A three-dimensional terrain reconstruction method and system based on unmanned aerial vehicle images

Abstract

This invention relates to the field of photogrammetry, specifically to a method and system for 3D terrain reconstruction based on UAV images. A 3D terrain reconstruction system based on UAV images includes: an information acquisition module, a terrain difference judgment module, a flight mission planning module, and a terrain reconstruction module. This invention utilizes a difference index-driven periodic adjustment mechanism and a neural network task parameter solving model to dynamically update key parameters such as flight altitude, overlap rate, speed, and exposure interval during flight. When the survey area experiences a sudden increase in wind speed, changes in illumination, or drastic terrain undulations, the system automatically shortens the refresh cycle and performs global or local adjustments, thereby promptly filling potential holes or blurry images and achieving high coverage and high-definition acquisition of complex terrain without manual intervention or the addition of expensive hardware.

Description

Technical Field

[0001] This invention relates to the field of photogrammetry, and more specifically to a method and system for three-dimensional terrain reconstruction based on UAV images. Background Technology

[0002] Unmanned aerial photogrammetry (UAV), relying on lightweight flight platforms, digital cameras, and post-processing software, has become the main method for acquiring high-resolution terrain data and orthophotos. The existing workflow typically involves flight path planning in one go before takeoff: based on factors such as desired resolution, average regional undulation, and conventional overlap, fixed flight altitude, track spacing, heading-lateral overlap rate, speed, and exposure interval are given. The UAV then performs shooting according to the predetermined parameters. After the mission is completed, image preprocessing, feature matching, structure-motion reconstruction (SfM), multi-view stereo matching (MVS), and orthophoto conversion are performed at the ground workstation. This "static planning-offline reconstruction" mode can meet the accuracy requirements in flat, weather-stable, small-scale scenes, but it shows significant shortcomings in complex terrain and dynamic environments.

[0003] First, environmental factors (such as wind speed, visibility, and illumination) often experience short-term, drastic fluctuations during flight. Lightweight drones have limited wind resistance; a sudden increase in wind speed can cause the drone's attitude to vibrate, leading to striped distortion. A sudden drop in visibility or rapid changes in illumination can easily cause underexposure, overexposure, or motion blur. Static flight paths cannot respond to these changes in real time, making it difficult to guarantee image quality. Second, large-area survey areas often include plains, hills or valleys with significant elevation differences, and forested areas with severe visibility obstruction. Uniform flight altitude and overlap settings cannot cover all sub-regions: at low altitudes, image holes appear in undulating areas; at high altitudes, redundant shooting occurs in flat areas, wasting energy and increasing the amount of data for post-processing. Furthermore, traditional parameter settings heavily rely on the experience of operators, making it difficult to maintain consistency between different teams or batches of tasks. In the face of sudden weather or complex terrain, real-time manual intervention is both time-consuming and prone to errors.

[0004] To address the aforementioned pain points, this invention proposes a method and system for three-dimensional terrain reconstruction based on UAV images. Summary of the Invention

[0005] This invention utilizes a differential index-driven periodic adjustment mechanism and a neural network task parameter solving model to dynamically update key parameters such as flight altitude, overlap rate, speed, and exposure interval during flight. When the survey area experiences a sudden increase in wind speed, changes in illumination, or drastic terrain undulations, the system automatically shortens the refresh cycle and performs global or local adjustments to promptly compensate for potential holes or blurred images, achieving high coverage and high-definition acquisition of complex terrain without the need for manual intervention or the addition of expensive hardware.

[0006] A method for 3D terrain reconstruction based on UAV images includes:

[0007] Obtain environmental factors and terrain features of the target area that can affect flight mission planning, and organize them into environmental factor vectors and terrain feature vectors respectively;

[0008] The terrain difference index of the target area is calculated based on the terrain feature vector. The index is compared with a preset difference threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene.

[0009] If the target area is a highly consistent scene, the trained task parameter solution model is used, taking the environmental factor vector and terrain feature vector as input, and outputting a flight mission planning parameter set, including flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval; then the UAV uses the flight mission planning parameter set to execute the flight mission; during the UAV's flight, the environmental factor vector is updated periodically at a preset base period, and the flight mission planning parameter set is locally adjusted based on the changes in the environmental factor vector, and then executed.

[0010] If the target area is a low-consistency scenario, the duration of the basic cycle is first adjusted according to the terrain difference index of the target area to obtain the adaptation update cycle. Then, the terrain feature vector and environmental factor vector of the UAV takeoff sub-area are input into the task parameter solving model to obtain the initial flight mission planning parameter set. Subsequently, the UAV uses the initial flight mission planning parameter set to execute the flight mission. During the flight of the UAV, the environmental factor vector is updated periodically according to the adaptation update cycle. At the same time, the terrain feature vector of the current sub-area of ​​the UAV is obtained. Based on the changes in the environmental factor vector and the current terrain feature vector, the initial flight mission planning parameter set is globally adjusted and then executed.

[0011] After image acquisition is completed, image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification are performed on all images in sequence, and finally the digital elevation model and orthophoto results are output.

[0012] Preferably, the terrain difference index of the target area is calculated based on the terrain feature vector, and the index is compared with a preset difference threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene. The specific steps include:

[0013] The grid side length is selected based on the expected ground resolution, so that the diagonal distance of the grid is no greater than half of the planned heading spacing; the target area is divided into several sub-regions; the lead images acquired before the flight mission are called up for each sub-region, and the following four types of terrain indicators that can be obtained based on the lead images are extracted: texture complexity index, edge density index, brightness variance index and parallax amplitude index.

[0014] Range normalization is applied to the above four types of terrain indicators respectively; then, they are linearly superimposed according to the preset weight coefficients to obtain a single scalar as the terrain complexity score of the sub-region.

[0015] The terrain complexity scores of all sub-regions are compiled into a sample set. First, the mean of the set is calculated. Then, the absolute value of each score deviating from the mean is calculated and averaged to obtain the average absolute deviation. At the same time, the interquartile range of the sample set is calculated, and the larger value between the average absolute deviation and the interquartile range is selected as the final terrain difference index.

[0016] Set a difference threshold. If the terrain difference index is not greater than the difference threshold, the target area is determined to be a high consistency scene; if the terrain difference index is greater than the difference threshold, it is determined to be a low consistency scene.

[0017] Preferably, the local adjustment operation for high consistency scenarios specifically includes:

[0018] At the end of each basic cycle, the latest environmental factor vector is collected and the difference is calculated with the environmental factor vector of the previous cycle to obtain the environmental change amount; if the change amount does not exceed the preset tolerance threshold, the current flight mission planning parameter set is kept unchanged and the flight mission continues to be executed.

[0019] When the environmental change is between the tolerance threshold and the preset warning threshold, only the flight speed and exposure interval are adjusted synchronously: when the wind speed increases or the visibility decreases, the flight speed is reduced proportionally and the exposure interval is shortened; when the wind speed decreases or the light intensity increases, the flight speed is restored proportionally and the exposure interval is extended. The adjustment range is limited to the preset maximum adjustment limit for a single cycle.

[0020] When environmental changes exceed the warning threshold, a safety strategy is triggered: pause filming or enter a hovering state while maintaining the current posture and waiting for the environment to recover.

[0021] Preferably, the duration of the base cycle is adjusted according to the terrain difference index of the target area to obtain an adaptive update cycle. The specific operation is as follows:

[0022] Set a base period and simultaneously set two fixed proportional boundary values: the first proportional boundary value is equal to 1, used for the case of minimum difference, and the second proportional boundary value is between 0 and 1, used for the case of maximum difference;

[0023] Read the calculated terrain difference index and use linear interpolation to map the terrain difference index to a periodic adjustment ratio k, so that k decreases monotonically with the terrain difference index and satisfies: when the terrain difference index is equal to 0, k is equal to 1 (i.e., the first ratio boundary value), and when the terrain difference index is equal to 1, k is equal to the second ratio boundary value.

[0024] Multiply the base cycle by the cycle adjustment ratio k to obtain the adaptation update cycle.

[0025] Preferably, the global adjustment operation for low consistency scenarios specifically includes:

[0026] At the end of each adaptation update cycle, the current environmental factor vector is first synchronously collected, and the terrain feature vector of the sub-region where the UAV is located is obtained. The environmental factor vector and the terrain feature vector are combined and used as input to the task parameter solution model to output a new flight mission planning parameter set. Then, the new flight mission planning parameter set is compared with the parameter set executed in the previous cycle. If the change of any parameter exceeds the preset global change safety limit, the parameter is interpolated and smoothed in a linear decreasing manner until the change is reduced to within the global change safety limit. Other parameters keep the model output value unchanged. In subsequent adaptation update cycles, the UAV continues to perform shooting tasks according to the new flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval.

[0027] Preferably, the task parameter solving model is based on a feedforward neural network and includes an input layer, several hidden layers, and an output layer.

[0028] A 3D terrain reconstruction system based on UAV imagery includes:

[0029] The information acquisition module is used to acquire environmental factors that can affect flight mission planning and terrain features of the target area, and organize them into environmental factor vectors and terrain feature vectors respectively.

[0030] The terrain difference judgment module is used to calculate the terrain difference index of the target area based on the terrain feature vector, compare the index with a preset difference threshold, and divide the target area into a high consistency scene or a low consistency scene based on the comparison result.

[0031] The flight mission planning module is used to address high-consistency scenarios in the target area. It utilizes a trained mission parameter solving model, taking environmental factor vectors and terrain feature vectors as input, and outputs a flight mission planning parameter set. The UAV then executes the flight mission using this parameter set. During UAV flight, the environmental factor vectors are updated periodically at a preset base period. Based on changes in the environmental factor vectors, the flight mission planning parameter set is locally adjusted and then executed. In low-consistency scenarios, the module first adjusts the base period duration based on the terrain difference index of the target area to obtain an adaptive update period. Then, the terrain feature vectors and environmental factor vectors of the UAV's takeoff sub-region are input into the mission parameter solving model to obtain an initial flight mission planning parameter set. The UAV then executes the flight mission using this initial parameter set. During UAV flight, the environmental factor vectors are updated periodically at an adaptive update period, and the terrain feature vectors of the UAV's current sub-region are obtained. Based on changes in the environmental factor vectors and the current terrain feature vectors, the initial flight mission planning parameter set is globally adjusted and then executed.

[0032] The terrain reconstruction module is used to perform image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification on all images after image acquisition, and finally output digital elevation model and orthophoto results.

[0033] The present invention has the following advantages:

[0034] 1. This invention uses a periodic adjustment mechanism driven by differential indicators and a neural network task parameter solving model to dynamically update key parameters such as flight altitude, overlap rate, speed and exposure interval during flight. When there is a sudden increase in wind speed, changes in illumination or drastic terrain undulations in the survey area, the system automatically shortens the refresh cycle and performs global or local adjustments to promptly fill potential holes or blurry images, achieving high coverage and high definition acquisition of complex terrain without manual intervention or the addition of expensive hardware.

[0035] 2. This invention maintains a relatively long basic update cycle for areas with low variability and relatively uniform terrain, and only performs limited fine-tuning when there are minor environmental fluctuations, avoiding unnecessary multiple retakes and parameter recalculations; at the same time, it calls the model frequently for global optimization only in areas with high variability, so that the amount of collected data matches the complexity of the scene; this strategy significantly reduces redundant images and point cloud scale, alleviates the storage and computing pressure of backend 3D reconstruction, and improves overall operation efficiency and drone endurance utilization. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the structure of a three-dimensional terrain reconstruction system based on UAV images used in an embodiment of the present invention. Detailed Implementation

[0037] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0038] Example 1: A method for 3D terrain reconstruction based on UAV images, comprising:

[0039] The environmental factors that can affect flight mission planning and the terrain features of the target area are obtained and organized into environmental factor vectors and terrain feature vectors, respectively. The environmental factor vectors include wind speed, visibility, light intensity, air pressure, air temperature and relative humidity; the terrain feature vectors include texture complexity, edge density, brightness variance and parallax amplitude.

[0040] The terrain difference index of the target area is calculated based on the terrain feature vector. The index is compared with a preset difference threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene.

[0041] If the target area is a highly consistent scene, the trained task parameter solution model is used, taking the environmental factor vector and terrain feature vector as input, and outputting a flight mission planning parameter set, including flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval; then the UAV uses the flight mission planning parameter set to execute the flight mission; during the UAV's flight, the environmental factor vector is updated periodically at a preset base period, and the flight mission planning parameter set is locally adjusted based on the changes in the environmental factor vector, and then executed.

[0042] If the target area is a low-consistency scenario, the duration of the basic cycle is first adjusted according to the terrain difference index of the target area to obtain the adaptation update cycle. Then, the terrain feature vector and environmental factor vector of the UAV takeoff sub-area are input into the task parameter solving model to obtain the initial flight mission planning parameter set. Subsequently, the UAV uses the initial flight mission planning parameter set to execute the flight mission. During the flight of the UAV, the environmental factor vector is updated periodically according to the adaptation update cycle. At the same time, the terrain feature vector of the current sub-area of ​​the UAV is obtained. Based on the changes in the environmental factor vector and the current terrain feature vector, the initial flight mission planning parameter set is globally adjusted and then executed.

[0043] After image acquisition is completed, image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification are performed on all images in sequence, and finally the digital elevation model and orthophoto results are output.

[0044] Based on the terrain feature vector, a terrain dissimilarity index for the target area is calculated. This index is compared with a preset dissimilarity threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene. The specific steps include:

[0045] The grid side length is selected based on the expected ground resolution, so that the diagonal distance of the grid is no greater than half of the planned heading spacing; the target area is divided into several sub-areas along the projection coordinate system of the survey area, and each sub-area is assigned a unique number and center coordinates to facilitate rapid positioning and data updates during flight.

[0046] For each sub-region, the following four types of terrain indicators, which can be obtained from the preliminary images acquired before the flight mission, are extracted:

[0047] Texture complexity index: Construct a gray-level co-occurrence matrix and take the average of the energy term and the entropy term to represent the surface texture details;

[0048] Edge density index: After extracting edges using edge operators, the ratio of the number of edge pixels to the total number of pixels is calculated.

[0049] Brightness variance index: The average of the squared deviations of gray values ​​relative to the mean of the sub-region is used to measure the magnitude of brightness variation;

[0050] Parallax magnitude index: Run the fast optical flow algorithm between two overlapping lead images, record the pixel displacement of the tracked feature points, and take the average absolute value of all pixel displacements as the parallax magnitude index.

[0051] Range normalization is performed on the above four types of terrain indicators respectively; then, they are linearly superimposed according to the preset weight coefficients to obtain a single scalar as the terrain complexity score of the sub-region; at the same time, the four-dimensional original indicator vector is saved for subsequent investigation and quality traceability; the weights are determined by historical samples or expert experience.

[0052] The terrain complexity scores of all sub-regions are combined into a sample set. First, the mean of the set is calculated. Then, the absolute value of each score deviating from the mean is calculated and averaged to obtain the average absolute deviation. At the same time, the interquartile range of the sample set is calculated, and the larger value between the average absolute deviation and the interquartile range is selected as the final difference index to suppress the influence of extreme values ​​on the judgment results.

[0053] A difference threshold is set. If the terrain difference index is not greater than the difference threshold, the target area is determined to be a high consistency scenario; if the terrain difference index is greater than the difference threshold, it is determined to be a low consistency scenario. This threshold can be adaptively adjusted by professionals in relevant fields based on historical statistical data or operational experience using a rolling average strategy to balance universality and judgment stability under multiple regional conditions.

[0054] Local adjustment operations for high consistency scenarios specifically include:

[0055] At the end of each basic cycle, the latest environmental factor vector is collected and the difference is calculated with the environmental factor vector of the previous cycle to obtain the environmental change amount. If the change amount does not exceed the preset tolerance threshold, the current flight mission planning parameter set is kept unchanged and the flight mission continues to be executed. The tolerance threshold can be given by historical statistical data of similar operations or expert experience, reflecting the acceptable range of small environmental fluctuations on image quality.

[0056] When environmental changes fall between the tolerance threshold and the preset warning threshold, only the flight speed and exposure interval are simultaneously fine-tuned: when wind speed increases or visibility decreases, the flight speed is reduced proportionally and the exposure interval is shortened; when wind speed decreases or light intensity increases, the flight speed is restored proportionally and the exposure interval is extended. The adjustment range is limited to the preset maximum adjustment limit per cycle, without changing the flight altitude, track spacing, or overlap rate. The maximum adjustment limit per cycle is determined and written into the parameter area by professionals in the relevant field during the system design phase, taking into account the dynamic response performance of the UAV platform, gimbal stability margin, and image clarity tolerance. When the limit value is high, the system allows for a larger range of speed and exposure adjustments within a single basic cycle, which can enhance the adaptability to rapidly deteriorating environments, but the track tracking error increases accordingly. When the limit value is low, the system only allows for smaller adjustments, which can maximize track stability and track geometry consistency, but may result in a short-term decrease in image clarity in sudden environmental changes.

[0057] When environmental changes exceed the warning threshold, a safety strategy is triggered: pause filming or enter a hovering state while maintaining the current posture and waiting for the environment to recover.

[0058] The duration of the base cycle is adjusted based on the terrain difference index of the target area to obtain an adaptive update cycle. The specific operation is as follows:

[0059] A base cycle is set, and two fixed proportional boundary values ​​are set simultaneously: the first proportional boundary value is equal to 1, which is used for the case of minimum difference, and the second proportional boundary value is between 0 and 1, which is used for the case of maximum difference; the second proportional boundary value is determined by professionals in the relevant field during the system design phase and remains unchanged during a single flight mission;

[0060] Read the calculated terrain difference index and use linear interpolation to map the terrain difference index to a periodic adjustment ratio k, so that k decreases monotonically with the terrain difference index and satisfies: when the terrain difference index is equal to 0, k is equal to 1 (i.e., the first ratio boundary value), and when the terrain difference index is equal to 1, k is equal to the second ratio boundary value.

[0061] Multiply the base cycle by the cycle adjustment ratio k to obtain the adaptation update cycle.

[0062] Global adjustment operations for low consistency scenarios specifically include:

[0063] At the end of each adaptation update cycle, the current environmental factor vector is first synchronously collected, and the terrain feature vector of the sub-region where the UAV is located is obtained. The environmental factor vector and the terrain feature vector are combined and used as input to the task parameter solution model to output a new flight mission planning parameter set. Then, the new flight mission planning parameter set is compared with the parameter set executed in the previous cycle. If the change of any parameter exceeds the preset global change safety limit, the parameter is interpolated and smoothed in a linear decreasing manner until the change is reduced to within the global change safety limit. Other parameters keep the model output value unchanged. In subsequent adaptation update cycles, the UAV continues to perform shooting tasks according to the new flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval.

[0064] The task parameter solving model is based on a feedforward neural network and includes an input layer, several hidden layers, and an output layer. The input layer simultaneously receives environmental factor vectors and terrain feature vectors, concatenating them into a unified feature sequence in a predetermined order. The first hidden layer performs a fully connected mapping on this feature sequence and completes the initial fusion of environmental and terrain information. At least one subsequent hidden layer continues to perform nonlinear mapping and feature abstraction to learn the complex correspondence between the flight mission planning parameter set and the input features. The output layer is a fully connected structure, with the number of nodes corresponding one-to-one with the number of parameters in the flight mission planning parameter set. It directly outputs flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval. The resulting output serves as the flight mission planning parameter set.

[0065] The specific steps for training the model for solving task parameters are as follows:

[0066] Obtain several model training samples with labeled flight mission planning parameter sets. Each model training sample contains environmental factor vectors and terrain feature vectors from a specific historical moment in the UAV's historical flight records. Divide all model training samples into training and validation sets. Use the training set to train the parameter-initialized task parameter solving model. Then, input the validation set into the task parameter solving model for validation and obtain the validation results. Set training conditions and determine whether the obtained validation results meet the training conditions. If yes, output the trained task parameter solving model; otherwise, continue training the task parameter solving model using the training set.

[0067] Example 2: A 3D terrain reconstruction system based on UAV images, such as... Figure 1 As shown, it includes:

[0068] The information acquisition module is used to acquire environmental factors that can affect flight mission planning and terrain features of the target area, and organize them into environmental factor vectors and terrain feature vectors respectively.

[0069] The terrain difference judgment module is used to calculate the terrain difference index of the target area based on the terrain feature vector, compare the index with a preset difference threshold, and divide the target area into a high consistency scene or a low consistency scene based on the comparison result.

[0070] The flight mission planning module is used to address high-consistency scenarios in the target area. It utilizes a trained mission parameter solving model, taking environmental factor vectors and terrain feature vectors as input, and outputs a flight mission planning parameter set. The UAV then executes the flight mission using this parameter set. During UAV flight, the environmental factor vectors are updated periodically at a preset base period. Based on changes in the environmental factor vectors, the flight mission planning parameter set is locally adjusted and then executed. In low-consistency scenarios, the module first adjusts the base period duration based on the terrain difference index of the target area to obtain an adaptive update period. Then, the terrain feature vectors and environmental factor vectors of the UAV's takeoff sub-region are input into the mission parameter solving model to obtain an initial flight mission planning parameter set. The UAV then executes the flight mission using this initial parameter set. During UAV flight, the environmental factor vectors are updated periodically at an adaptive update period, and the terrain feature vectors of the UAV's current sub-region are obtained. Based on changes in the environmental factor vectors and the current terrain feature vectors, the initial flight mission planning parameter set is globally adjusted and then executed.

[0071] The terrain reconstruction module is used to perform image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification on all images after image acquisition, and finally output digital elevation model and orthophoto results.

[0072] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for three-dimensional terrain reconstruction based on UAV images, characterized in that, include: Obtain environmental factors and terrain features of the target area that can affect flight mission planning, and organize them into environmental factor vectors and terrain feature vectors respectively; The terrain difference index of the target area is calculated based on the terrain feature vector. The index is compared with a preset difference threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene. If the target area is a highly consistent scene, the trained task parameter solution model is used, taking the environmental factor vector and terrain feature vector as input, and outputting a flight mission planning parameter set, including flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval; then the UAV uses the flight mission planning parameter set to execute the flight mission; during the UAV's flight, the environmental factor vector is updated periodically at a preset base period, and the flight mission planning parameter set is locally adjusted based on the changes in the environmental factor vector, and then executed. If the target area is a low-consistency scenario, the duration of the basic cycle is first adjusted according to the terrain difference index of the target area to obtain the adaptation update cycle. Then, the terrain feature vector and environmental factor vector of the UAV takeoff sub-area are input into the task parameter solving model to obtain the initial flight mission planning parameter set. Subsequently, the UAV uses the initial flight mission planning parameter set to execute the flight mission. During the flight of the UAV, the environmental factor vector is updated periodically according to the adaptation update cycle. At the same time, the terrain feature vector of the current sub-area of ​​the UAV is obtained. Based on the changes in the environmental factor vector and the current terrain feature vector, the initial flight mission planning parameter set is globally adjusted and then executed. After image acquisition is completed, image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification are performed on all images in sequence, and finally the digital elevation model and orthophoto results are output. Based on the terrain feature vector, a terrain dissimilarity index for the target area is calculated. This index is compared with a preset dissimilarity threshold. Based on the comparison result, the target area is divided into a high consistency scene or a low consistency scene. The specific steps include: The grid side length is selected based on the expected ground resolution, so that the diagonal distance of the grid is no greater than half of the planned heading spacing; the target area is divided into several sub-regions; the lead images acquired before the flight mission are called up for each sub-region, and the following four types of terrain indicators that can be obtained based on the lead images are extracted: texture complexity index, edge density index, brightness variance index and parallax amplitude index. Range normalization is applied to the above four types of terrain indicators respectively; then, they are linearly superimposed according to the preset weight coefficients to obtain a single scalar as the terrain complexity score of the sub-region. The terrain complexity scores of all sub-regions are compiled into a sample set. First, the mean of the set is calculated. Then, the absolute value of each score deviating from the mean is calculated and averaged to obtain the average absolute deviation. At the same time, the interquartile range of the sample set is calculated, and the larger value between the average absolute deviation and the interquartile range is selected as the final terrain difference index. Set a difference threshold. If the terrain difference index is not greater than the difference threshold, the target area is determined to be a high consistency scene; if the terrain difference index is greater than the difference threshold, it is determined to be a low consistency scene.

2. The method for three-dimensional terrain reconstruction based on UAV images according to claim 1, characterized in that, Local adjustment operations for high consistency scenarios specifically include: At the end of each basic cycle, the latest environmental factor vector is collected and the difference is calculated with the environmental factor vector of the previous cycle to obtain the environmental change amount; if the environmental change amount does not exceed the preset tolerance threshold, the current flight mission planning parameter set is kept unchanged and the flight mission continues to be executed. When the environmental change is between the tolerance threshold and the preset warning threshold, only the flight speed and exposure interval are adjusted synchronously: when the wind speed increases or the visibility decreases, the flight speed is reduced proportionally and the exposure interval is shortened; when the wind speed decreases or the light intensity increases, the flight speed is restored proportionally and the exposure interval is extended. The adjustment range is limited to the preset maximum adjustment limit for a single cycle. When environmental changes exceed the warning threshold, a safety strategy is triggered: pause filming or enter a hovering state while maintaining the current posture and waiting for the environment to recover.

3. The method for three-dimensional terrain reconstruction based on UAV images according to claim 2, characterized in that, The duration of the base cycle is adjusted based on the terrain difference index of the target area to obtain an adaptive update cycle. The specific operation is as follows: Set a base period and simultaneously set two fixed proportional boundary values: the first proportional boundary value is equal to 1, used for the case of minimum difference, and the second proportional boundary value is between 0 and 1, used for the case of maximum difference; Read the calculated terrain difference index and use linear interpolation to map the terrain difference index to a periodic adjustment ratio k, so that k decreases monotonically with the terrain difference index and satisfies: when the terrain difference index is equal to 0, k is equal to the first ratio boundary value, and when the terrain difference index is equal to 1, k is equal to the second ratio boundary value. Multiply the base cycle by the cycle adjustment ratio k to obtain the adaptation update cycle.

4. The method for three-dimensional terrain reconstruction based on UAV images according to claim 3, characterized in that, Global adjustment operations for low consistency scenarios specifically include: At the end of each adaptation update cycle, the current environmental factor vector is first synchronously collected, and the terrain feature vector of the sub-region where the UAV is located is obtained. The environmental factor vector and the terrain feature vector are combined and used as input to the task parameter solution model to output a new flight mission planning parameter set. Then, the new flight mission planning parameter set is compared with the parameter set executed in the previous cycle. If the change of any parameter exceeds the preset global change safety limit, the parameter is interpolated and smoothed in a linear decreasing manner until the change is reduced to within the global change safety limit. Other parameters keep the model output value unchanged. In subsequent adaptation update cycles, the UAV continues to perform shooting tasks according to the new flight altitude, track spacing, heading overlap rate, lateral overlap rate, flight speed, and exposure interval.

5. The method for three-dimensional terrain reconstruction based on UAV images according to claim 4, characterized in that, The task parameter solution model is based on a feedforward neural network and includes an input layer, several hidden layers, and an output layer.

6. A three-dimensional terrain reconstruction system based on UAV imagery, characterized in that, The system is applied to a three-dimensional terrain reconstruction method based on UAV images as described in any one of claims 1-5, comprising: The information acquisition module is used to acquire environmental factors that can affect flight mission planning and terrain features of the target area, and organize them into environmental factor vectors and terrain feature vectors respectively. The terrain difference judgment module is used to calculate the terrain difference index of the target area based on the terrain feature vector, compare the index with a preset difference threshold, and divide the target area into a high consistency scene or a low consistency scene based on the comparison result. The flight mission planning module is used to address high-consistency scenarios in the target area. It utilizes a trained mission parameter solving model, taking environmental factor vectors and terrain feature vectors as input, and outputs a flight mission planning parameter set. The UAV then executes the flight mission using this parameter set. During UAV flight, the environmental factor vectors are updated periodically at a preset base period. Based on changes in the environmental factor vectors, the flight mission planning parameter set is locally adjusted and then executed. In low-consistency scenarios, the module first adjusts the base period duration based on the terrain difference index of the target area to obtain an adaptive update period. Then, the terrain feature vectors and environmental factor vectors of the UAV's takeoff sub-region are input into the mission parameter solving model to obtain an initial flight mission planning parameter set. The UAV then executes the flight mission using this initial parameter set. During UAV flight, the environmental factor vectors are updated periodically at an adaptive update period, and the terrain feature vectors of the UAV's current sub-region are obtained. Based on changes in the environmental factor vectors and the current terrain feature vectors, the initial flight mission planning parameter set is globally adjusted and then executed. The terrain reconstruction module is used to perform image preprocessing, feature matching, sparse 3D reconstruction, dense matching, surface reconstruction and orthorectification on all images after image acquisition, and finally output digital elevation model and orthophoto results.

Images