjusted based on the grid coefficient to obtain the gridded hyperparameter space at the current moment.

[0075] As an optional embodiment, the specific method for calculating the grid coefficient based on the UAV's environmental data and hardware information includes: acquiring atmospheric pressure data, wind speed data, and wind direction data from the environmental data, as well as the UAV's battery power data, and performing standardization processing; presetting the prediction length, and using the ARIMA model to predict the atmospheric pressure data, wind speed data, and battery power data respectively, obtaining the predicted values ​​of the atmospheric pressure data, wind speed data, and battery power data after the current time; acquiring several data points within the neighborhood radius centered on the current time, as the neighborhood data points for the current time; and calculating the grid coefficient at the current time based on the changes and fluctuations of the neighborhood data points corresponding to the atmospheric pressure data, wind speed data, and battery power data at the current time.

[0076] In one specific embodiment, atmospheric pressure data, wind speed data, wind direction data, and drone battery power data from the acquired environmental data are standardized. The specific standardization method can be Z-Score standardization.

[0077] It should be noted that the prediction length is preset to 10s based on experience, but it can be adjusted according to the actual situation. This embodiment of the invention does not impose a specific limitation.

[0078] As an optional embodiment, the specific calculation method for the grid coefficient at the current moment is as follows:

[0079]

[0080] in, Indicates the current time The grid coefficient below; This indicates the current time in the atmospheric pressure data. The average value of all neighboring data points; This indicates the current time in the wind speed data. The average value of all neighboring data points; This indicates the current time in the wind speed data. The normalized standard deviation of all neighboring data points; This indicates the current time in the atmospheric pressure data. The normalized standard deviation of all neighboring data points; This indicates the current time in the battery data. The average value of all neighboring data points; This indicates the current time in the battery data. The neighborhood slope; Represents the absolute value function; This indicates the preset first parameter; This represents the logistic normalization function.

[0081] As an optional implementation, for the current moment in the wind speed data... The normalized standard deviation of all neighboring data points, and the atmospheric pressure data at the current time. The normalized standard deviation of all neighboring data points, wherein the normalized standard deviation is obtained by using the maximum-minimum normalization method to normalize the wind speed data at the current time. The standard deviation of all neighboring data points, and the current time in the atmospheric pressure data. The standard deviations of all neighboring data points are normalized to obtain the corresponding normalized standard deviations.

[0082] As an optional embodiment, the current time in the power data The method for obtaining the neighborhood slope is as follows: The slope of the current time interval in the power data is obtained using the least squares method. A straight line is fitted to all neighboring data points, and the slope corresponding to the fitted line is used as the current time value in the power data. The neighborhood slope.

[0083] It should be noted that the first parameter in the specific calculation method of the grid coefficient is preset to 0.1 based on experience to avoid the denominator being 0 in the formula. The specific value can be adjusted according to the actual situation, and this embodiment of the invention does not impose specific limitations. The grid coefficient, calculated based on the changes and fluctuations of the neighborhood data points corresponding to atmospheric pressure data, wind speed data, and power data at the current moment, reflects the degree of adjustment of the grid point density in the gridded hyperparameter space when searching for the UAV's flight altitude and camera tilt angle through the hyperparameter space, under the constraints of environmental complexity and hardware capabilities. Environmental data (air pressure fluctuations, strong winds) reflects the severity of the flight environment. The more severe the environment, the greater the physical error in the UAV maintaining attitude and executing commands. In this case, excessively high-precision parameter search is meaningless (execution will not be in place), and computational resources must prioritize ensuring flight control safety. Therefore, in harsh environments, increasing the search step size (reducing the grid density) can quickly obtain a "suboptimal solution" and save computational power; in favorable environments, decreasing the step size can pursue an "optimal solution"; when the environment is complex and the hardware capabilities are limited, the grid coefficient should be small and the grid density should be reduced accordingly to ensure computational efficiency; when the environment is complex and the hardware capabilities are sufficient, the grid coefficient should be large and the grid density should be increased accordingly to ensure search accuracy; in high-dimensional hyperparameter spaces, the reasonable setting of the grid density is crucial for avoiding the curse of dimensionality and ensuring search efficiency.

[0084] As an optional embodiment, the method of dynamically adjusting the spacing of each dimension in the hyperparameter space based on the grid coefficient to obtain the gridded hyperparameter space at the current moment includes: ... As step size coefficients, these coefficients are multiplied by the flight altitude step size and the tilt angle step size, respectively, to obtain the gridded hyperparameter space at the current moment; where... Indicates the current time The grid coefficient below; This represents an exponential function with the natural constant as its base.

[0085] It should be noted that the larger the value of the grid coefficient, the greater the density of grid points in the hyperparameter space should be. Therefore, in order to increase the density of grid points, the step size corresponding to the division of each dimension of the hyperparameter space into a grid at equal intervals should be reduced, thereby obtaining more combinations of hyperparameters and increasing the density of grid points. Conversely, when it is necessary to reduce the density of grid points, a larger grid coefficient should be used to adjust the step size to avoid the step size being too small.

[0086] Thus, the gridded hyperparameter space at the current moment is obtained using the above method.

[0087] Step S003: Perform iterative hyperparameter search in the gridded hyperparameter space. During the search process, the grid density is locally adjusted and the target coefficients are calculated for each iteration based on the search process. The optimal hyperparameter matrix is ​​then determined using the target coefficients.

[0088] It should be noted that after obtaining the hyperparameter space, in order to obtain the optimal values ​​corresponding to the UAV's flight altitude and the tilt angle of the tilt camera at the current moment, it is necessary to search within the hyperparameter space. Typically, all possible values ​​of hyperparameters in the hyperparameter space are traversed to determine whether the combination of hyperparameters meets the current requirements for photographic accuracy. However, traversing all possible values ​​of hyperparameters in the hyperparameter space will generate a huge amount of computation. Therefore, this embodiment of the invention chooses to use the beetle whisker search algorithm to search within the hyperparameter space. During the search process, in order to ensure the accuracy of the search results, the resolution of hyperparameter values ​​in local areas of the hyperparameter space is further improved.

[0089] Specifically, as a preferred embodiment, the iterative hyperparameter search in the gridded hyperparameter space, which involves locally adjusting the grid density during the search process and calculating the target coefficients at each iteration number based on the search process, thereby determining the optimal hyperparameter matrix using the target coefficients, includes the following specific steps:

[0090] Step S301: Use the beetle whisker search algorithm to search in the gridded hyperparameter space, and adjust the density of local regions in the gridded hyperparameter space during the search process.

[0091] It should be noted that since the hyperparameter space after gridding contains a combination of hyperparameters formed by the flight altitudes of several UAVs and the tilt angles of four tilt cameras, if the granularity of the grid is not fine enough during the search process in the hyperparameter space using the beetle whisker search algorithm, the search results will not be able to approach the optimal solution. However, if the granularity of the grid is too fine, it will increase the computational cost. Therefore, according to the search process of the beetle whisker search algorithm in the hyperparameter space, the granularity of the hyperparameter grid is locally adjusted to balance the computational cost and the accuracy of the search results.

[0092] As a preferred embodiment, the method of using the beetle whisker search algorithm to search in the gridded hyperparameter space, and adjusting the density of local regions in the gridded hyperparameter space during the search process, includes the following specific methods:

[0093] First, each combination of hyperparameters in the gridded hyperparameter space is treated as a grid point, and then uniformly distributed within the gridded hyperparameter space. A longhorn beetle uses grid points as its position in each iteration within a gridded hyperparameter space; the step size of the beetle in each iteration is set to... There are grid points, among which The preset number of longhorn beetles, The preset step size parameter is used; when the beetle searches in the gridded hyperparameter space, it obtains the matrix formed by each hyperparameter at the corresponding grid point, which is denoted as the hyperparameter matrix at the corresponding iteration number.

[0094] It should be noted that, based on experience, the preset number of longhorn beetles and the step length parameters are 10 and 1 respectively, which can be adjusted according to the actual situation. This embodiment of the invention does not impose specific limitations.

[0095] Then, in the digital simulation model, the hyperparameter matrix for the corresponding iteration number is input into the UAV's control module, thereby adjusting the parameters of the components corresponding to each element in the hyperparameter matrix. Based on the digital elevation model (DEM) and camera parameters, geometric projection simulation is performed to calculate the theoretical overlap, which is denoted as the theoretical overlap for the corresponding iteration number. The absolute value of the difference between the theoretical overlap and the baseline overlap is obtained, and the quality coefficient for the corresponding hyperparameter matrix is ​​calculated based on the absolute value of the difference. This absolute value of the difference is negatively correlated with the quality coefficient.

[0096] It should be noted that during the drone's flight, images are taken at different locations to determine the digital elevation model (DEM) for each location. However, since a high-precision DEM requires extensive calculations and analysis, it is usually built later by combining data collected by the drone using various technologies. A high-precision DEM cannot be generated during the drone's flight. Therefore, to estimate the theoretical overlap of the drone during flight, this embodiment of the invention uses geometric projection simulation based on the DEM and camera parameters to calculate the theoretical overlap. The DEM used in this process is based on the terrain undulation estimated from the coarse sparse point cloud or real-time ranging data from the airborne radar generated by the drone at the previous moment. The quality coefficient describes the degree of overlap required by the image data captured by the drone at the location represented by the corresponding hyperparameter matrix. A higher quality coefficient indicates that the image data captured by the drone at that location better meets the overlap requirements, and therefore, the higher the likelihood that the subsequent photographic data at that location will be used as the final image data to be uploaded.

[0097] Finally, based on the changes in the corresponding mass coefficients of each longhorn beetle during position iteration in the gridded hyperparameter space, the grid point density of the gridded hyperparameter space is locally adjusted.

[0098] It should be noted that when the longhorn beetle performs position iteration in the gridded hyperparameter space, there may be situations where the quality coefficient is high but there is still room for optimization. That is, the grid density of the local area where the grid points searched by the longhorn beetle are located may be further increased, so that the longhorn beetle can search for more specific locations in the local area. Therefore, this embodiment of the invention utilizes the change of the quality coefficient of the longhorn beetle during the position iteration process in the gridded hyperparameter space to adjust the grid density of the local area.

[0099] As an optional embodiment, the method for locally adjusting the grid point density of the gridded hyperparameter space based on the changes in the corresponding mass coefficients of each longhorn beetle during position iteration in the gridded hyperparameter space includes:

[0100] Obtain the sequence of mass coefficients corresponding to all iterations before the current time for any beetle, and use it as the mass coefficient sequence of the beetle at the current time; calculate the search coefficient of the beetle at the corresponding iteration number based on the changing trend of the element values ​​in the mass coefficient sequence; preset the search radius, and use the search coefficient to multiply the flight altitude step and tilt angle step between grids within the search radius to obtain the local adjustment result of the gridded hyperparameter space.

[0101] It should be noted that the search radius is preset to 5 based on experience, but it can be adjusted according to the actual situation. This embodiment of the invention does not impose any specific limitations.

[0102] As an optional embodiment, the specific calculation method for calculating the search coefficient of the longhorn beetle at the current moment based on the changing trend of the element values ​​in the quality coefficient sequence is as follows: ;

[0103] in, Indicates the current moment Next The search coefficient for each longhorn beetle; Indicates the current moment The slope of the fitted line of the quality coefficient sequence; Indicates the current moment Next The mass coefficients of the hyperparameter matrix corresponding to the number of iterations for each longhorn beetle; This represents the logistic normalization function.

[0104] It should be noted that the search coefficient is used to describe how close the beetle is to the optimal position at the corresponding iteration number. The larger the value of the search coefficient, the closer the beetle is to the optimal position, and the better the photographic data obtained by the UAV at that position can meet the requirements of photogrammetry accuracy.

[0105] Step S302: Based on the movement of all longhorn beetles and the changes in their corresponding quality coefficients during the search process of the longhorn beetle whisker search algorithm in the gridded hyperparameter space, calculate the target coefficients for each iteration number, and determine the optimal hyperparameter matrix using the target coefficients.

[0106] It should be noted that, in the process of searching in a hyperparameter grid with locally adjustable granularity using the beetle whisker search algorithm, in order to determine whether the searched hyperparameter combination can serve as the optimal solution for the UAV under the current conditions, this embodiment of the invention selects to analyze the positional distribution of beetles in multiple hyperparameter spaces and the search process of multiple iterations, thereby obtaining the target coefficients at the corresponding number of iterations, so as to reflect the possibility that the search results at the corresponding number of iterations have approximated the optimal solution to the greatest extent possible.

[0107] As a preferred embodiment, the method for calculating the target coefficients at each iteration number based on the movement of all longhorn beetles and the changes in their corresponding quality coefficients during the search process in the gridded hyperparameter space using the longhorn beetle whisker search algorithm, and determining the optimal hyperparameter matrix using the target coefficients, includes the following specific methods:

[0108] First, obtain the mass coefficients of all longhorn beetles in the gridded hyperparameter space, and calculate the target coefficients for each iteration number based on the position distribution of the longhorn beetles in the gridded hyperparameter space.

[0109] As an optional embodiment, the specific calculation method for the target coefficient is as follows:

[0110]

[0111] in, Indicates the first The target coefficient under a number of iterations; This represents the number of longhorn beetles in the gridded hyperparameter space; Represents the th in the gridded hyperparameter space At the time corresponding to the iteration number, the first... The search coefficient for each longhorn beetle; Represents the th hyperparameter in the gridded hyperparameter space At the time corresponding to the iteration number, the first iteration... The search coefficient for each longhorn beetle; Represents the th hyperparameter in the gridded hyperparameter space At the time corresponding to the nth iteration number, the nth iteration The first longhorn beetle and the first The Euclidean distance between the longhorn beetles; Represents a linear normalization function; Represents the maximum value function; express belong .

[0112] Since there are multiple longhorn beetles in the gridded hyperparameter space, and the extent to which they reach the optimal position varies during the search process, but the longhorn beetles continuously approach the optimal position during multiple iterations, the more clustered the longhorn beetles are in the gridded hyperparameter space and the higher the search coefficient level of the clustered region, the more likely that region is the optimal position, i.e., the hyperparameter matrix corresponding to that position is the optimal hyperparameter matrix. Therefore, the target coefficient is used to describe the probability of obtaining the optimal hyperparameter matrix at the corresponding number of iterations. The larger the value of the target coefficient, the greater the probability of obtaining the optimal hyperparameter matrix.

[0113] Then, a two-dimensional rectangular coordinate system is constructed, with the iteration number as the horizontal axis and the target coefficient as the vertical axis. A scatter plot of the target coefficients corresponding to all iteration numbers is obtained in the two-dimensional rectangular coordinate system. The elbow method is used to determine the inflection point in the scatter plot, and the iteration number corresponding to the inflection point is recorded as the target iteration number. The hyperparameter matrix of the beetle corresponding to the maximum value among all search coefficients under the target iteration number is obtained as the optimal hyperparameter matrix.

[0114] It should be noted that the quality coefficient sequence of the longhorn beetles as they travel in the gridded hyperparameter space reflects the image overlap matching degree, further calculating the search coefficients and dynamically adjusting the local grid density accordingly. Secondly, the positional distribution of all longhorn beetles and the search coefficients are fused to calculate the target coefficients. These coefficients, weighted by Euclidean distance, reflect the beetle aggregation degree and the potential to approach the optimal solution. Finally, the elbow method is used to automatically identify the inflection point (i.e., the optimal number of iterations) from the target coefficient curve and select the corresponding hyperparameter matrix as the global optimal solution. This achieves a dynamic balance between "accuracy and efficiency" in the search process; local density adjustment avoids computational redundancy caused by excessive global grid density while ensuring high-precision search in key areas. The combination of the target coefficients and the elbow method eliminates the need for manual intervention, significantly improving the algorithm's convergence speed and the reliability of the optimal solution, providing an efficient and adaptive decision-making basis for optimizing UAV oblique photogrammetry parameters.

[0115] Thus, the optimal hyperparameter matrix of the UAV is obtained through the above method.

[0116] Step S004: Use the optimal hyperparameter matrix to control the UAV, thereby acquiring and uploading image data under the optimal hyperparameter matrix.

[0117] It should be noted that the optimal hyperparameter matrix obtained through the above steps reflects the state in which the UAV can ensure the most suitable accuracy of oblique photogrammetry through multiple cameras under the current environment. Therefore, the optimal hyperparameter matrix is ​​directly used as the input of the UAV's control module to obtain the corresponding image data.

[0118] Specifically, firstly, the optimal hyperparameter matrix is ​​input into the UAV's flight control system, thereby parsing the optimal hyperparameter matrix and extracting the various control parameters contained therein; the optimal hyperparameter matrix includes a flight altitude parameter and the tilt angle parameters of each of the four tilt cameras; the flight control system generates corresponding command signals based on the matrix and sends them to the UAV's altitude adjustment module and the attitude adjustment modules of each camera gimbal, respectively.

[0119] Then, once the drone's flight status and the attitude of each camera meet the preset requirements, the main control unit issues a synchronous shooting command and uses the AES-256 encryption algorithm to encrypt and encapsulate the image files, which are then uploaded to the ground station server or cloud storage platform through the communication module on the drone.

[0120] This concludes the embodiment.

[0121] It should be noted that the embodiments used in this example The model is only used to represent negative correlations and the results of the constraint model output are in Within this range, in specific implementations, other models with the same purpose can be substituted; this embodiment is merely an example. The description will be based on a model, without making specific limitations on it. This refers to the input of the model.

[0122] Please see Figure 4 The diagram illustrates a structural block diagram of a system for improving the accuracy of oblique photogrammetry of a UAV according to an embodiment of the present invention. The system includes a memory 402, a processor 401, and a computer program 4021 stored in the memory 402 and executable on the processor. When the processor 401 executes the computer program 4021, it implements steps S001 to S004 of the method for improving the accuracy of oblique photogrammetry of a UAV.

[0123] Further, in an optional embodiment, the memory 402 described above may include read-only memory and random access memory, and provide instructions and data to the processor. Memory 402 may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory, wherein the non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may be random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Sync Link DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0124] The aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting Advanced Reduced Instruction Set Machines (ARM) architecture.

[0125] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for improving the precision of unmanned aerial vehicle photogrammetry, characterized in that, The method includes the following steps: Acquire hardware information, environmental data, and operational data of the drone; A hyperparameter space is constructed and meshed. The mesh density is initialized based on environmental data and hardware information to obtain the meshed hyperparameter space. Iterative hyperparameter search is performed in a gridded hyperparameter space. During the search process, the grid density is locally adjusted and the target coefficients are calculated for each iteration based on the search process. The optimal hyperparameter matrix is ​​then determined using the target coefficients. The drone is controlled using the optimal hyperparameter matrix, thereby acquiring and uploading image data under the optimal hyperparameter matrix. The iterative hyperparameter search in the gridded hyperparameter space, with local adjustment of the grid density during the search process, includes the following specific method: treating each hyperparameter combination in the gridded hyperparameter space as a grid point, and uniformly distributing the grid density within the gridded hyperparameter space. A longhorn beetle uses grid points as its position in each iteration within a gridded hyperparameter space; the step size of the beetle in each iteration is set to... There are grid points, among which The preset number of longhorn beetles, The preset step length parameter is used. When the beetle searches in the gridded hyperparameter space, it obtains the matrix formed by each hyperparameter at the corresponding grid point, which is denoted as the hyperparameter matrix under the corresponding iteration number. In the digital simulation model, the hyperparameter matrix under the corresponding iteration number is input into the UAV's control module, thereby adjusting the parameters of the components corresponding to each element in the hyperparameter matrix. Based on the digital elevation model (DEM) and camera parameters, geometric projection simulation is performed to calculate the theoretical overlap, which is denoted as the theoretical overlap under the corresponding iteration number. The absolute value of the difference between the theoretical overlap and the baseline overlap is obtained. The quality coefficient under the corresponding hyperparameter matrix is ​​calculated based on the absolute value of the difference. The absolute value of the difference is negatively correlated with the quality coefficient. According to the change of the corresponding quality coefficient during the position iteration of each beetle in the gridded hyperparameter space, the grid point density of the gridded hyperparameter space is locally adjusted.

2. The method for improving the accuracy of UAV oblique photogrammetry according to claim 1, characterized in that, The specific methods for constructing the hyperparameter space and performing meshing processing, initializing the mesh density based on environmental data and hardware information to obtain the meshed hyperparameter space, include: Construct a hyperparameter space and determine its size based on the UAV's environmental data and hardware information; The hyperparameter space is meshed, and the mesh density is initialized using environmental data and hardware information.

3. The method for improving the accuracy of UAV oblique photogrammetry according to claim 2, characterized in that, The specific methods for constructing the hyperparameter space and determining its size based on the UAV's environmental data and hardware information are as follows: Create a hyperparameter space, which corresponds to a spatial matrix. The elements of the spatial matrix include the flight altitude of the UAV and the tilt angles of the four tilt cameras. Each dimension of the hyperparameter space corresponds to an element in the spatial matrix. The spatial matrix of the hyperparameter space is initialized based on the hardware information of the UAV.

4. The method for improving the accuracy of UAV oblique photogrammetry according to claim 2, characterized in that, The specific methods for meshing the hyperparameter space and initializing the mesh density using environmental data and hardware information are as follows: The flight altitude step size and tilt angle step size are preset respectively. The corresponding dimensions in the hyperparameter space are divided at equal intervals by the flight altitude step size and tilt angle step size to form an initial grid point set. The initial grid point set contains several initial grid points, and each grid point corresponds to a matrix, denoted as the hyperparameter matrix. The dimension of the hyperparameter matrix is ​​the same as the dimension of the hyperparameter space. The grid coefficient is calculated based on the environmental data and hardware information of the UAV. The spacing of each dimension in the hyperparameter space is dynamically adjusted based on the grid coefficient to obtain the gridded hyperparameter space at the current moment.

5. The method for improving the accuracy of UAV oblique photogrammetry according to claim 4, characterized in that, The specific method for calculating the grid coefficient based on the UAV's environmental data and hardware information is as follows: Acquire atmospheric pressure, wind speed, and wind direction data from environmental data, as well as the drone's battery power data, and perform standardization processing. Preset the prediction length, and use the ARIMA model to predict the atmospheric pressure, wind speed, and battery power data respectively, obtaining the predicted values ​​of atmospheric pressure, wind speed, and battery power data after the current time. Obtain several data points within the neighborhood radius centered on the current time, as the neighborhood data points for the current time. Calculate the grid coefficient for the current time based on the changes and fluctuations of the neighborhood data points corresponding to the atmospheric pressure, wind speed, and battery power data at the current time.

6. The method for improving the accuracy of UAV oblique photogrammetry according to claim 4, characterized in that, The method for dynamically adjusting the spacing of each dimension in the hyperparameter space based on the grid coefficient to obtain the gridded hyperparameter space at the current moment includes the following specific methods: Will As step size coefficients, these coefficients are multiplied by the flight altitude step size and the tilt angle step size, respectively, to obtain the gridded hyperparameter space at the current moment; where... Indicates the current time The grid coefficient below; This represents an exponential function with the natural constant as its base.

7. The method for improving the accuracy of UAV oblique photogrammetry according to claim 6, characterized in that, The method for locally adjusting the grid point density in the gridded hyperparameter space based on the changes in the corresponding mass coefficients of each longhorn beetle during position iteration in the gridded hyperparameter space includes the following specific methods: Obtain the sequence of quality coefficients for any longhorn beetle at all iterations prior to the current moment, and use it as the quality coefficient sequence for the longhorn beetle at the current moment; calculate the search coefficient for the longhorn beetle at the corresponding iteration number based on the changing trend of the element values ​​in the quality coefficient sequence. A preset search radius is used, and the flight altitude step and tilt angle step between grids within the search radius are multiplied by the search coefficient to obtain the local adjustment result of the gridded hyperparameter space.

8. The method for improving the accuracy of UAV oblique photogrammetry according to claim 1, characterized in that, The specific method for calculating the target coefficients at each iteration number based on the search process, and then using the target coefficients to determine the optimal hyperparameter matrix, includes: Obtain the mass coefficients of all longhorn beetles in the gridded hyperparameter space, and calculate the target coefficients for each iteration number by combining the position distribution of the longhorn beetles in the gridded hyperparameter space. Construct a two-dimensional rectangular coordinate system, with the iteration number as the horizontal axis and the target coefficient as the vertical axis, and obtain a scatter plot of the target coefficient in the two-dimensional rectangular coordinate system for all iteration numbers; The elbow method is used to determine the inflection point in the scatter plot. The iteration number corresponding to the inflection point is recorded as the target iteration number. The hyperparameter matrix of the beetle corresponding to the maximum value among all search coefficients under the target iteration number is obtained as the optimal hyperparameter matrix.

9. The method for improving the accuracy of UAV oblique photogrammetry according to claim 1, characterized in that, The method for controlling the UAV using the optimal hyperparameter matrix to acquire and upload image data under the optimal hyperparameter matrix includes the following specific steps: The optimal hyperparameter matrix is ​​input into the flight control system of the UAV, thereby parsing the optimal hyperparameter matrix and extracting the various control parameters contained therein; the optimal hyperparameter matrix includes a flight altitude parameter and the tilt angle parameters of each of the four tilt cameras; the flight control system generates corresponding command signals based on the matrix and sends them to the altitude adjustment module of the UAV and the attitude adjustment module of each camera gimbal respectively; Once the drone's flight status and the attitude of each camera meet the preset requirements, the main control unit issues a synchronous shooting command and uses the AES-256 encryption algorithm to encrypt and encapsulate the image file, which is then uploaded to the ground station server or cloud storage platform through the communication module on the drone.

Images