Three-dimensional modeling and unmanned aerial vehicle landing area address selection method based on laser point cloud data

Abstract

The invention relates to the field of unmanned aerial vehicle automation, in particular to a three-dimensional modeling and unmanned aerial vehicle landing area address selection method based on laser point cloud data. The method comprises the steps of collection and dynamic compensation of laser radar point cloud data, multi-scale adaptive registration of three-dimensional point cloud, topographic feature extraction and evaluation, selection of an optimal landing point, path planning based on an improved A algorithm, and real-time environmental data updating and dynamic adjustment mechanism. According to the method, point cloud data is dynamically compensated through Kalman filtering, a 3D-NDT algorithm is utilized to perform high-precision three-dimensional environment modeling, key topographic features are extracted through an RANSAC algorithm to comprehensively evaluate topographic complexity and safety distance, an optimal landing point is selected, an optimal path of an unmanned aerial vehicle is planned in combination with an improved A algorithm, and path safety and flight energy efficiency are ensured. By updating environment data in real time and dynamically adjusting a flight path and a landing point, the unmanned aerial vehicle is adapted to sudden environment changes, and safe and stable autonomous landing of the unmanned aerial vehicle in a complex and changeable environment is ensured.

Description

Technical Field

[0001] The present invention relates to the field of UAV automation, and in particular to a three-dimensional modeling and UAV landing area selection method based on laser point cloud data, which is used for autonomous landing tasks of UAVs in low-visibility environments such as complex terrain and low-light conditions at night. Background Art

[0002] With the rapid development of drone technology, they are widely used in various tasks, including logistics transportation, environmental monitoring, etc. Autonomous flight and autonomous landing of drones are key technologies, especially at night or in complex terrain conditions, the safe landing of drones becomes particularly important.

[0003] In traditional drone landing systems, visual sensors are usually the main source of data to assist drones in environmental perception. However, visual systems are heavily dependent on external lighting conditions, which greatly reduces the effectiveness of visual sensors at night or in low light conditions. In addition, complex terrain (such as rugged mountainous areas or areas covered with dense vegetation) also poses challenges to visual sensors, making it difficult to ensure the safety and accuracy of autonomous landing.

[0004] The introduction of laser radar (LiDAR) technology provides a new solution for the environmental perception of UAVs. LiDAR generates high-precision three-dimensional point cloud data by emitting laser pulses and receiving reflected light, which enables it to work effectively under any lighting conditions. However, how to use LiDAR point cloud data to efficiently select the landing area for UAVs remains a complex technical problem. Existing LiDAR data processing methods have the disadvantages of large computational complexity and poor real-time performance, and traditional terrain feature extraction methods have limited processing capabilities for complex terrains and cannot meet the needs of UAVs for fast and safe landing.

[0005] In addition, during the autonomous flight of the drone, the dynamic environment (such as moving obstacles or wind speed changes) may affect the drone's path planning. Therefore, updating environmental data in real time and adjusting the flight path are also difficult problems that have not been fully solved in the existing technology.

[0006] Based on the above background, the present invention aims to provide an efficient and reliable method for UAV three-dimensional environment modeling and optimal landing point selection by utilizing lidar point cloud data, improved algorithms and dynamic environment update mechanism, which is particularly suitable for UAV autonomous landing missions at night or under complex terrain conditions. Summary of the invention

[0007] The present invention aims to solve the problem of autonomous landing of UAVs at night and in complex terrain conditions, and proposes a method for 3D environment modeling and optimal landing point selection based on LiDAR point cloud data. By integrating multi-sensor data (such as IMU and LiDAR), and combining improved point cloud registration algorithms, terrain feature extraction methods, and path planning algorithms, the present invention can perceive complex terrain environments in real time during the flight of the UAV, select the optimal landing area, and plan a safe autonomous landing path.

[0008] The present invention provides a systematic solution, which mainly includes the following aspects:

[0009] (1) Data acquisition: LiDAR is used to obtain point cloud data, and IMU sensors are used to obtain position information that cannot be obtained by humans or machines.

[0010] (2) Data preprocessing: The point cloud data is dynamically compensated by combining the IMU data with the Kalman filter to correct the data error caused by the UAV movement. The displacement and attitude data provided by the IMU can effectively reduce the motion distortion in the point cloud and ensure the temporal and spatial consistency and accuracy of the data.

[0011] (3) 3D point cloud registration: Based on the improved 3D normal distribution transform (3D-NDT) algorithm, multi-scale adaptive registration of point clouds is performed. This algorithm provides higher accuracy in complex terrains, while reducing the computational burden in simpler terrains, ensuring the real-time performance of the system. Through high-precision point cloud registration, an accurate 3D model of the drone's environment is established.

[0012] (4) Terrain feature extraction and evaluation: The RANSAC plane fitting algorithm is used to extract terrain features from point clouds and extract key terrain parameters, including slope, roughness, undulation, and safety distance. The robustness of the RANSAC algorithm can effectively handle noise and outliers in point cloud data, ensuring that stable terrain features are extracted under complex terrain conditions. Through the quantitative evaluation of these features, the possible landing area of ​​the drone is determined.

[0013] (5) Optimal landing point selection: This invention proposes a landing point selection mechanism based on multiple parameters through comprehensive evaluation of terrain complexity and safety distance. A comprehensive evaluation index is obtained by weighted calculation of terrain complexity (including slope, roughness, undulation) and safety distance. The area with the smallest comprehensive evaluation index is selected as the optimal landing point, thereby improving the safety and accuracy of UAV landing.

[0014] (6) Path planning and autonomous landing: An improved A* path planning algorithm is used to generate an optimal path for the drone from its current location to the best landing point. The algorithm combines obstacle risk assessment and flight energy optimization mechanisms to ensure that the drone can safely and efficiently plan the optimal path in complex environments. During flight, the algorithm can dynamically adjust the path based on the latest environmental data to ensure that the drone avoids obstacles and lands safely.

[0015] (7) Dynamic update and adjustment mechanism: During the flight of the drone, the lidar continuously scans the terrain, and the system continuously adjusts the optimal landing point and path planning by updating the point cloud data in real time. This mechanism can effectively respond to environmental changes, such as moving obstacles or sudden changes in wind speed, to ensure that the drone always flies on the optimal path and lands safely.

[0016] Technical effect of the present invention: The present invention proposes a dynamic and adaptive UAV autonomous landing method through in-depth processing of LiDAR point cloud data and comprehensive evaluation of various terrain features (slope, roughness, undulation and safety distance). This method is not only suitable for low-visibility environments at night, but also for various complex terrain conditions, such as mountainous areas, forests and other areas with dense obstacles. The present invention effectively solves the limitations of traditional UAV autonomous landing systems at night and in complex terrain conditions. By combining IMU and LiDAR data, a high-precision three-dimensional environment modeling and terrain feature extraction method is provided, which can accurately identify and select a safe landing point. The improved A* path planning algorithm ensures the path safety and energy efficiency optimization of UAVs in complex environments. The technical solution of the present invention is suitable for various types of UAV autonomous flight missions, especially in complex flight environments such as night, mountainous areas, forests, etc., and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 The following is a flowchart of the software and hardware system of the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the 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. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0020] The present invention proposes an autonomous landing system for UAVs that combines LiDAR point cloud data, inertial measurement unit (IMU) data, improved algorithms, and real-time dynamic adjustment. In complex and dynamic environments, the system can accurately perceive environmental information, select a safe landing point, and plan the best path. The following are the specific implementation steps of the present invention.

[0021] Step 1: Data collection: UAVs need to accurately perceive their surroundings during flight, especially in complex terrain, dense obstacles or low visibility. Environmental perception systems are essential. Traditional visual sensors rely on lighting conditions and are not effective at night or in low-light environments. LiDAR is a sensor that is very suitable for these scenarios and can provide high-resolution 3D point cloud data under any lighting conditions.

[0022] Point cloud data contains the three-dimensional coordinates of objects around the drone, which constitute the terrain model and the distribution of obstacles. However, since the drone moves during flight, these movements will affect the accuracy of the point cloud, especially the speed, acceleration and attitude changes of the drone, which will cause distortion or inaccuracy in the point cloud.

[0023] Therefore, the present invention introduces an IMU (inertial measurement unit) to assist in correcting the motion error of the drone. The IMU can provide real-time position information (such as acceleration, angular velocity, etc.) of the drone.

[0024] Step 2: Data preprocessing, combining the Kalman filter algorithm with the LiDAR point cloud data for dynamic compensation, thereby eliminating motion errors during flight. The following is the specific process of data processing and compensation:

[0025] Integration of IMU data: IMU provides the real-time motion status of the drone, including acceleration, angular velocity, attitude angle, etc. These data will be synchronized with the LiDAR data to obtain accurate motion information at every moment.

[0026] Kalman filter for state prediction: Kalman filter is a recursive estimation method that predicts the motion state of the drone and makes corrections based on actual observation data. The formulas for state prediction and covariance matrix prediction are as follows:

[0027]

[0028] P k|k-1 =F·P k-1|k-1 ·F T +Q+Q imu (2)

[0029] State update and correction: The second step of Kalman filtering is to correct the point cloud data through Kalman gain. Compared with the traditional Kalman gain, we dynamically adjust the Kalman gain by combining the displacement information provided by IMU and the point cloud data observed by LiDAR. The formula is as follows:

[0030] K k =P k|k-1 ·H T ·(H·P k|k-1 ·H T +R+R imu ) -1 (3)

[0031] Then the state is updated and the covariance matrix is ​​updated through the Kalman gain. Traditional Kalman filtering only relies on the sensor's observations for state updates and cannot handle the rapid attitude changes in dynamic systems such as drones. We introduce the rotation correction Δθ provided by the IMU imu Used to make additional corrections to the state, especially dynamic compensation for rotation angles. This can more accurately account for the actual movement of systems such as drones, especially when there are attitude changes. The update formula is as follows:

[0032]

[0033] P k|k =(IK k ·H)·P k|k-1 (5)

[0034] This dynamic compensation mechanism is particularly suitable for scenarios where the drone is flying at high speed or has large posture changes, such as obstacle avoidance or rapid navigation. The compensated point cloud data can provide an accurate description of the 3D environment and eliminate errors caused by motion. This provides a solid data foundation for subsequent terrain modeling, feature extraction, and path planning.

[0035] Step 3, point cloud registration is the process of aligning point cloud data collected at different times. During the flight of the drone, the lidar will continuously collect new point cloud data over time. In order to generate a complete 3D map, the system needs to accurately align the point cloud data collected at different times.

[0036] Traditional point cloud registration methods are difficult to handle both complex and simple terrains, because complex terrain requires high-precision processing, while using high precision in simple terrain will lead to unnecessary computational overhead. Therefore, the present invention proposes a multi-scale adaptive registration method based on 3D-NDT (normal distribution transform).

[0037] The 3D-NDT algorithm divides the point cloud data into multiple regions and performs Gaussian distribution modeling on each region. The expression is:

[0038]

[0039] In addition, the present invention adds an adaptive multi-scale mechanism to the system, and dynamically adjusts the registration scale by calculating the complexity parameter λi of each area. For complex terrain, the system will increase the resolution to ensure the accuracy of the point cloud data; for relatively flat areas, the system will reduce the resolution to reduce the amount of calculation. The adaptively adjusted covariance matrix is:

[0040]

[0041] This mechanism ensures that the system can work efficiently under different terrain conditions. The registration process is completed through iteration. The system calculates the error between the source point cloud and the target point cloud in each iteration and adjusts the pose of the point cloud according to the error until the error meets the preset convergence condition.

[0042] Finally, the point cloud registration is completed through iteration. In each iteration, the system first calculates the error between the target point cloud and the source point cloud, and then adjusts the pose of the source point cloud according to the error value, so that the error between the two is gradually reduced. Through continuous iterative adjustment, the error eventually meets the preset convergence condition, and the registration of the source point cloud and the target point cloud is completed.

[0043] The error is calculated based on the following formula:

[0044]

[0045] In this formula, E(T) represents the registration error. The system gradually adjusts the position and posture of the source point cloud to achieve the best alignment with the target point cloud by minimizing the error.

[0046] In each iteration, the source point cloud is adjusted by the transformation matrix T. The adjustment is based on the current registration error. The system will correct the transformation matrix T according to the size of the error until the error reaches the preset convergence condition. Finally, the registration error between the source point cloud and the target point cloud approaches the minimum, achieving accurate registration of the two. Iteration can effectively handle noise and outliers in point cloud data, and can also handle feature extraction of complex terrain, ensuring point cloud registration with high accuracy.

[0047] After the registration is completed, the map is updated through the active window strategy, and the newly added point cloud is added to the point cloud map. The update formula is as follows:

[0048] M k+1 =αM k +(1-α)N k (9)

[0049] This multi-scale adaptive registration method is particularly effective in complex three-dimensional environments (such as mountains and forests). It not only ensures accuracy, but also saves computing resources when processing simple terrain. During the UAV mission, real-time and efficient registration can ensure the accuracy of the environmental map and provide a basis for subsequent terrain feature extraction and landing point selection.

[0050] Step 4, terrain feature extraction and evaluation, when the drone selects the landing area, it must understand the detailed characteristics of the terrain, including slope, roughness, and height changes. Traditional terrain feature extraction methods may be affected by noise and outliers, resulting in inaccurate results. Therefore, the present invention uses the RANSAC (random sampling consensus) algorithm to extract terrain features to ensure that useful information can be accurately extracted even in noisy data.

[0051] Before feature extraction, the lidar point cloud data is first processed into a virtual grid. Each grid cell contains a number of point cloud data points, representing the terrain features within the cell. The entire terrain area is divided into N×M grid cells, and the size of each cell is Δx×Δy. Then:

[0052] Before feature extraction, the LiDAR point cloud data is first processed into a virtual grid. Each grid cell contains a number of point cloud data points, representing the terrain features within the cell. The entire terrain area is divided into N×M grid cells, and the size of each cell is Δx×Δy. The specific definition is as follows:

[0053] G(i,j)={p i,j |x min +i·Δx≤x <x min +(i+1)·Δx,y min +j·Δy≤y <y min +(j+1)·Δy} (10)

[0055] The boundary of each grid cell G(i,j) is defined by x min and min The decision represents the terrain area within the unit range, and the point cloud data contained in this range is used to describe the terrain characteristics of the area.

[0056] The RANSAC algorithm selects the best plane model from a large amount of point cloud data and removes outliers through random sampling and consistency checking. The specific steps are as follows:

[0057] RANSAC first randomly selects three points from the point cloud for plane fitting, and the fitted plane equation is:

[0058] z=a·x+b·y+c

[0059] Among them, a, b, and c are the parameters of the fitting plane.

[0060] The system determines whether these points belong to the fitted model by calculating the distance between other points and the fitted plane. If the distance from the point to the plane is less than the preset threshold, the point is considered an inlier, otherwise it is considered a noise or outlier. RANSAC will continue to randomly sample until it finds a plane model with the most inliers.

[0061] After obtaining the plane model, the system further calculates the key parameters of the terrain:

[0062] Slope represents the inclination of the terrain, and its formula is:

[0063]

[0064] Roughness indicates the flatness of the terrain, and the calculation formula is:

[0065]

[0066] The relief represents the height change of the terrain, and the calculation formula is:

[0067] U=z max -z min (13)

[0068] The safety distance represents the distance between the grid cell G(i,j) and the nearest obstacle, and the calculation formula is:

[0069] D s =min(||G(i,j)-O k ||)(14)

[0070] Through the RANSAC algorithm, the system can accurately extract flat areas suitable for drone landing, ensuring that useful terrain features can still be identified in the case of high environmental noise. This feature extraction result will directly affect the subsequent landing point selection.

[0071] Step 5: In complex terrain, selecting the best landing point is the key to the safe landing of the drone. The best landing point requires not only a flat terrain, but also the distribution of surrounding obstacles and a safe distance. To this end, the present invention adopts a multi-parameter comprehensive evaluation method to evaluate each potential landing point.

[0072] The system combines the two main parameters of terrain complexity and safety, and obtains the comprehensive evaluation index Li of each area through weighted calculation.

[0073] Terrain complexity assessment: Terrain complexity is determined by the weighted sum of slope, roughness and undulation. The formula is:

[0074] T i =λ θ ·θ+λ R ·R+λ U ·U(15)

[0075] Among them, λθ, λR, and λU are weight coefficients, which respectively represent the influence of slope, roughness, and undulation on terrain complexity.

[0076] Safety distance assessment: The safety distance is determined by calculating the distance between the landing point and the surrounding obstacles. The formula is:

[0077]

[0078] Where Ds: represents the distance between the landing point and the nearest obstacle. The larger the safety distance, the safer the area.

[0079] Comprehensive evaluation: The system takes terrain complexity and safety distance into consideration, and the final evaluation index is determined by the following formula:

[0080] L i =T i +τ·S i (17)

[0081] Among them, τ is the safety weight coefficient, which indicates the importance of safety relative to the complexity of the terrain.

[0082] Through this multi-parameter comprehensive evaluation method, the system can accurately select an area suitable for landing and avoid landing in areas with dense obstacles or uneven terrain. This mechanism ensures the safety of the drone in complex environments.

[0083] Step 6: After determining the best landing point, the drone needs a safe and efficient path to reach the landing point. The A* algorithm is a heuristic search algorithm that can find the path with the lowest cost between a given starting point and an end point. Its core is to find the optimal path from the starting point to the target point by estimating the total generation f(n) of the path. The path cost calculation formula of the A* algorithm is as follows:

[0084] f(n)=g(n)+h(n)(18)

[0085] Where: f(n): represents the estimated total cost from the starting point to the target point through node n, g(n): represents the actual path cost from the starting point to node n, h(n): represents the heuristic estimated cost from node n to the target point.

[0086] In the traditional A* algorithm, h(n) usually uses Euclidean distance or Manhattan distance to estimate the distance from node n to the target point, which usually only considers the length of the path and ignores the distribution of obstacles and flight energy consumption. Therefore, the present invention improves the A* algorithm, not only considering the path length, but also combining obstacle risk assessment and energy consumption optimization.

[0087] The improved A* algorithm combines three factors, path length, obstacle risk, and flight energy consumption, when planning the path. The details are as follows:

[0088] Path cost function: The cost function of the A* algorithm is:

[0089] f′(n)=g(n)+h(n)+λ r ·R(n)(19)

[0090] Among them, g(n) is the actual cost from the starting point to the current node n, h(n) is the heuristic estimate from the current node to the target point, R(n) is the obstacle risk assessment value, which indicates the density of obstacles passed by the path, and λr is the weight coefficient of risk assessment.

[0091] Energy consumption optimization: In view of the limited energy of drones, energy consumption optimization is introduced in the path planning process. The energy consumption of each path node is calculated based on factors such as its flight distance, flight altitude change, and ambient wind speed. The improved cost function is:

[0092] f″(n)=f′(n)+λ e ·E(n)(20)

[0093] Among them, E(n) is the estimated value of flight energy consumption, which represents the energy consumption of the drone from the current position to the target point, and λe is the energy consumption optimization weight.

[0094] By introducing a flight energy consumption optimization mechanism, the flight energy consumption of the UAV can be effectively reduced, especially in long-distance flight missions, which can significantly improve the endurance.

[0095] The present invention needs to be applicable to path planning in a three-dimensional environment, especially in complex terrain or areas with dense obstacles. Path planning needs to consider not only obstacles on the plane, but also the distribution of obstacles at different heights. The specific implementation is as follows:

[0096] Hierarchical path planning: Divide the 3D environment into multiple altitude layers, each of which corresponds to a 2D path planning problem. By running the A* algorithm on each altitude layer, selecting the optimal path, and combining the conversion costs between different altitude layers, a 3D path is finally generated.

[0097] Altitude conversion cost: If the drone needs to convert between different altitudes, the energy consumption and safety of altitude changes must be considered. The calculation formula for the altitude conversion cost is:

[0098] C h (n) = λ h ·Δh(21)

[0099] Among them, C h (n): represents the altitude transition cost at node n, λ h : Weight parameter of altitude conversion, Δh: altitude difference between current node and target node.

[0100] Through the altitude layer stratification and conversion mechanism, drones can flexibly plan paths in a three-dimensional environment, avoid high-risk areas, and ensure flight safety.

[0101] Implementation process of UAV path planning:

[0102] 1. Obtain the starting point, target point and environmental information (such as obstacle distribution, wind speed, etc.) of the drone, and divide the three-dimensional environment into multiple height layers.

[0103] 2. Based on the A* algorithm, path planning is performed at each altitude layer to calculate the optimal path from the starting point to the target point, taking into account factors such as path length, risk and energy consumption.

[0104] 3. Combine the conversion costs between different altitude layers to generate a three-dimensional path, and select the path with the lowest cost as the final flight path.

[0105] During the flight of the drone, environmental information (such as new obstacles or changes in wind speed) is updated in real time, and the flight path is dynamically adjusted based on the latest data.

[0106] Step 6: In actual application scenarios, the environment may change dynamically. For example, obstacles may move and wind speed may increase, which may affect the flight mission of the drone. Therefore, the present invention designs a dynamic update and adjustment mechanism to ensure that the drone can adjust the flight path and landing point according to real-time environmental information during flight.

[0107] LiDAR and IMU provide real-time updated environmental information, including terrain changes, obstacle movement, etc. The system continuously updates the 3D environmental model based on this real-time data and recalculates the optimal landing point and path planning.

[0108] When the system detects changes in the environment (such as new obstacles or changes in wind speed), the system will dynamically adjust the flight path based on the latest data. The purpose of path adjustment is to ensure that the drone always chooses a safe and optimal route to reach the target point, while ensuring that the drone's energy efficiency is not significantly reduced.

Claims

1. A three-dimensional modeling and UAV landing area selection method based on laser point cloud data, characterized in that: The following steps are involved: Step 1, obtain point cloud data: use the lidar system to collect three-dimensional point cloud data of the drone's surrounding environment; Step 2, data preprocessing: performing Kalman filtering and coordinate transformation on the acquired 3D point cloud data; Step 3: Build a 3D point cloud map: Based on the preprocessed point cloud data, use the 3D normal distribution transform (3D-NDT) algorithm to perform point cloud registration and generate a high-precision 3D point cloud map. Step 4, extract terrain features: perform virtual grid processing on the 3D point cloud map, and extract terrain parameters including slope, roughness, undulation and safety distance through sliding window plane fitting; Step 5, select the best landing area: calculate the best landing point selection index based on terrain complexity and safety distance, and select the area with the smallest index as the landing area; Step 6, path planning and autonomous landing: Based on the selected optimal landing area, the improved A* path planning algorithm is used to plan the landing path of the UAV and perform autonomous landing.

2. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: The three-dimensional point cloud data in step 1 includes the three-dimensional coordinates of objects around the drone, and these coordinates constitute the terrain model and the distribution of obstacles.

3. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: The step 2 is specifically as follows: 1) State prediction: The motion state of the drone is predicted by combining Kalman filtering with IMU data, and the process noise covariance matrix Q of the IMU is introduced imu , to improve the accuracy of the prediction, first predict the current state of the drone based on the state and input of the previous moment: in, represents the state prediction at the current time k, F: state transfer matrix, describing the state change of the system from time k-1 to time k, represents the state estimate at time k-1, B: control input matrix, u k : Control input, ΔP imu : displacement correction provided by IMU; Then make predictions on the covariance matrix: P k|k-1 =F·P k-1|k-1 ·F T +Q+Q imu (2) Where: P k|k-1 : represents the state covariance matrix at prediction time k, P k-1|k-1 : represents the state covariance matrix at time k-1, F T : transpose of the state transfer matrix, Q : process noise covariance matrix, Q imu : IMU noise covariance matrix; 2) Dynamically compensate for point cloud errors: Combine the IMU noise information data and the LiDAR point cloud data to dynamically adjust the Kalman gain, and combine the IMU attitude data to correct the point cloud's position and rotation to dynamically compensate for point cloud errors. The process is as follows: First calculate the Kalman gain: K k =P k|k-1 ·H T ·(H·P k|k-1 ·H T +R+R imu ) -1 (3) Among them, K k : represents the Kalman gain, H: observation matrix, H T : transpose of the observation matrix, R : measurement noise covariance matrix, R imu : IMU measurement noise covariance matrix; Then update the state prediction by observing the data: in, The updated state at time k is z k : represents the actual observation value at time k, Δθ imu : Rotation correction provided by IMU; Finally, update the state covariance matrix of the current moment k: P k|k =(I-K k ·H)·P k|k-1 (5) Where P k|k : Updated state covariance matrix, I: identity matrix.

4. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: Step 3 shown uses the 3D normal distribution transform 3D-NDT algorithm to perform point cloud registration method: 1) The 3D-NDT algorithm divides the point cloud data into multiple regions and performs Gaussian distribution modeling on each region. The expression is: Among them, P i (x): represents the probability distribution of the i-th region in the point cloud, x: the point with registration, μ i : represents the mean vector of the i-th region, ∑ i : represents the covariance matrix of the ith region, d: the dimension of the point cloud, T : Indicates transpose. 2) Based on the basic 3D-NDT algorithm, an adaptive multi-scale registration mechanism is added to dynamically adjust the registration scale by calculating the complexity parameter λi of each region. The covariance matrix of the adaptive adjustment is: in, represents the covariance matrix after multi-scale adjustment, f(∑ i ,λ i ): represents the multi-scale function, λ i : represents the regional complexity parameter; 3) The registration process is completed through iteration. In each iteration, the error between the target point cloud and the source point cloud is calculated, and the pose of the point cloud is adjusted according to the error until the error meets the preset convergence condition. The error between the target point cloud and the source point cloud is minimized according to the following formula to complete the registration: Among them, E(T): represents the error function, represents a point in the target point cloud, Represents the point in the source point cloud, T: transformation matrix, t: represents the coordinates of the point in the target point cloud, s: represents the coordinates of the point in the source point cloud.

5. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: The three-dimensional point cloud map generation method in step 3 is as follows: According to the registered point cloud, at each time step, the sliding window local map is updated using the following formula to reduce error accumulation: M k+1 =αM k +(1-α)N k (9) Among them, M k+1 : represents the local map of the next frame, M k : Current local map, N k : new point cloud data, α: smoothing coefficient, k: represents the number of time steps, which is used to represent the state update of the algorithm at different times.

6. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: Step 4 shown is specifically: First, the LiDAR point cloud data is processed into a virtual grid. Each grid cell contains a number of point cloud data points, which represent the terrain features within the cell. The entire terrain area is divided into N×M grid cells, and the size of each cell is Δx×Δy. Then: G(i,j)={p i,j |x min +i·Δx≤x<x min +(i+1)·Δx,y min +j·Δy≤y<y min +(j+1)·Δy} (10) Where, G(i,j): the i-th and j-th grid cells, Δx is the horizontal grid cell size, and Δy is the vertical grid cell size. Secondly, the RANSAC algorithm is used to perform plane fitting on the point cloud data in each grid unit. The RANSAC random sampling algorithm is used to fit a plane model that conforms to the point cloud data in the presence of noise and outliers in the data.

7. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: Step 5 shown is specifically: Based on the plane model obtained by RANSAC fitting, the terrain parameters of the grid unit are calculated. The terrain parameters include: slope, roughness, undulation, and safety distance. The calculation formula is as follows: Slope: The inclination of the fitted plane, indicating the steepness of the terrain. The calculation formula is: Among them, θ: represents the slope of the area, a and b: represent the parameters of the plane equation; Coarse degree: The average distance from each point in the point cloud to the fitting plane, indicating the smoothness of the terrain. The calculation formula is: Where R: represents the terrain roughness, z i : represents the height of the i-th point in the point cloud, N: represents the number of point clouds in the grid unit; Relief: The height difference between the highest point and the lowest point in a local area indicates the degree of relief of the terrain. The calculation formula is: U=z max -With min (13) Among them, U: represents the undulation of the terrain, z max and z min : Indicates the highest and lowest points in the grid area; Safety distance: The distance between the grid cell G(i,j) and the nearest obstacle. The safety of landing is ensured by calculating the distance between the center point of each grid cell and the surrounding dangerous grids: D s =min(||G(i,j)-O k ||)(14) Where: D s : Indicates safe distance, O k : Indicates the grid center of surrounding obstacles; Based on the above parameters, the terrain complexity index Ti and safety index Si of the grid cell are calculated to evaluate the best landing point. The terrain complexity index Ti is calculated by the weighted sum of slope, roughness and undulation: T i =λ θ ·θ+λ R ·R+λ U ·U(15) Among them, λ θ ,λ R ,λ U : Indicates slope, roughness, and undulation The safety index Si is determined by the safety distance Ds: Combining the terrain complexity index and safety index, the landing evaluation index Li of each grid cell is finally calculated: L i =T i +τ·S i (17) Among them, L i : represents the landing evaluation index of the grid unit, τ: represents the safety weight coefficient.

8. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: The process of selecting a landing area is as follows: 1) For the current flight area of ​​the drone, the terrain is modeled and features are extracted based on the lidar point cloud data to obtain the terrain complexity index Ti and safety distance index Si of each grid unit; 2) Calculate the comprehensive landing assessment index Li of all grid cells and sort them from small to large. Generally, the smaller the Li value, the flatter the terrain in the area, the less obvious obstacles there are, and the area is far away from potential danger areas. 3) Select the grid cell with the minimum Li value as the best landing point; if the area cannot meet the actual landing requirements, select the next area with a smaller Li value; 4) Send the coordinates of the best landing point to the drone flight control system and perform path planning based on the selected area to ensure that the drone can land safely.

9. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1 is characterized in that: The improved A* path planning algorithm in step 6 is as follows: A* algorithm is used, and its formula is f(n)=g(n)+h(n)(18) 1) In the path planning process, the length of the path is considered and the risk assessment of obstacles is added. The risk of each path node is determined by the density and distance of the surrounding obstacles. The improved cost function is: f′(n)=g(n)+h(n)+λ r ·R(n)(19) Among them, R(n): represents the risk assessment value of node n, λ r : represents the risk weight parameter; 2) Energy consumption optimization is introduced in the path planning process. The energy consumption of each path node is calculated based on factors such as its flight distance, flight altitude change, and ambient wind speed. The improved cost function is: f″(n)=f′(n)+λ e ·E(n)(20) Where, E(n): represents the flight energy consumption of node n, λ e : represents the energy consumption weight parameter; 3) Hierarchical path planning: Divide the 3D environment into multiple altitude layers, each of which corresponds to a 2D path planning problem. By running the A* algorithm on each altitude layer, selecting the optimal path, and combining the conversion costs between different altitude layers, a 3D path is finally generated; 4) Altitude conversion cost: If the UAV needs to convert between different altitudes, the energy consumption and safety of altitude changes must be considered. The calculation formula for the altitude conversion cost is: C h (n)=λ h ·Δh(21) Among them, C h (n): represents the altitude level switching cost at node n, λ h : Weight parameter of altitude conversion, Δh: altitude difference between current node and target node.

10. The three-dimensional modeling and UAV landing area selection method based on laser point cloud data according to claim 1, characterized in that: Implementation process of planning the landing path of the drone: 1) Obtain the starting point, target point and environmental information (such as obstacle distribution, wind speed, etc.) of the drone, and divide the three-dimensional environment into multiple height layers; 2) Based on the A* algorithm, path planning is performed at each altitude layer to calculate the optimal path from the starting point to the target point, taking into account factors such as path length, risk, and energy consumption; 3) Combine the conversion costs between different altitude layers to generate a three-dimensional path, and select the path with the lowest cost as the final flight path; 4) During the flight of the drone, environmental information (such as new obstacles or wind speed changes) is updated in real time, and the flight path is dynamically adjusted based on the latest data.

Images