Unmanned aerial vehicle parking apron dynamic landing point adjusting method and system based on environment perception

Abstract

The invention relates to the technical field of unmanned aerial vehicles, and discloses an unmanned aerial vehicle parking apron dynamic landing point adjustment method and system based on environment awareness, and the method comprises the steps: obtaining environment data and flight parameters, constructing a three-dimensional terrain grid model, and analyzing the stability; loading obstacle data and path constraints to generate a comprehensive evaluation model, and simulating to obtain regulation and control data; training the model to generate initial adjustment parameters, and combining the parameters with the model to obtain dynamic adjustment simulation information; constructing a multi-path optimization model to obtain optimal parameters; and combining real-time sensing to generate a safety probability, and adjusting a landing strategy. The system comprises a data acquisition module, a terrain modeling module and the like. According to the scheme, intelligent optimization and real-time adjustment of a landing point are realized through multi-dimensional environment perception and dynamic modeling, the landing safety and reliability of the unmanned aerial vehicle in a complex environment are improved, and the method is suitable for an intelligent landing scene of the unmanned aerial vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicles, and particularly to a method and system for dynamically adjusting the landing point of an unmanned aerial vehicle apron based on environmental perception. Background Art

[0002] With the wide application of unmanned aerial vehicle technology, the application of unmanned aerial vehicles in fields such as logistics distribution, power inspection, and emergency rescue is becoming increasingly popular. The safe and stable landing of an unmanned aerial vehicle is one of the key links to ensure the successful completion of its mission. As an important infrastructure for the landing of an unmanned aerial vehicle, the rationality and adaptability of the landing point directly affect the safety and reliability of the unmanned aerial vehicle landing.

[0003] In the traditional unmanned aerial vehicle landing process, the selection of the landing point is usually based on a fixed preset position, lacking the ability of real-time perception and dynamic adjustment of environmental changes. However, the environment where the unmanned aerial vehicle apron is located is often complex and changeable. For example, terrain undulations, changes in the distribution of obstacles, changes in weather conditions, etc. These factors will have an adverse impact on the landing of the unmanned aerial vehicle.

[0004] Specifically, terrain undulations may cause the landing surface to be uneven, increasing the impact and vibration when the unmanned aerial vehicle lands, and may even cause the unmanned aerial vehicle to overturn; the appearance of dynamic obstacles, such as suddenly flying birds, temporarily built facilities, etc., may block the landing path of the unmanned aerial vehicle and cause collision accidents; different weather conditions, such as strong winds, heavy rains, heavy snows, etc., will affect the flight state and control performance of the unmanned aerial vehicle, and thus affect the accuracy and safety of landing.

[0005] In addition, the flight state parameters of the unmanned aerial vehicle itself, such as flight speed, altitude, attitude, etc., also need to be comprehensively considered during the landing point adjustment process. The traditional fixed landing point strategy cannot be dynamically adjusted according to these real-time changing parameters, resulting in a low landing success rate of the unmanned aerial vehicle in a complex environment and posing a great safety hazard.

[0006] Although some existing unmanned aerial vehicle landing point adjustment methods can consider the influence of environmental factors to a certain extent, most of them have the following deficiencies: First, the environmental perception ability is limited, and it is impossible to realize multi-dimensional and real-time environmental data collection and analysis of the apron area; second, the dynamic assessment of terrain stability is lacking, and it is impossible to accurately judge the landing safety of different regions; third, the path planning and obstacle avoidance strategies are not flexible enough to cope with the complex and changeable distribution of obstacles and the change of the unmanned aerial vehicle flight state; fourth, the self-adaptability and intelligence level of the system are relatively low, and it is impossible to realize the dynamic optimization and real-time adjustment of the landing point.

[0007] Therefore, how to provide a landing point adjustment method and system that can perceive environmental changes in real time, dynamically evaluate terrain stability, flexibly plan landing paths, and make real-time adjustments according to the flight state parameters of the unmanned aerial vehicle has become a technical problem that needs to be solved urgently in the current field of unmanned aerial vehicles. Summary of the Invention

[0008] The purpose of the present invention is to provide a method and system for dynamically adjusting the landing point of an unmanned aerial vehicle apron based on environmental perception to solve the problems mentioned in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solutions: A method and system for dynamically adjusting the landing point of an unmanned aerial vehicle apron based on environmental perception, the method includes:

[0010] Obtain the environmental perception data of the unmanned aerial vehicle apron area and the flight state parameters of the unmanned aerial vehicle;

[0011] Perform multi-dimensional terrain scanning on the apron area and construct a three-dimensional terrain grid model, and conduct dynamic terrain stability analysis on the three-dimensional terrain grid model in combination with the environmental perception data to generate an initial terrain stability evaluation model;

[0012] Load the dynamic obstacle distribution data and the path constraint conditions of the unmanned aerial vehicle into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, and conduct dynamic simulation of the landing point in combination with the flight state parameters to obtain the landing point regulation simulation data;

[0013] Construct a dynamic obstacle prediction model and train it through the landing point regulation simulation data to obtain a landing point adjustment model, and then generate initial adjustment parameters, where the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path, and the initial adjustment response parameters;

[0014] Based on the initial adjustment parameters, the comprehensive terrain stability evaluation model, and the flight state parameters, obtain the dynamic adjustment simulation information;

[0015] Combine the dynamic adjustment simulation information with the apron design parameters to construct a multi-path optimization model and iteratively optimize the design parameters to obtain the optimal path planning parameters;

[0016] Infer the path planning parameters of the current period based on the real-time environmental perception model to generate the safe landing probability of the current period, and combine the optimal path planning parameters of the current period with the actual path parameters to adjust the real-time landing strategy of the unmanned aerial vehicle to achieve the goal of dynamic landing point matching.

[0017] Preferably, the landing point regulation simulation data at least includes the terrain undulation distribution, the obstacle density distribution, the safe landing area set, and the dynamic adjustment priority sequence;

[0018] The dynamically adjusted simulation information at least includes the landing point offset, the path obstacle avoidance success rate, and the stability evaluation index.

[0019] Preferably, the multi-dimensional terrain scanning of the apron area and the construction of a three-dimensional terrain grid model, combined with environmental perception data, are used to perform dynamic terrain stability analysis on the three-dimensional terrain grid model to generate an initial terrain stability evaluation model, including the following steps:

[0020] Perform lidar-vision fusion scanning on the apron area to obtain a stereoscopic terrain image sequence of the surface structure and obstacle distribution;

[0021] Preprocess the stereoscopic terrain image sequence to obtain standardized terrain grid data, where the preprocessing includes one or more of point cloud denoising, coordinate registration, feature extraction, grid subdivision, and data alignment;

[0022] Based on the standardized terrain grid data, combined with three-dimensional surface reconstruction technology, generate a three-dimensional terrain grid model;

[0023] Perform dynamic terrain stability analysis on the three-dimensional terrain grid model to generate an initial terrain stability evaluation model, where the dynamic terrain stability analysis at least includes grid weight assignment and stability parameter mapping, divides the terrain safety level through grid weight assignment, and assigns corresponding environmental perception data to each level area.

[0024] Preferably, the dynamic obstacle distribution data and the UAV path constraint conditions are loaded into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, and the landing point dynamic simulation is performed in combination with the flight state parameters to obtain the landing point regulation simulation data, including the following steps:

[0025] Load the dynamic obstacle distribution data into the initial terrain stability evaluation model to simulate real-time obstacle changes, and generate a first terrain stability evaluation model;

[0026] Load the UAV path constraint conditions into the first terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, where the UAV path constraint conditions include the maximum pitch angle limit, the minimum turning radius, and the power system redundancy parameters;

[0027] Based on the comprehensive terrain stability evaluation model and the flight state parameters, perform landing point dynamic simulation, and the specific process includes:

[0028] Construct a multi-constraint path planning equation, which at least includes an obstacle avoidance distance equation, a trajectory smoothness equation, and a power consumption equation. Combine the comprehensive terrain stability evaluation model and numerically solve the multi-constraint path planning equation by the piecewise linearization method to obtain the terrain undulation distribution, the obstacle density distribution, the set of safe landing areas, and the dynamic adjustment priority sequence;

[0029] The steps to obtain the set of safe landing areas include the following:

[0030] Based on the obstacle density distribution, calculate the obstacle coverage index of each grid cell;

[0031] Identify the areas in the comprehensive terrain stability evaluation model where the obstacle coverage index is not greater than the preset threshold, and generate the set of safe landing areas, which is expressed as follows:

[0032] Among them, the set of safe landing areas is the spatial distribution mapping result of the obstacle coverage index and the preset threshold.

[0033] Preferably, the steps to construct a dynamic obstacle prediction model, train it with the landing point regulation simulation data to obtain a landing point adjustment model, and then generate initial adjustment parameters include the following:

[0034] Construct a dynamic obstacle prediction model based on the three-dimensional terrain grid characteristics;

[0035] Train and verify the dynamic obstacle prediction model with the landing point regulation simulation data to obtain a landing point adjustment model;

[0036] Input the real-time flight state parameters into the landing point adjustment model to predict the set of safe landing areas;

[0037] Based on the predicted set of safe landing areas, generate initial adjustment parameters, where the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path, and the initial adjustment response parameters.

[0038] Preferably, it also includes data reconstruction of the landing point regulation simulation data, specifically:

[0039] Based on the terrain undulation distribution, the obstacle density distribution, and the set of safe landing areas, construct an initial multi-dimensional path input tensor;

[0040] Perform standardization and feature stratification processing on the initial multi-dimensional path input tensor to generate a final multi-dimensional path input tensor;

[0041] Based on the dynamic adjustment priority sequence, construct a landing point adjustment label tensor;

[0042] Combine the final multi-dimensional path input tensor and the landing point adjustment label tensor to form a training sample set.

[0043] Preferably, generating the initial adjustment parameters based on the predicted set of safe landing areas includes the following steps:

[0044] Extract the grid cell with the lowest obstacle density in the predicted set of safe landing areas to generate the initial landing point position;

[0045] According to the spatial connectivity of the predicted set of safe landing areas, fit the feasible topological structure of the initial obstacle avoidance path to generate the initial adjustment response parameters;

[0046] Calculate the terrain undulation gradient direction of the predicted set of safe landing areas and normalize it to the path reference vector, and the path reference vector is the initial obstacle avoidance path direction.

[0047] Preferably, obtaining the dynamic adjustment simulation information based on the initial adjustment parameters, the comprehensive terrain stability evaluation model, and the flight state parameters includes the following steps:

[0048] Map the initial landing point position to the comprehensive terrain stability evaluation model, match the initial obstacle avoidance path with the initial adjustment response parameters, encrypt the path nodes in the adjustment area, update the comprehensive terrain stability evaluation model, define the dynamic adjustment trigger condition, the path update rule, and the response parameter increment, and generate the dynamic adjustment simulation model;

[0049] Based on the dynamic adjustment simulation model, iteratively solve the multi-constraint path planning equation by the piecewise linearization method to obtain the dynamic adjustment simulation information, specifically including:

[0050] When the dynamic adjustment trigger condition is satisfied, update the landing point offset, the path obstacle avoidance success rate, and the stability evaluation index, and re-solve the multi-constraint path planning equation until the simulation termination condition is reached;

[0051] The dynamic adjustment trigger condition includes triggering the response parameter update when the current path obstacle avoidance success rate is not greater than the preset success rate threshold; the path update rule includes adjusting the initial obstacle avoidance path based on the terrain undulation gradient direction; the response parameter increment has a piecewise linear relationship with the current stability evaluation index.

[0052] Preferably, constructing the multi-path optimization model and iteratively optimizing the design parameters to obtain the optimal path planning parameters includes the following steps:

[0053] Construct a multi-path optimization model, where the optimization variables include the path node density, the turning angle threshold, and the power distribution ratio, the optimization objectives include minimizing the path length and maximizing the adjustment response speed, and the constraint conditions include the upper limit of the UAV power and the sensor accuracy threshold;

[0054] The multi-path optimization model is initially solved by the A* algorithm to generate an initial optimized path population;

[0055] Based on the initial optimized path population, the Dijkstra algorithm is used for global optimization to generate optimal path planning parameters.

[0056] Preferably, the present invention further includes a dynamic landing point adjustment system for an unmanned aerial vehicle apron based on environmental perception, and the system includes:

[0057] A data acquisition module for obtaining environmental perception data of the unmanned aerial vehicle apron area and unmanned aerial vehicle flight state parameters;

[0058] A terrain modeling and analysis module for performing multi-dimensional terrain scanning on the apron area and constructing a three-dimensional terrain grid model, and performing dynamic terrain stability analysis on the three-dimensional terrain grid model in combination with environmental perception data to generate an initial terrain stability evaluation model;

[0059] A comprehensive evaluation and simulation module for loading dynamic obstacle distribution data and unmanned aerial vehicle path constraint conditions into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, and performing dynamic simulation of the landing point in combination with flight state parameters to obtain landing point regulation simulation data;

[0060] A parameter generation module for constructing a dynamic obstacle prediction model and training it with the landing point regulation simulation data to obtain a landing point adjustment model, and further generating initial adjustment parameters, where the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path, and the initial adjustment response parameters;

[0061] A dynamic adjustment simulation module for obtaining dynamic adjustment simulation information based on the initial adjustment parameters, the comprehensive terrain stability evaluation model, and the flight state parameters;

[0062] A path optimization module for constructing a multi-path optimization model and iteratively optimizing the design parameters in combination with the dynamic adjustment simulation information and the apron design parameters to obtain optimal path planning parameters;

[0063] A strategy adjustment module for inferring the path planning parameters of the current period based on the real-time environmental perception model to generate the safe landing probability of the current period, and adjusting the real-time landing strategy of the unmanned aerial vehicle in combination with the optimal path planning parameters of the current period and the actual path parameters to achieve the goal of dynamic landing point matching.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] In terms of environmental perception and terrain modeling, a sequence of stereo terrain images is obtained through the fusion scanning of lidar and vision, and preprocessing such as point cloud denoising and coordinate registration is carried out. Combining with three-dimensional surface reconstruction technology, a high-precision three-dimensional terrain grid model is generated. At the same time, through the analysis of dynamic terrain stability, the classification of terrain safety levels and the assignment of environmental perception data are realized, which can accurately reflect the terrain characteristics and stability conditions of the apron area, providing a solid foundation for the selection of landing points. This multi-dimensional and high-precision terrain modeling method overcomes the problems of incomplete and inaccurate terrain information acquisition in traditional methods, can more realistically simulate the actual terrain environment, and improves the system's adaptability to complex terrains.

[0066] In terms of comprehensive evaluation and dynamic simulation, by loading dynamic obstacle distribution data and UAV path constraint conditions, a comprehensive terrain stability evaluation model is generated, and landing point dynamic simulation is carried out in combination with flight state parameters. The constructed multi-constraint path planning equation covers multiple aspects such as obstacle avoidance distance, trajectory smoothness, and power consumption. Through the piecewise linearization method for numerical solution, detailed landing point regulation simulation data such as terrain undulation distribution, obstacle density distribution, and safe landing area set can be obtained. This enables the system to comprehensively evaluate the feasibility and safety of different landing points considering various factors, predict potential problems in advance, and provide a scientific basis for the adjustment of landing points. Compared with traditional methods, this comprehensive evaluation and dynamic simulation method is more comprehensive and accurate, and can effectively reduce the landing risks caused by environmental factors and changes in UAV flight states.

[0067] In terms of parameter generation and dynamic adjustment, by constructing a dynamic obstacle prediction model and training it with landing point regulation simulation data, a landing point adjustment model is obtained, which can predict the set of safe landing areas according to real-time flight state parameters and generate initial adjustment parameters. At the same time, through the dynamic adjustment simulation model, when the trigger conditions are met, the dynamic adjustment simulation information such as landing point offset and path obstacle avoidance success rate is updated in real time, realizing the dynamic optimization of landing points and obstacle avoidance paths. This parameter generation and dynamic adjustment mechanism based on data-driven and model training endows the system with strong self-adaptability and intelligent level, can make timely adjustments according to the real-time changing environment and UAV state, and improves the flexibility and reliability of the landing process.

[0068] In terms of path optimization and strategy adjustment, the constructed multi-path optimization model aims to minimize the path length and maximize the adjustment response speed. Through iterative optimization using the A* algorithm and Dijkstra algorithm, the optimal path planning parameters are obtained. Meanwhile, based on the real-time environment perception model, the safe landing probability is generated, and combined with the optimal path planning parameters and actual path parameters, the real-time landing strategy is adjusted, realizing dynamic landing point matching. This multi-path optimization and strategy adjustment method can optimize the landing path of the UAV, improve the landing efficiency and accuracy, reduce power consumption, and extend the endurance time of the UAV while ensuring landing safety.

[0069] In addition, through the collaborative work of each module of the entire system, a complete closed-loop is formed from data collection, modeling analysis, simulation evaluation to parameter generation, path optimization, and strategy adjustment, realizing the full-process automation and intelligence of UAV landing point adjustment. This system can not only cope with complex and changeable environmental conditions and changes in UAV flight states, but also continuously improve its own performance through data training and model optimization, with strong scalability and sustainability. In summary, the present invention effectively solves the deficiencies of traditional UAV landing point adjustment methods in aspects such as environmental adaptability, evaluation accuracy, adjustment flexibility, and intelligence level, significantly improving the safety and reliability of UAV landing, and having broad application prospects and important technical value. Brief Description of the Drawings

[0070] Figure 1 is the working principle diagram of the dynamic landing point adjustment method for UAV apron based on environment perception described in the present invention;

[0071] Figure 2 is the flow chart for constructing the three-dimensional terrain grid model and the initial terrain stability evaluation model;

[0072] Figure 3 is the flow chart for training the dynamic obstacle prediction model and generating the initial adjustment parameters;

[0073] Figure 4 is the flow chart for constructing the dynamic adjustment simulation model and generating simulation information. Detailed Embodiments

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0075] Please refer to Figures 1 - 4, the dynamic landing point adjustment method and system for an unmanned aerial vehicle (UAV) apron based on environmental perception according to the present invention are specifically implemented as follows:

[0076] Through devices such as lidar and vision sensors deployed in the UAV apron area, environmental perception data of this area are collected in real time, including terrain features, obstacle distribution, meteorological parameters, etc.; meanwhile, flight state parameters of the UAV are obtained, such as flight altitude, speed, attitude angle, power system state, etc.

[0077] Perform multi-dimensional terrain scanning on the apron area, use lidar and vision fusion technology to obtain a sequence of stereo terrain images, and after preprocessing such as point cloud denoising and coordinate registration, use three-dimensional surface reconstruction technology to construct a three-dimensional terrain grid model. Combining the environmental perception data, through grid weight assignment and stability parameter mapping, perform dynamic terrain stability analysis on the grid model, divide the terrain safety level, and generate an initial terrain stability evaluation model.

[0078] Load dynamic obstacle distribution data and UAV path constraint conditions (such as maximum pitch angle limit, minimum turning radius, etc.) into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model. Based on this model and flight state parameters, construct a multi-constraint path planning equation (including equations such as obstacle avoidance distance, trajectory smoothness, power consumption, etc.), numerically solve it through the piecewise linearization method, perform dynamic simulation of the landing point, and obtain landing point regulation simulation data such as terrain undulation distribution, obstacle density distribution, safe landing area set, and dynamic adjustment priority sequence.

[0079] Construct a dynamic obstacle prediction model based on three-dimensional terrain grid features, train and verify the model using the landing point regulation simulation data to obtain a landing point adjustment model. Input the real-time flight state parameters into the model, predict the safe landing area set, and generate initial adjustment parameters including the initial landing point position, initial obstacle avoidance path, and initial adjustment response parameters.

[0080] Map the initial adjustment parameters to the comprehensive terrain stability evaluation model, generate a dynamic adjustment simulation model by encrypting path nodes, defining dynamic adjustment trigger conditions (such as the path obstacle avoidance success rate being lower than a preset threshold) and path update rules (such as adjusting the path based on the terrain undulation gradient). By iteratively solving the multi-constraint path planning equation, obtain dynamic adjustment simulation information such as landing point offset, path obstacle avoidance success rate, and stability evaluation index.

[0081] Combine the dynamic adjustment simulation information and apron design parameters to construct a multi-path optimization model with path node density, turning angle threshold, etc. as optimization variables and minimizing the path length and maximizing the adjustment response speed as objectives. Through iterative optimization of the A* algorithm and Dijkstra algorithm, obtain the optimal path planning parameters.

[0082] Based on the real-time environment perception model, the path planning parameters of the current period are inferred to generate the probability of safe landing. The optimal path planning parameters are combined with the actual path parameters to dynamically adjust the UAV landing strategy and achieve landing point matching.

[0083] The present invention will be further described below in conjunction with Examples 1 to 5:

[0084] Embodiment 1:

[0085] First, a composite perception system consisting of a laser radar and a visual sensor is used to synchronously scan the drone landing area. The laser radar emits lasers in the form of pulses, and obtains the target distance by measuring the time difference of the reflected signal, forming high-density point cloud data, which can accurately depict the terrain undulations and obstacle contours; the visual sensor collects texture and color information in the area. After the two are fused, a three-dimensional terrain image sequence containing surface structure, obstacle location and morphological characteristics is generated, providing multi-dimensional raw data for subsequent modeling.

[0086] After obtaining the stereo terrain image sequence, preprocessing is required to improve the data quality. The preprocessing process starts with point cloud denoising, using a statistical filtering algorithm to remove abnormal points: by calculating the distance distribution between each point and its k-neighboring points, setting a distance threshold (such as the mean plus 3 times the standard deviation), and removing points that exceed the threshold as noise, so that the point cloud data can more realistically reflect the original appearance of the terrain. Then coordinate registration is performed, and the iterative closest point (ICP) algorithm is used to unify the multi-sensor coordinate system: feature points with the same name are selected in the lidar point cloud and the visual image, and the translation matrix and rotation matrix are iteratively calculated to completely align the two sets of data in spatial position to ensure the spatial consistency of subsequent modeling.

[0087] The feature extraction stage uses the Harris corner detection algorithm to extract feature points with significant edges and corners by calculating the grayscale gradient matrix of the local area of ​​the image. These points correspond to key locations such as protruding parts in the terrain and vertices of obstacle contours, which are used for subsequent mesh subdivision and model construction to enhance the model's ability to express terrain details. Mesh subdivision encrypts the original mesh based on feature points, recursively divides the mesh using the Loop subdivision algorithm, inserts new vertices at the midpoints and vertices of the edges, and generates a denser mesh structure, which improves the model's accuracy in depicting tiny terrain undulations (such as small mounds and depressions) while keeping the mesh smooth and continuous.

[0088] Data alignment unifies the pre-processed terrain data into a global coordinate system through translation, rotation and scaling operations, and adjusts the scale to the actual physical unit (such as meters), ensuring seamless splicing of scanned data from different periods and regions to form standardized terrain grid data. Each grid cell contains unique coordinates (x, y, z) and attribute parameters such as altitude, slope, and roughness, laying the foundation for 3D modeling.

[0089] Based on standardized data, a three-dimensional terrain grid model is constructed using the moving least squares (MLS) method. This method fits a quadratic polynomial surface within the neighborhood of each grid cell, smoothly connects the height values of adjacent cells, and generates a continuous terrain surface. During the fitting process, the weights are automatically adjusted according to the density of feature points in the grid cell, increasing the fitting accuracy in complex terrain areas (such as around obstacles) and simplifying the calculation in flat areas to balance the model accuracy and computational efficiency. After the surface fitting, the triangulation technique is used to discretize the surface into triangular cells, and the vertex coordinates and normal vectors of each cell record the terrain geometry, facilitating computer processing and stability analysis.

[0090] The dynamic terrain stability analysis is achieved by assigning weights to grid cells. The weight factors include terrain slope, surface roughness, historical settlement data, etc. The terrain slope is calculated by the elevation difference and horizontal distance between adjacent grid cells. Areas with a slope ≤ 5° are assigned a high weight (0.8 - 1.0), indicating stability; 5° < slope ≤ 15° are assigned a medium weight (0.5 - 0.8); slopes > 15° are assigned a low weight (0.1 - 0.5). The surface roughness is measured by the change in point cloud density, with higher weights in smooth areas (such as cement floors) and lower weights in rough areas (such as gravel areas). The historical settlement data comes from geological reports, and the weight in the settlement risk area is reduced according to the settlement amplitude (for example, the weight in areas with an annual settlement > 10 mm is ≤ 0.3).

[0091] The terrain is divided into three levels according to the weights: safe area (weight 0.8 - 1.0), sub-safe area (0.5 - 0.8), and dangerous area (< 0.5). The terrain in the safe area is flat and stable, and it is preferred as the landing area; the sub-safe area needs to be used after combining real-time environmental assessment; landing is prohibited in the dangerous area. Environmental perception data is associated with each level area. For example, when the wind speed ≥ level 5, the weight of the safe area is reduced by 0.1 - 0.2; when the precipitation probability ≥ 60%, the weight of the sub-safe area is reduced by 0.1 to achieve dynamic update of the stability assessment.

[0092] In the model verification process, the actual measurement data is compared with the model output, and the root mean square error (RMSE) and mean absolute error (MAE) are calculated. If RMSE > 0.2 m or MAE > 0.15 m, the grid subdivision density (such as from 1 m spacing to 0.5 m) or the surface fitting parameters (such as increasing the polynomial degree) are adjusted, and the model is rebuilt until the error meets the standard. Through iterative optimization, it is ensured that the initial terrain stability assessment model accurately reflects the terrain state, providing a reliable basis for subsequent landing point selection.

[0093] During the whole process, technical details such as the synchronous triggering of lidar and vision sensors, data transmission rate (e.g., 10Gbps fiber optic transmission), and parallel computing for point cloud processing (accelerated by GPU) ensure the real-time performance and accuracy of modeling. Standardized data formats (such as storing point clouds in PLY format and mesh models in OBJ format) ensure smooth data interaction between modules and provide basic support for the dynamic adjustment of the UAV's landing point.

[0094] Embodiment 2:

[0095] In the process of generating the comprehensive evaluation model and dynamic simulation of the landing point, the specific implementation method is as follows: First, load the dynamic obstacle distribution data on the basis of the initial terrain stability evaluation model. The dynamic obstacle distribution data is obtained by real-time collection through millimeter-wave radar and vision sensors deployed around the UAV apron. The millimeter-wave radar has the characteristics of strong penetration and being unaffected by lighting conditions, and can accurately detect the distance, speed, and azimuth information of obstacles in bad weather such as rain and fog; the vision sensor can provide details such as the color, shape, and texture of obstacles. After time synchronization and spatial registration processing of the data of the two types of sensors, a dynamic data stream containing information such as the position, size, and movement trajectory of obstacles is generated.

[0096] Overlay the dynamic obstacle distribution data onto the three-dimensional terrain grid of the initial terrain stability evaluation model through a grid mapping algorithm. Specifically, first map the spatial coordinates of each obstacle to the corresponding grid cell, and calculate the occupied area or volume ratio within the grid cell according to the geometric shape of the obstacle (such as sphere, cube, etc.). For moving obstacles, predict their positions at multiple future moments according to their movement trajectories and map them to the corresponding grid cells in sequence, so as to simulate the dynamic change process of real-time obstacles in the three-dimensional terrain grid model and generate the first terrain stability evaluation model. This model clearly identifies the grid cells currently occupied by obstacles and the areas that may be affected in the future for a period of time, providing real-time obstacle constraint information for subsequent path planning.

[0097] Furthermore, load the UAV path constraint conditions into the first terrain stability assessment model to generate a comprehensive terrain stability assessment model. The UAV path constraint conditions mainly include the maximum pitch angle limit, the minimum turning radius, and the power system redundancy parameters, etc. The maximum pitch angle limit is determined according to the aerodynamic design and power performance of the UAV. For example, the maximum allowable pitch angle of a certain type of UAV is ±30°. Exceeding this angle may cause the UAV to stall or lose attitude control. The minimum turning radius is determined by the wheelbase and motor thrust of the UAV. Suppose the wheelbase of a certain UAV is 1.5 meters, and its minimum turning radius is set to 2 meters to ensure the flight stability of the UAV during turning. The power system redundancy parameter requires that the remaining power of the UAV must meet 1.5 times the energy consumption required to complete the current path planning to cope with the additional energy consumption caused by unexpected situations.

[0098] After converting the above path constraint conditions into mathematical expressions, conduct a quantitative assessment through the attribute parameters of the grid cells. For example, for each grid cell, calculate its pitch angle relative to the current position of the UAV. If it exceeds the maximum pitch angle limit, mark this grid cell as an impassable area. According to the connection relationship between grid cells, calculate the curvature radius when turning the path. If it is less than the minimum turning radius, adjust the position of the path inflection point to meet the constraint conditions. At the same time, combine the flight speed and path length of the UAV to calculate the energy consumption of each path segment and compare it with the remaining power to ensure that the power system redundancy parameter is met. Through the above processing, the comprehensive terrain stability assessment model organically combines terrain stability, obstacle distribution, and UAV physical limitations to form a complete feasible region constraint for path planning.

[0099] Based on the comprehensive terrain stability assessment model and UAV flight state parameters, carry out the landing point dynamic simulation work. First, construct a multi-constraint path planning equation, which includes multiple sub-equations such as an obstacle avoidance distance equation, a trajectory smoothness equation, and a power consumption equation. The obstacle avoidance distance equation requires that the horizontal distance between the UAV and obstacles during flight is not less than 1.2 meters, and the vertical distance is not less than 0.8 meters to avoid the risk of collision. By calculating the spatial distance between the UAV flight path and the grid cell where the obstacle is located, judge whether the obstacle avoidance requirement is met. If not, re-plan the path.

[0100] The trajectory smoothness equation constrains the degree of path curvature through the radius of curvature, avoiding sharp turns of the UAV during flight. Specifically, for each inflection point on the path, calculate the radius of curvature of its adjacent line segments. If the radius of curvature is less than a preset threshold (e.g., 5 meters), intermediate inflection points are added through interpolation to smooth the path curve and ensure the UAV can fly smoothly. The power consumption equation calculates the energy consumption of each path segment based on the flight resistance model of the UAV. The flight resistance model considers factors such as air density, the angle of attack of the UAV, and flight speed, determines the drag coefficient through empirical formulas or wind tunnel experiment data, and then calculates the energy consumption per unit distance, providing a basis for power optimization in path planning.

[0101] The piecewise linearization method is used to numerically solve the multi-constraint path planning equation. First, the continuous flight path is discretized into a piecewise linear path composed of multiple line segments, and each line segment corresponds to a movement between grid cells. By setting an appropriate discretization step size (e.g., 0.5 meters), the computational complexity is controlled while ensuring the calculation accuracy. For each discretized path segment, the obstacle avoidance distance equation, the trajectory smoothness equation, and the power consumption equation are solved in sequence to determine whether all constraint conditions are met. If not, the direction or length of the path segment is adjusted and the solution is redone until all constraint conditions are satisfied.

[0102] Through the above numerical solution process, landing point regulation simulation data such as terrain undulation distribution, obstacle density distribution, safe landing area set, and dynamically adjusted priority sequence are obtained. The terrain undulation distribution is generated from the elevation data of each grid cell and visually shows the height changes of the terrain in the form of a contour map or a 3D surface map to help identify complex terrain areas. The obstacle density distribution calculates the number of grid cells occupied by obstacles per unit area, and the higher the density value, the more dense the obstacles in that area and the greater the landing risk.

[0103] The generation process of the safe landing area set is as follows: First, calculate the obstacle coverage index of each grid cell, which is defined as the ratio of the area occupied by obstacles in the grid cell to the total area of the grid cell. By setting a preset threshold (e.g., 0.3), grid cells with an obstacle coverage index not greater than this threshold are identified as safe landing areas, and the spatial position information of these areas is integrated to generate the safe landing area set. This set is stored in the form of a list of grid cells, and each grid cell contains parameters such as coordinates, elevation, and stability weight, providing a clear range of feasible areas for the subsequent selection of landing points.

[0104] The dynamic adjustment of the priority sequence is determined according to factors such as the stability weight of grid cells and the obstacle density. First, the grid cells within the safe landing area are sorted in descending order of stability weight. For grid cells with the same weight, they are then sorted in ascending order of obstacle density, thus forming a dynamically adjusted priority sequence. Grid cells with higher priorities are preferentially used as candidate landing points to ensure that the UAV lands in areas with high stability and few obstacles, reducing the landing risk.

[0105] During the dynamic simulation of the landing point, it is also necessary to consider the real-time flight state parameters of the UAV, such as flight altitude, speed, attitude angle, etc. These parameters are obtained in real time through sensors such as the inertial measurement unit (IMU) and global positioning system (GPS) on the UAV and transmitted to the simulation system. The simulation system adjusts the initial conditions and boundary conditions of the path planning equation in real time according to these parameters to ensure that the simulation results can accurately reflect the actual flight state of the UAV, improving the reliability and practicality of the simulation data for landing point control.

[0106] To verify the effectiveness of the comprehensive terrain stability assessment model and the dynamic simulation method of the landing point, multiple simulation experiments and parameter adjustments are required. In the simulation experiments, different environmental conditions and obstacle distributions are simulated, the changing rules of the set of safe landing areas and the dynamically adjusted priority sequence are observed, and the solution efficiency and accuracy of the multi-constraint path planning equation are analyzed. By adjusting parameters such as the discretization step size and the constraint condition threshold, the simulation model is optimized to ensure that it can accurately and efficiently generate simulation data for landing point control in various complex scenarios, providing a reliable basis for subsequent landing point adjustment and path optimization.

[0107] Example 3:

[0108] In the process of constructing the landing point adjustment model and generating the initial adjustment parameters, the specific implementation method is as follows: First, a dynamic obstacle prediction model is constructed based on the geometric features and time series data of the three-dimensional terrain grid model. The geometric features of the three-dimensional terrain grid model include spatial dimension information such as the coordinates (x, y, z), slope, roughness, and historical obstacle positions of each grid cell, and the time series data covers time dimension information such as the obstacle movement trajectories and speed changes in the previous N moments. Considering the temporal dependence of the obstacle movement trajectories and the spatial correlation of the terrain features, a long short-term memory network (LSTM) is used as the model architecture. This network can effectively capture the long-term dependence relationships in the time series through the gating mechanism and is suitable for dealing with the prediction problem of dynamic obstacles.

[0109] The input layer of the dynamic obstacle prediction model is designed as a multi-dimensional feature vector, which includes the geometric features of the grid cell at the current moment (such as coordinates, slope, roughness) and the obstacle position coordinates, movement speed, direction angle, etc. of the previous N moments. For example, when N = 5, the input vector will include the grid features at the current moment and the obstacle state data of the past 5 moments, with a total of (3 + 3 + 1) × (5 + 1) = 42-dimensional features (assuming 3-dimensional grid coordinates, 1-dimensional slope, and 3-dimensional position, 1-dimensional speed for obstacle state). The input layer passes the feature vector to the hidden layer through a fully connected manner. The hidden layer contains multiple LSTM units, and each unit processes the input sequence layer by layer through the collaborative action of the forget gate, input gate, and output gate to extract the potential laws of obstacle movement and the influence mode of terrain features on it.

[0110] The training process of the model is carried out based on the landing point regulation simulation data. First, it is necessary to reconstruct the simulation data to construct input and output samples suitable for model training. The specific steps are as follows: Based on the terrain undulation distribution, obstacle density distribution, and safe landing area set, an initial multi-dimensional path input tensor is constructed. The dimension of this tensor is designed as [number of samples, time step, number of features], where the number of samples corresponds to the number of simulation scenarios, the time step corresponds to the time series length of obstacle movement in each scenario, and the number of features includes terrain features (such as grid coordinates, slope, roughness) and obstacle features (such as position, speed, density). For example, for a sample set containing 1000 simulation scenarios and each scenario records 20 time step data, the dimension of the initial input tensor is [1000, 20, 10] (assuming 6-dimensional terrain features and 4-dimensional obstacle features).

[0111] Standardization and feature stratification processing are performed on the initial multi-dimensional path input tensor. The standardization processing uses the Z-score standardization method to normalize the feature values of each dimension to a distribution with a mean of 0 and a standard deviation of 1 to eliminate the influence of different feature dimensions on model training. The feature stratification processing maps the terrain features and obstacle features to different channels respectively. For example, the first 6 channels store the terrain features, and the last 4 channels store the obstacle features, which is convenient for the model to learn the correlation between spatial features and movement features respectively. After processing, the final multi-dimensional path input tensor is generated, with its dimension remaining unchanged, but each feature value is in the standardized interval and the structure is clear.

[0112] Construct a landing point adjustment label tensor based on dynamically adjusting the priority sequence. Each element of the label tensor corresponds to the expected output of a sample, including information such as the recommended landing point coordinates (x, y, z), the obstacle avoidance path direction angles (θ, φ), etc. Among them, the recommended landing point coordinates are selected as the center coordinates of the grid cell with the lowest obstacle density from the set of safe landing areas, and the obstacle avoidance path direction angles are determined according to the terrain undulation gradient direction and the obstacle movement trend. The dimension of the label tensor is [number of samples, number of output features]. For example, when the output features include 3D coordinates and 2D direction angles, the dimension is [1000, 5].

[0113] Combine the final multi-dimensional path input tensor with the landing point adjustment label tensor to form a training sample set, and use the supervised learning method to train the dynamic obstacle prediction model. During the training process, the mean squared error (MSE) is used as the loss function, and the weight parameters of the model are optimized through the backpropagation algorithm. To avoid overfitting, a Dropout layer is added after the hidden layer to randomly discard a certain proportion (such as 20%) of the neuron connections to enhance the generalization ability of the model. The training process is divided into multiple epochs. In each epoch, the sample set is randomly shuffled and input into the model in batches of size 32 for training until the loss function converges to a preset threshold or reaches the maximum number of training rounds (such as 200 rounds).

[0114] After the model training is completed, it is verified through an independent validation sample set. During the verification process, the simulation data that has not participated in the training is input into the model, and indicators such as the overlap rate between the predicted set of safe landing areas and the actual safe area, and the error of the recommended landing point coordinates are calculated to evaluate the prediction accuracy of the model. If the verification result does not meet the requirements (such as the overlap rate is lower than 70%), then adjust the model architecture (such as increasing the number of LSTM layers, changing the number of hidden units) or optimize the training parameters (such as adjusting the learning rate, Dropout ratio), and retrain until the model performance meets the standard.

[0115] The dynamically trained and verified obstacle prediction model is the landing point adjustment model. Input the real-time flight state parameters of the UAV (such as the current position coordinates, flight speed, remaining battery power, attitude angles, etc.) and the latest environmental perception data (such as the real-time obstacle distribution, terrain stability weight) into this model. Through forward propagation calculation, the model outputs the prediction result of the set of safe landing areas in the future time period (such as the next 10 seconds). The prediction result is presented in the form of a list of grid cells, and each grid cell contains parameters such as the obstacle coverage index and the stability weight correction value at the prediction moment, reflecting the safety and applicability of the area in the future time period.

[0116] The process of generating the initial adjustment parameters based on the predicted set of safe landing areas is as follows: First, extract the grid cell with the lowest obstacle coverage index in the set, calculate the geometric center coordinates of this cell, and determine it as the initial landing point position. The obstacle coverage index is calculated as the ratio of the predicted area occupied by obstacles in the grid cell to the total area of the cell. The lower this value, the fewer obstacles there are in the area and the higher the landing safety. For example, in the predicted set of safe landing areas, the obstacle coverage index of a certain grid cell is 0.1, which is significantly lower than the average level, and its center coordinates (x0, y0, z0) are used as the initial landing point position.

[0117] Generate the initial obstacle avoidance path according to the spatial connectivity of the predicted set of safe landing areas. The spatial connectivity analysis is achieved through the adjacency matrix method in graph theory. Each safe grid cell is regarded as a node of the graph. If two nodes are adjacent in space (i.e., the grid cells share an edge or a face), an edge connection is established between the nodes. The minimum spanning tree algorithm (such as Kruskal's algorithm) is used to generate a tree structure that contains all the nodes in the connected graph. The set of edges of this tree structure constitutes the feasible topological structure of the initial obstacle avoidance path. The initial obstacle avoidance path consists of the key edges in the tree structure. By extracting the endpoint coordinates of each edge, a path sequence containing a series of inflection points is formed. For example, from the current position, pass through node A, node B, and node C in sequence to reach the initial landing point, and the coordinates of each inflection point constitute the key parameters of the initial obstacle avoidance path.

[0118] At the same time, calculate the terrain undulation gradient direction of the predicted set of safe landing areas to determine the initial obstacle avoidance path direction. The terrain undulation gradient is obtained by performing spatial difference calculations on the altitude data of the grid cells. Specifically, for each grid cell, calculate the height change rates (∂z / ∂x, ∂z / ∂y) in the x-axis and y-axis directions to form a gradient vector (gx, gy). The direction of this vector is the direction of the maximum terrain undulation. Normalize the gradient vector to a unit vector to obtain the path reference vector (gx_normalized, gy_normalized, 0), which is the initial obstacle avoidance path direction. For example, if the gradient vector is (2, -3), the normalized path reference vector is (2 / √13, -3 / √13, 0), indicating that the obstacle avoidance path should be preferentially planned along the downhill direction of the terrain to reduce flight resistance and energy consumption.

[0119] The generation of the initial adjustment response parameters combines the results of terrain stability assessment with the power state of the UAV. The results of terrain stability assessment are obtained through a comprehensive terrain stability assessment model. Each safety grid cell corresponds to a stability weight value (such as 0.8, 0.7, etc.). The higher the weight value, the more stable the terrain and the larger the allowable path adjustment step. The UAV power state parameters include remaining battery power, motor thrust output, etc. When the remaining battery power is sufficient, a larger adjustment step is allowed. Otherwise, the step needs to be reduced to save energy. The initial adjustment response parameters specifically include the path adjustment step (such as adjusting 0.5 meters per step), the speed correction coefficient (such as adjusting between 0.9 - 1.1 according to the stability weight), etc. These parameters are related to the stability weight and power state parameters through a linear mapping relationship. For example, the path adjustment step = base step × stability weight × power coefficient, where the power coefficient is determined according to the percentage of remaining battery power. When the remaining battery power is 80%, the power coefficient is 1.0, and for every 10% decrease, the coefficient decreases by 0.1.

[0120] During the entire implementation process, it is necessary to ensure the real-time and accuracy of data transmission. The UAV transmits real-time flight state parameters to the ground control system through a wireless communication module. The sensor network in the ground control system synchronously collects environmental perception data and performs fusion processing with simulation data, model parameters, etc. The data processing module adopts a distributed computing architecture and uses GPU acceleration technology to improve the efficiency of model inference and parameter calculation, ensuring that the generation of initial adjustment parameters can be completed within milliseconds to meet the real-time control requirements of the UAV. In addition, by establishing a data verification mechanism, the validity of real-time data input into the model is verified, and outliers and error data are eliminated to ensure the reliability of the model output results.

[0121] Example 4:

[0122] In the process of generating dynamic adjustment simulation information and path optimization, the specific implementation method is as follows: First, map the initial landing point position to the corresponding grid cell of the comprehensive terrain stability assessment model. The initial landing point position is determined by the previous steps. For example, it is the center coordinates of the grid cell with the lowest obstacle density selected from the set of safe landing areas. The position parameter is mapped to the coordinate system of the three-dimensional terrain grid model through a coordinate conversion algorithm to ensure its one-to-one correspondence with the grid cells in the model. Then, match the initial obstacle avoidance path with the initial adjustment response parameters. The initial obstacle avoidance path is generated by fitting the spatial connectivity of the safe landing area and contains a series of inflection point coordinates. Match these inflection point coordinates with the grid cell path in the comprehensive model to check whether the path is completely within the safe landing area and meets the UAV path constraint conditions (such as maximum pitch angle, minimum turning radius, etc.).

[0123] Perform path node encryption operation on the adjustment area to improve the simulation accuracy. The encryption rules are determined according to the terrain complexity and obstacle distribution density. For example, in areas with large terrain undulations or dense obstacles, the path node spacing is shortened from the original 1 meter to 0.5 meter, and new nodes are inserted between the original nodes through linear interpolation method to generate a denser path node sequence. After node encryption, update the path constraint conditions in the comprehensive terrain stability evaluation model, including parameters such as pitch angle, turning radius, and energy consumption at each node, to ensure that the model can more accurately reflect the flight state of the UAV on the encrypted path.

[0124] Define the dynamic adjustment trigger condition, path update rule, and response parameter increment to generate a dynamic adjustment simulation model. The dynamic adjustment trigger condition is set that the current path obstacle avoidance success rate is not greater than the preset success rate threshold, and the preset success rate threshold is determined according to the safety level requirements of the UAV, for example, set to 85%. When the path obstacle avoidance success rate is detected to be lower than this threshold during the simulation process, trigger the response parameter update mechanism. The path update rule is based on the terrain undulation gradient direction, specifically, the current obstacle avoidance path direction is adjusted to within ±15° of the terrain undulation gradient descent direction to utilize the terrain advantage to reduce flight resistance and obstacle influence. The response parameter increment has a piecewise linear relationship with the current stability evaluation index, and the stability evaluation index is output by the comprehensive terrain stability evaluation model, with a value range of 0 - 1. For example, when the index ≥ 0.7, the response parameter increment is 0.1 times the current value; when 0.4 < index < 0.7, the increment is 0.05 times the current value; when the index ≤ 0.4, the increment is 0. In this way, the dynamic coupling of the response parameter and terrain stability is achieved.

[0125] Based on the dynamic adjustment simulation model, iteratively solve the multi-constraint path planning equation through the piecewise linearization method. The multi-constraint path planning equation includes the obstacle avoidance distance equation, trajectory smoothness equation, power consumption equation, etc. The piecewise linearization method discretizes the continuous flight path into multiple linear segments, and each linear segment corresponds to a simulation step (such as 0.1 second). In each simulation step, first calculate the current path obstacle avoidance success rate, and the obstacle avoidance success rate is determined by statistically calculating the proportion of whether the spatial distance between the path segment and the obstacle grid cell meets the safety threshold (such as horizontal distance ≥ 1.2 meters, vertical distance ≥ 0.8 meters). If the success rate is lower than the preset threshold, adjust the obstacle avoidance path direction according to the path update rule. For example, perform a weighted average of the original path direction vector and the terrain gradient descent direction vector to obtain a new path direction vector, and update the path node coordinates. At the same time, calculate the response parameter increment according to the stability evaluation index, and adjust parameters such as the flight speed and acceleration of the UAV. For example, when the stability index is 0.6, the speed correction coefficient is 0.95, and multiply the current flight speed by this coefficient to reduce the speed and improve flight stability.

[0126] Re-solve the multi-constraint path planning equation and calculate whether the adjusted path segment meets all the constraint conditions. If it still does not meet the requirements, continue to adjust the path direction and response parameters until the obstacle avoidance success rate is higher than the preset threshold or the maximum number of iterations (such as 5 times) is reached. After each iteration, update the landing point offset, the path obstacle avoidance success rate, and the stability evaluation index. The landing point offset is centered on the initial landing point, and the maximum allowable translation distance within the safe landing area is 5 meters, which is represented by the coordinate offset (Δx, Δy, Δz); the stability evaluation index is calculated based on the average stability weight of the grid cells passed by the adjusted path, reflecting the overall stability level of the path.

[0127] The simulation termination condition is set to reach the preset maximum simulation duration (such as 30 seconds) or the landing point offset is less than 0.1 meter. At this time, it is considered that the path adjustment has tended to be stable, and the final dynamic adjustment simulation information is generated, including the landing point offset sequence, the change curve of the path obstacle avoidance success rate, the stability evaluation index time series, etc. This information is stored in a structured data format and can be called by the subsequent path optimization module.

[0128] In the path optimization section, first construct a multi-path optimization model. The optimization variables include the path node density, the turning angle threshold, and the power distribution ratio. The path node density is defined as the number of nodes per meter of the path, and the value range is 5 - 15. The higher the node density, the finer the path, but the higher the computational complexity; the turning angle threshold is the maximum allowable turning angle of the UAV at the path inflection point, and the value range is 30° - 90°. The larger the threshold, the higher the path flexibility, but the higher the requirement for the UAV's maneuverability; the power distribution ratio is the power output ratio of horizontal flight to vertical flight, and the value range is 0.3 - 0.7. This ratio affects the flight attitude and energy consumption of the UAV.

[0129] The optimization objectives include minimizing the path length and maximizing the adjustment response speed. Minimizing the path length calculates the straight-line distance between the path start point and the end point through the Euclidean distance formula, combined with the detour distance at the path inflection points, and the goal is to make the total path length the shortest; maximizing the adjustment response speed is measured by the number of path adjustments per unit time. The more adjustments, the more sensitive the system is to environmental changes. The constraint conditions include the UAV power upper limit (such as the maximum output power of the motor) and the sensor accuracy threshold (such as the GPS positioning accuracy of ±0.1 meter) to ensure that the optimized path is feasible within the physical capabilities of the UAV and the sensor accuracy range.

[0130] The A* algorithm is used to perform a preliminary solution on the multi-path optimization model to generate an initial optimized path population. The A* algorithm is a heuristic search algorithm that, by combining the actual cost from the current node to the starting point and the estimated cost to the end point (heuristic function), preferentially searches for paths with lower costs. During implementation, the three-dimensional terrain grid model is converted into a cost grid, and the cost of each grid cell is determined by factors such as terrain stability weight, obstacle density, and path node density. The grid cells with higher stability weights and lower obstacle densities have lower costs. The A* algorithm is used to search for the shortest path from the current position to the initial landing point in the cost grid, generating an initial optimized path population containing 10 candidate paths, and each path contains parameters such as node coordinates, turning angles, and power distribution ratios.

[0131] Based on the initial optimized path population, the Dijkstra algorithm is used for global optimization. The Dijkstra algorithm is a single-source shortest path algorithm suitable for finding the shortest paths from a starting point to all other nodes in a weighted graph. Each path in the initial optimized path population is used as the starting point to construct a weighted graph including dimensions such as path nodes, turning angles, and power distribution ratios, and the weight values are calculated according to the optimization objective function. For example, the path length weight is 0.6, and the weight for adjusting the response speed is 0.4. The Dijkstra algorithm is used to search for the globally optimal path in the weighted graph, comprehensively considering the path length and response speed, and generating the optimal path planning parameters. For example, the optimal parameters obtained after optimization are a path node density of 10 / m, a turning angle threshold of 60°, and a power distribution ratio of 0.5. This set of parameters achieves a balance between the path length and response speed, while meeting the constraints of the UAV power upper limit and sensor accuracy.

[0132] During the path optimization process, it is necessary to verify the feasibility of the optimization results. The optimal path planning parameters are substituted into the comprehensive terrain stability assessment model to check whether the path completely avoids dangerous areas and meets all UAV path constraint conditions, and a simulation flight is carried out by dynamically adjusting the simulation model to observe whether parameters such as the obstacle avoidance success rate and stability index of the UAV on the optimized path meet the safety requirements. If it is found that there are locally infeasible areas in the path (such as the pitch angle of a certain section of the path exceeding the maximum allowable value), it will return to the A* algorithm stage, adjust the heuristic function parameters or cost grid weights, and regenerate the candidate paths until the optimization results meet all constraint conditions and safety indicators.

[0133] During the entire implementation process, data processing and algorithm calculations are both carried out on a high-performance computing platform. Parallel computing technology is utilized to accelerate the solution processes of the A* algorithm and the Dijkstra algorithm, ensuring that path optimization can be completed within the limited time before the UAV lands. Meanwhile, a parameter storage and backtracking mechanism is established to save the intermediate results and final parameters of each optimization, so that it can quickly roll back to the previous feasible solution when the environmental conditions suddenly change, guaranteeing the flight safety of the UAV.

[0134] Example 5:

[0135] During the process of adjusting the real-time landing strategy and the coordination of system modules, the specific implementation method is as follows: The strategy adjustment module infers the path planning parameters for the current period based on the real-time environment perception model. The real-time environment perception model is constructed by fusing the latest lidar point cloud data, visual image data, and meteorological sensor data (such as wind speed, temperature, humidity), and can update parameters such as the obstacle distribution and terrain stability weight in the three-dimensional terrain grid model in real time. For example, when the lidar detects a new static obstacle in the apron area, the model will immediately map the position and size of the obstacle to the corresponding grid cells, update the obstacle coverage index, and recalculate the set of safe landing areas.

[0136] Generate the safe landing probability for the current period through the Bayesian network model. The nodes of the Bayesian network include environmental variables (wind speed, visibility, obstacle density), terrain variables (stability weight, slope), and landing safety status (safe / dangerous), and the edges between the nodes represent the probability dependence relationships between the variables. The calculation expression of the safe landing probability is:

[0137]

[0138] Where, represents the safe landing probability under the given environmental variables and terrain variables , is the probability distribution of environmental variables in the safe state, is the probability distribution of terrain variables in the safe state, is the prior probability of the safe state, and are the marginal probabilities of environmental variables and terrain variables respectively. This formula integrates multi-source data through Bayes' theorem to quantify the comprehensive impact of the environment and terrain on landing safety.

[0139] Adjust the real-time landing strategy of the UAV by combining the optimal path planning parameters in the current period with the actual path parameters. The optimal path planning parameters are output by the path optimization module and include path node coordinates, turning angle thresholds, power distribution ratios, etc.; the actual path parameters are obtained in real time through the inertial navigation system (INS) and global navigation satellite system (GNSS) carried by the UAV, including the actual position, speed, heading angle, etc. of the UAV. When the safe landing probability is lower than the preset threshold (such as 0.7), adjustment instructions are calculated through a proportional-integral-derivative (PID) controller. The input of the controller is the path deviation (the coordinate difference between the optimal path and the actual path), and the output is the correction amount of the landing point position , the adjusted value of the flight speed and the turning angle of the obstacle avoidance path .

[0140] The coordinated operation of each module of the system is scheduled through a distributed real-time operating system (RTOS). The data acquisition module consists of devices such as lidar, vision sensors, and IMUs, and continuously acquires environmental perception data and UAV flight state parameters at a frequency of 50 Hz, and transmits them to the terrain modeling and analysis module through an Ethernet bus. The terrain modeling and analysis module performs preprocessing such as point cloud denoising and coordinate registration on the received data, updates the three-dimensional terrain grid model and the initial terrain stability evaluation model every 1 second, and sends the updated model data to the comprehensive evaluation and simulation module.

[0141] Based on the latest model data, the comprehensive evaluation and simulation module loads the dynamic obstacle distribution and UAV path constraint conditions, generates a comprehensive terrain stability evaluation model, and performs a dynamic simulation of the landing point every 0.5 seconds, outputting landing point regulation simulation data such as terrain undulation distribution and safe landing area set, and transmitting it to the parameter generation module. The parameter generation module uses the landing point regulation simulation data to train the dynamic obstacle prediction model, updates the landing point adjustment model every 2 seconds, and generates initial adjustment parameters (initial landing point position, initial obstacle avoidance path, etc.) according to the real-time flight state parameters, and sends them to the dynamic adjustment simulation module.

[0142] The dynamic adjustment simulation module combines the initial adjustment parameters and the comprehensive terrain stability evaluation model to perform a dynamic adjustment simulation with a step size of 0.1 second, generates dynamic adjustment simulation information such as landing point offset and path obstacle avoidance success rate, and feeds it back to the path optimization module every 0.2 second. Based on the dynamic adjustment simulation information and the helipad design parameters, the path optimization module iteratively optimizes through the A* algorithm and Dijkstra algorithm, outputs the optimal path planning parameters every 1 second, and transmits them to the strategy adjustment module. The strategy adjustment module generates landing strategy adjustment instructions in real time according to the optimal path planning parameters, actual path parameters, and safe landing probability, and sends them to the UAV through a wireless communication link (such as 5G) at a command frequency of 10 Hz to ensure that the UAV can respond to environmental changes in a timely manner.

[0143] During the data transmission between modules, the TCP / IP protocol is adopted to ensure the reliability of data, and sensitive information (such as the location of the drone and path planning parameters) is encrypted and transmitted through a data encryption algorithm (such as AES-128) to prevent data leakage or malicious tampering. At the same time, a heartbeat detection mechanism is established, and each module regularly sends status information to the main controller. If the main controller does not receive the heartbeat signal of a certain module within the specified time, a fault alarm is triggered, and the system switches to the standby module to continue running, ensuring the high availability and fault tolerance of the system.

[0144] The entire real-time adjustment process forms a closed-loop control: from environmental data collection to model update, then to path planning and strategy adjustment, each link is closely connected and the timing is strictly synchronized, ensuring that the drone can continuously perceive the surrounding status in a complex dynamic environment, dynamically optimize the landing point and flight path, and finally achieve the goal of safe and accurate landing. Through modular design and standardized interfaces, the system supports the flexible expansion of sensor devices and algorithm models, and can adapt to the needs of different types of drones and diverse landing scenarios.

[0145] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0146] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamically adjusting the landing point of a drone based on environmental perception, characterized in that, It includes the following steps: Obtain the environmental perception data of the drone apron area and the drone flight state parameters; Conduct multi-dimensional terrain scanning on the apron area and construct a three-dimensional terrain grid model. Combine the environmental perception data to perform dynamic terrain stability analysis on the three-dimensional terrain grid model and generate an initial terrain stability evaluation model; Load the dynamic obstacle distribution data and the drone path constraint conditions into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model. Combine the flight state parameters to conduct a landing point dynamic simulation and obtain the landing point regulation simulation data; Construct a dynamic obstacle prediction model and train it with the landing point regulation simulation data to obtain a landing point adjustment model, and then generate initial adjustment parameters. Among them, the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path, and the initial adjustment response parameters; Based on the initial adjustment parameters, the comprehensive terrain stability evaluation model, and the flight state parameters, obtain the dynamic adjustment simulation information; Combine the dynamic adjustment simulation information with the apron design parameters, construct a multi-path optimization model and perform iterative optimization on the design parameters to obtain the optimal path planning parameters; Infer the path planning parameters of the current period based on the real-time environmental perception model, generate the safety landing probability of the current period, and combine the optimal path planning parameters and the actual path parameters of the current period to adjust the real-time landing strategy of the drone to achieve the dynamic landing point matching target.

2. The method for dynamically adjusting the landing point of a drone apron based on environmental perception according to claim 1, wherein The landing point regulation simulation data at least includes the terrain undulation distribution, the obstacle density distribution, the set of safe landing areas, and the dynamic adjustment priority sequence; The dynamic adjustment simulation information at least includes the landing point offset, the path obstacle avoidance success rate, and the stability evaluation index.

3. The method for dynamically adjusting the landing point of a drone landing pad based on environmental perception according to claim 1, characterized in that, The step of conducting multi-dimensional terrain scanning on the apron area and constructing a three-dimensional terrain grid model, combining the environmental perception data to perform dynamic terrain stability analysis on the three-dimensional terrain grid model, and generating an initial terrain stability evaluation model includes the following steps: Conduct lidar and vision fusion scanning on the apron area to obtain a sequence of three-dimensional terrain images of the surface structure and obstacle distribution; Preprocess the sequence of three-dimensional terrain images to obtain standardized terrain grid data. Among them, the preprocessing includes one or more of point cloud denoising, coordinate registration, feature extraction, grid subdivision, and data alignment; Based on the standardized terrain grid data, combine the three-dimensional surface reconstruction technology to generate a three-dimensional terrain grid model; Perform dynamic terrain stability analysis on the three-dimensional terrain grid model to generate an initial terrain stability evaluation model. Among them, the dynamic terrain stability analysis at least includes grid weight assignment and stability parameter mapping. Divide the terrain safety level through grid weight assignment and assign corresponding environmental perception data to each level area.

4. The method for dynamically adjusting the landing point of a drone landing pad based on environmental perception according to claim 1, characterized in that, The step of loading the dynamic obstacle distribution data and the drone path constraint conditions into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, combining the flight state parameters to conduct a landing point dynamic simulation, and obtaining the landing point regulation simulation data includes the following steps: Load the dynamic obstacle distribution data into the initial terrain stability evaluation model to simulate the real-time obstacle changes and generate a first terrain stability evaluation model; Load the UAV path constraint conditions into the first terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, where the UAV path constraint conditions include the maximum pitch angle limit, the minimum turning radius, and the power system redundancy parameters; Based on the comprehensive terrain stability evaluation model and the flight state parameters, conduct a dynamic simulation of the landing point. The specific process includes: Construct a multi-constraint path planning equation, which at least includes an obstacle avoidance distance equation, a trajectory smoothness equation, and a power consumption equation. Combine the comprehensive terrain stability evaluation model and numerically solve the multi-constraint path planning equation by the piecewise linearization method to obtain the terrain undulation distribution, the obstacle density distribution, the set of safe landing areas, and the dynamic adjustment priority sequence; The steps to obtain the set of safe landing areas include: Based on the obstacle density distribution, calculate the obstacle coverage index of each grid cell; Identify the areas in the comprehensive terrain stability evaluation model where the obstacle coverage index is not greater than the preset threshold, and generate a set of safe landing areas, which is expressed as follows: Among them, the set of safe landing areas is the spatial distribution mapping result of the obstacle coverage index and the preset threshold.

5. The method for dynamically adjusting the landing point of a drone apron based on environmental perception according to claim 1, wherein The steps of constructing a dynamic obstacle prediction model and training it with the landing point regulation simulation data to obtain a landing point adjustment model, and then generating initial adjustment parameters include: Construct a dynamic obstacle prediction model based on the three-dimensional terrain grid characteristics; Train and verify the dynamic obstacle prediction model with the landing point regulation simulation data to obtain a landing point adjustment model; Input the real-time flight state parameters into the landing point adjustment model to predict the set of safe landing areas; Based on the predicted set of safe landing areas, generate initial adjustment parameters, where the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path, and the initial adjustment response parameters.

6. The method for dynamically adjusting the landing point of a drone landing pad based on environmental perception according to claim 5, wherein It also includes data reconstruction of the landing point regulation simulation data, specifically: Based on the terrain undulation distribution, the obstacle density distribution, and the set of safe landing areas, construct an initial multi-dimensional path input tensor; Perform standardization and feature stratification processing on the initial multi-dimensional path input tensor to generate a final multi-dimensional path input tensor; Based on the dynamic adjustment priority sequence, construct a landing point adjustment label tensor; Combine the final multi-dimensional path input tensor and the landing point adjustment label tensor to form a training sample set.

7. The method for dynamically adjusting the landing point of a drone landing pad based on environmental perception according to claim 6, characterized in that, The steps of generating initial adjustment parameters based on the predicted set of safe landing areas include: Extract the grid cell with the lowest obstacle density in the predicted set of safe landing areas to generate the initial landing point position; According to the spatial connectivity of the predicted set of safe landing areas, fit the feasible topological structure of the initial obstacle avoidance path to generate the initial adjustment response parameters; Calculate the terrain undulation gradient direction of the predicted set of safe landing areas and normalize it to a path reference vector, and the path reference vector is the initial obstacle avoidance path direction.

8. The method for dynamically adjusting the landing point of a drone apron based on environmental perception according to claim 1, wherein The steps of obtaining dynamic adjustment simulation information based on the initial adjustment parameters, the comprehensive terrain stability evaluation model, and the flight state parameters include: Map the initial landing point position into the comprehensive terrain stability evaluation model, match the initial obstacle avoidance path and the initial adjustment response parameters, encrypt the path nodes in the adjustment area, update the comprehensive terrain stability evaluation model, define the dynamic adjustment trigger conditions, path update rules and response parameter increments, and generate a dynamic adjustment simulation model; Based on the dynamic adjustment simulation model, iteratively solve the multi-constraint path planning equation by the piecewise linearization method to obtain the dynamic adjustment simulation information, specifically including: When the dynamic adjustment trigger conditions are met, update the landing point offset, the path obstacle avoidance success rate and the stability evaluation index, and re-solve the multi-constraint path planning equation until the simulation termination conditions are reached; The dynamic adjustment trigger conditions include triggering the update of the response parameters when the current path obstacle avoidance success rate is not greater than the preset success rate threshold; the path update rules include adjusting the initial obstacle avoidance path based on the terrain undulation gradient direction; the response parameter increment has a piecewise linear relationship with the current stability evaluation index.

9. The method for dynamically adjusting the landing point of a drone apron based on environmental perception according to claim 1, wherein The steps of constructing a multi-path optimization model and iteratively optimizing the design parameters to obtain the optimal path planning parameters include: Construct a multi-path optimization model, where the optimization variables include the path node density, the turning angle threshold and the power distribution ratio, the optimization objectives include minimizing the path length and maximizing the adjustment response speed, and the constraint conditions include the upper limit of the UAV power and the sensor accuracy threshold; Perform a preliminary solution to the multi-path optimization model by the A* algorithm to generate an initial optimized path population; Based on the initial optimized path population, use the Dijkstra algorithm for global optimization to generate the optimal path planning parameters.

10. A dynamic landing point adjustment system for an unmanned aerial vehicle apron based on environmental perception, characterized in that, Including: A data acquisition module for obtaining the environmental perception data of the UAV apron area and the UAV flight state parameters; A terrain modeling and analysis module for performing multi-dimensional terrain scanning on the apron area and constructing a three-dimensional terrain grid model, and performing dynamic terrain stability analysis on the three-dimensional terrain grid model in combination with the environmental perception data to generate an initial terrain stability evaluation model; A comprehensive evaluation and simulation module for loading the dynamic obstacle distribution data and the UAV path constraint conditions into the initial terrain stability evaluation model to generate a comprehensive terrain stability evaluation model, and performing landing point dynamic simulation in combination with the flight state parameters to obtain the landing point regulation simulation data; A parameter generation module for constructing a dynamic obstacle prediction model and training it with the landing point regulation simulation data to obtain a landing point adjustment model, and then generating initial adjustment parameters, where the initial adjustment parameters at least include the initial landing point position, the initial obstacle avoidance path and the initial adjustment response parameters; A dynamic adjustment simulation module for obtaining dynamic adjustment simulation information based on the initial adjustment parameters, the comprehensive terrain stability evaluation model and the flight state parameters; A path optimization module for constructing a multi-path optimization model in combination with the dynamic adjustment simulation information and the apron design parameters and iteratively optimizing the design parameters to obtain the optimal path planning parameters; A strategy adjustment module, which is used to infer the path planning parameters of the current period based on the real-time environment perception model, generate the safe landing probability of the current period, and combine the optimal path planning parameters and the actual path parameters of the current period to adjust the real-time landing strategy of the drone to achieve the goal of dynamic landing point matching.

Images