Unmanned aerial vehicle point cloud terrain filtering method and system based on gradient constraint and improved distribution simulation

Abstract

The invention discloses an unmanned aerial vehicle point cloud terrain filtering method and system based on gradient constraint and improved distribution simulation, and aims to solve the problems that unmanned aerial vehicle photogrammetry point cloud has much noise and irregular distribution in a complex terrain, the method comprises the following steps: firstly, calculating a gradient change rate by constructing a point cloud neighborhood relationship, and establishing a terrain preliminary screening constraint model; rapid distinguishing between potential ground points and non-ground points is realized; on this basis, a distribution simulation model is constructed, a gradient threshold adjustment mechanism is introduced, and physical parameters of distribution particles are adaptively adjusted according to gradient features of local terrains, so that refined fitting of complex terrains is realized; and finally, calculating the distance from the preprocessed point cloud to the surface of the simulated cloth to complete ground point extraction. The digital elevation model generated by the method is high in elevation precision and low in earth-rock calculation error, can effectively overcome the problem that a traditional filtering method is low in precision under complex terrain conditions, and provides an effective technical path for high-precision terrain extraction of photogrammetry point clouds.

Description

A method and system for UAV point cloud terrain filtering based on gradient constraints and improved cloth simulation Technical Field

[0001] This invention belongs to the field of photogrammetry and data processing technology, and in particular relates to a method and system for UAV point cloud terrain filtering based on gradient constraints and improved cloth simulation. Background Technology

[0002] The development of UAV oblique photogrammetry technology has led to point cloud data acquired via UAV platforms gradually becoming an important source of terrain information. Compared to traditional surveying methods, UAV point cloud data offers advantages such as high acquisition efficiency and wide coverage, and has been widely used in topographic mapping, engineering surveys, and other fields. However, current UAV-acquired point cloud data often suffers from high noise levels and low filtering accuracy under complex terrain conditions. This is particularly true in photogrammetric point clouds generated through dense matching of multi-view images, where noise exhibits high frequency and irregular distribution, posing significant challenges to terrain data processing and analysis.

[0003] Existing point cloud filtering methods struggle to achieve an effective balance between noise removal and terrain feature preservation. Cloth simulation filtering, a relatively mature point cloud filtering method, is widely used due to its ability to closely fit terrain surfaces. However, the physical parameters in this method are typically set globally and fixedly. In complex survey areas containing various terrain types such as flat land, steep slopes, and cliffs, it is difficult to meet the filtering needs of different terrain regions, thus affecting the overall filtering effect. Summary of the Invention

[0004] Purpose of the invention: This invention provides a method and system for UAV point cloud terrain filtering based on gradient constraints and improved cloth simulation, aiming to solve the problems of low filtering accuracy and poor terrain feature preservation in existing technologies under complex terrain.

[0005] Technical solution: This invention provides a UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation, including:

[0006] Obtain the original point cloud data of the survey area;

[0007] The original point cloud data is preprocessed, and the spatial topology of the point cloud data is constructed based on the preprocessed point cloud data.

[0008] Based on the spatial topological relationship, the gradient feature values ​​of each point in the preprocessed point cloud data are calculated, and a terrain preliminary screening constraint model is constructed. Based on the gradient feature values ​​of each point in the preprocessed point cloud data, a candidate ground point set is obtained.

[0009] Flip the candidate ground point set to obtain the collision object, and initialize the traditional cloth simulation model above the collision object;

[0010] An improved cloth simulation iterative process is performed on the traditional cloth simulation model. During the iteration process, the stiffness value of the cloth particles is adaptively adjusted according to the gradient feature value of the midpoint of the collision object corresponding to the cloth particles until the parameter combination of the traditional cloth simulation model reaches the optimal value, thereby obtaining the cloth model after it is stationary.

[0011] An approximate terrain surface is generated based on the static cloth model; the distance from each point in the preprocessed point cloud data to the approximate terrain surface is calculated, and each point in the preprocessed point cloud data is classified based on the distance threshold.

[0012] Furthermore, the preprocessing includes statistical outlier removal filtering of the original point cloud and downsampling of the original point cloud data using a voxel grid method.

[0013] Furthermore, the construction of the spatial topology relationship of the point cloud data includes: using the KD Tree algorithm to construct a spatial index of the preprocessed point cloud data, determining the set of the k nearest neighbors of each point in the preprocessed point cloud data, and establishing the spatial topology relationship of the point cloud data.

[0014] Furthermore, the step of calculating the gradient feature values ​​of each point in the preprocessed point cloud data based on the spatial topological relationship includes:

[0015] Define the rate of change of elevation gradient The formula is:

[0016] ;

[0017] in, For the midpoint of the preprocessed point cloud data The three-dimensional coordinates For point The three-dimensional coordinates of the j-th point in the nearest neighbor set, where k is the point. The number of nearest neighbor sets; For point The largest elevation gradient change rate in the nearest neighbor set is also used as a point. The gradient eigenvalues.

[0018] Furthermore, the initial terrain screening constraint model includes:

[0019] The gradient characteristic values ​​of each point in the topography of the survey area are statistically analyzed, and the average gradient characteristic of the topography of the survey area is calculated using the following formula:

[0020] ;

[0021] Where N is the number of points in the topography of the survey area that are included in the statistics;

[0022] An initial gradient threshold is set based on the average gradient characteristics of the topography of the survey area. The formula is:

[0023] ;

[0024] in, This is the adjustment coefficient;

[0025] like Then the decision point Points in steep areas are marked as non-ground points;

[0026] like Then the decision point For candidate ground points, add them to the candidate ground point set.

[0027] Furthermore, the step of adaptively adjusting the stiffness value of the cloth particles based on the gradient feature value of the midpoint of the collision object during the iteration process includes:

[0028] In each iteration, the stiffness value of the fabric particles is set as a function of the gradient eigenvalues, with the following formula:

[0029] ;

[0030] in, Let be the hardness value of the i-th fabric particle. Based on basic hardness, Let be the normalized gradient feature value of a point in the collision object corresponding to the i-th cloth particle. As a regulating factor;

[0031] The position of the cloth particles is updated based on the Verlet integral method, and the next iteration is carried out until the cloth particles tend to be stationary. The cloth mesh resolution and the number of iterations are recorded at this time.

[0032] By adjusting the cloth mesh resolution and the number of iterations, an optimal set of parameters was finally determined.

[0033] Based on the optimal parameter combination, the cloth model after it comes to rest is obtained.

[0034] Furthermore, the process of generating an approximate terrain surface based on the stationary cloth model, calculating the distance from each point in the preprocessed point cloud data to the approximate terrain surface, and classifying each point in the preprocessed point cloud data based on a distance threshold includes:

[0035] The cloth model, after being brought to a standstill, is flipped back to the original coordinate system to generate an approximate terrain surface.

[0036] Calculate the distance from each point in the preprocessed point cloud data to the approximate terrain surface. , where i represents the i-th point, and a distance threshold is set. ,like If the i-th point in the original point cloud data is determined to be the final ground point, then it is determined to be a non-ground point.

[0037] This invention also provides a drone point cloud terrain filtering system based on gradient constraints and improved cloth simulation, comprising:

[0038] The data acquisition module is used to acquire raw point cloud data of the survey area;

[0039] The preprocessing module is used to preprocess the original point cloud data and construct the spatial topology relationship of the point cloud data based on the preprocessed point cloud data.

[0040] The gradient calculation module is used to calculate the gradient feature values ​​of each point in the preprocessed point cloud data based on the spatial topological relationship, and to construct a terrain preliminary screening constraint model. Based on the gradient feature values ​​of each point in the preprocessed point cloud data, a candidate ground point set is obtained by screening.

[0041] The initial cloth model module is used to flip the candidate ground point set to obtain the collision object, and initialize the traditional cloth simulation model on top of the collision object;

[0042] The iterative optimization module is used to perform an improved cloth simulation iterative process on the traditional cloth simulation model. During the iteration process, the stiffness value of the cloth particles is adaptively adjusted according to the gradient feature value of the midpoint of the collision object corresponding to the cloth particles until the parameter combination of the traditional cloth simulation model reaches the optimal value, thereby obtaining the cloth model after it is stationary.

[0043] The classification module is used to generate an approximate terrain surface based on the static cloth model; calculate the distance from each point in the preprocessed point cloud data to the approximate terrain surface, and classify each point in the preprocessed point cloud data based on the distance threshold.

[0044] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0045] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0046] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0047] 1. In view of the high-frequency and irregular distribution of noise in photogrammetric point clouds, a terrain initial screening constraint is constructed by using the gradient change rate to effectively eliminate non-ground point noise.

[0048] 2. An improved cloth simulation model is constructed, and a gradient threshold adjustment mechanism is introduced to enhance the algorithm's noise resistance in flat areas and make it more suitable for complex terrain areas, thus solving the problem of insufficient adaptability of traditional cloth simulation models in variable terrain environments.

[0049] 3. Experimental verification shows that the digital elevation model generated by this method has an elevation accuracy better than 3cm, and the earthwork calculation error is controlled within 3%, which meets the requirements of high-precision terrain extraction and mapping. Attached Figure Description

[0050] Figure 1 is a flowchart of the overall process of the method of the present invention.

[0051] Figure 2 is a schematic diagram comparing point clouds before and after the initial terrain screening.

[0052] Figure 3 is a schematic diagram of the improved fabric simulation based on gradient adaptation.

[0053] Figure 4 is a schematic diagram showing the comparison of effects. Detailed Implementation

[0054] As shown in Figure 1, the UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to the present invention includes:

[0055] S1. Obtain the original point cloud data of the survey area;

[0056] S2. Preprocess the original point cloud data and construct the spatial topology relationship of the point cloud data based on the preprocessed point cloud data;

[0057] S3. Calculate the gradient feature values ​​of each point in the preprocessed point cloud data based on the spatial topological relationship, and construct a terrain preliminary screening constraint model. Based on the gradient feature values ​​of each point in the preprocessed point cloud data, select and obtain a candidate ground point set.

[0058] S4. Flip the candidate ground point set to obtain the collision object, and initialize the traditional cloth simulation model above the collision object.

[0059] S5. An improved cloth simulation iterative process is performed on the traditional cloth simulation model. During the iteration process, the hardness value of the cloth particles is adaptively adjusted according to the gradient feature value of the midpoint of the collision object corresponding to the cloth particles until the parameter combination of the traditional cloth simulation model reaches the optimal value, thereby obtaining the cloth model after it is stationary.

[0060] S6. Generate an approximate terrain surface based on the static cloth model; calculate the distance from each point in the preprocessed point cloud data to the approximate terrain surface, and classify each point in the preprocessed point cloud data based on the distance threshold.

[0061] In this embodiment,

[0062] S1. Obtaining raw point cloud data for the survey area: This invention generates raw point cloud data containing three-dimensional coordinates and texture information using UAV oblique photogrammetry technology. This raw point cloud data is high-precision photogrammetric point cloud data for the survey area. The specific implementation process includes:

[0063] Acquire basic data for the survey area and conduct on-site reconnaissance. Based on the terrain undulations and accuracy requirements of the survey area, complete the detailed planning of the UAV's autonomous operation flight path, and set key parameters such as flight altitude, heading, lateral overlap, and gimbal tilt angle; simultaneously complete the deployment and measurement of image control points.

[0064] Distortion correction and quality checks are performed on the acquired multi-view tilted images; bundle adjustment is performed using image control points, and multiple aerial triangulation calculations are executed to ensure the accuracy of the densification points and image positions, generating high-precision sparse point clouds and image exterior orientation elements.

[0065] Based on the results of aerial triangulation, an irregular triangular mesh was constructed using multi-view image dense matching technology, and original high-density 3D point cloud data with RGB texture information was generated.

[0066] S2. Preprocess the original point cloud data, including denoising and thinning the original point cloud data: perform statistical outlier removal filtering on the point cloud data to remove floating points and discrete noise caused by matching errors, and use the voxel grid method to downsample the point cloud data to reduce data redundancy while preserving terrain features.

[0067] The KD Tree algorithm is used to construct a spatial index for the preprocessed point cloud data, thereby determining the location of each point in the preprocessed point cloud data. Establish the spatial topological relationship of the point cloud data by using the set of k nearest neighbors.

[0068] S3. By analyzing the local geometric features in the spatial topology of point cloud data, a terrain preliminary screening constraint model is constructed to quickly remove significant non-ground points. The specific steps are as follows:

[0069] S3.1 Define the elevation gradient change rate G of point cloud data i The calculation formula is as follows:

[0070] ;

[0071] in, For the current point The three-dimensional coordinates Let k be the coordinates of the j-th point in its neighborhood (the set of nearest neighbors), and k be the number of points in the neighborhood. For point The largest elevation gradient change rate in the nearest neighbor set is also used as a point. The gradient eigenvalues.

[0072] S3.2. Statistically analyze the gradient characteristic values ​​of each point in the topography of the survey area, and calculate the average gradient characteristic of the topography of the survey area. The formula is:

[0073] ;

[0074] Where N is the number of points in the topography of the survey area that are included in the statistics;

[0075] An initial gradient threshold is set based on the average gradient characteristics of the topography of the survey area. The formula is:

[0076] ;

[0077] in, This is an adjustment coefficient used to control the sensitivity of the gradient threshold to changes in terrain gradient;

[0078] As shown in Figure 2, if If the point is determined to be in a high-steep area, it is marked as a non-ground point and needs to be removed; Mark them as candidate ground points, add them to the candidate ground point set, and they will enter the subsequent cloth simulation stage for fine processing;

[0079] S4. Flip the candidate ground point set filtered in S3 to invert the terrain undulations and obtain the collision object. The points in the candidate ground point set correspond to the midpoints of the flipped collision objects.

[0080] The initialization of the traditional cloth simulation model is completed by initializing a virtual cloth grid above the inverted point cloud data (i.e., the collision object) and setting the initial position of the cloth particles above the collision object.

[0081] S5. As shown in Figure 3, an improved cloth simulation iterative process is performed on the traditional cloth simulation model. During the iteration process, the stiffness value of the cloth particles is adaptively adjusted based on the gradient characteristic value of the midpoint of the collision object corresponding to the cloth particles. This solves the problem of over-smoothing or penetration caused by the fixed cloth simulation model in the traditional algorithm, until the parameter combination of the traditional cloth simulation model reaches its optimal value, thus obtaining the static cloth model. Specifically, this includes:

[0082] In each iteration, the stiffness value of the cloth particles is set as a function of the gradient feature value, and the following mapping relationship is constructed:

[0083] ;

[0084] in, Let be the hardness value of the i-th fabric particle. Based on basic hardness, Let be the normalized gradient feature value of a point in the collision object corresponding to the i-th cloth particle. The adjustment factor is used; this formula makes the gradient small in flat regions, Approaching 0, hardness The fabric remains relatively rigid, effectively resisting noise interference in low-lying areas; in areas with large gradients and steep slopes, the hardness... The fabric becomes softer and can deform better to conform to terrain details;

[0085] The positions of the cloth particles are updated based on the Verlet integral method. Combined with the adaptive stiffness parameters mentioned above, the next iteration is entered to simulate the collision process between the cloth and the inverted point cloud under the action of gravity until the cloth particles tend to be stationary. The cloth mesh resolution and the number of iterations are recorded at this time.

[0086] By adjusting the cloth mesh resolution and the number of iterations, an optimal set of parameters was finally determined.

[0087] Based on the optimal parameter combination, the cloth model after it comes to rest is obtained.

[0088] S6. Flip the cloth model back to the original coordinate system after it has been stationary to generate an approximate terrain surface, and calculate the vertical distance from each point in the preprocessed point cloud data to the approximate terrain surface. , where i represents the i-th point, and a distance threshold is set. :like If the condition is met, the i-th point in the preprocessed point cloud data is determined to be the final ground point; otherwise, the point is determined to be a non-ground point.

[0089] The final output is high-precision ground point cloud data after classification, which is used to generate digital elevation models and perform subsequent earthwork calculations.

[0090] The present invention discloses a drone point cloud terrain filtering system based on gradient constraints and improved cloth simulation, comprising:

[0091] The data acquisition module is used to acquire raw point cloud data of the survey area;

[0092] The preprocessing module is used to preprocess the original point cloud data and construct the spatial topology relationship of the point cloud data based on the preprocessed point cloud data.

[0093] The gradient calculation module is used to calculate the gradient feature values ​​of each point in the preprocessed point cloud data based on the spatial topological relationship, and to construct a terrain preliminary screening constraint model. Based on the gradient feature values ​​of each point in the preprocessed point cloud data, a candidate ground point set is obtained by screening.

[0094] The initial cloth model module is used to flip the candidate ground point set to obtain the collision object, and initialize the traditional cloth simulation model on top of the collision object;

[0095] The iterative optimization module is used to perform an improved cloth simulation iterative process on the traditional cloth simulation model. During the iteration process, the stiffness value of the cloth particles is adaptively adjusted according to the gradient feature value of the midpoint of the collision object corresponding to the cloth particles until the parameter combination of the traditional cloth simulation model reaches the optimal value, thereby obtaining the cloth model after it is stationary.

[0096] The classification module is used to generate an approximate terrain surface based on the static cloth model; calculate the distance from each point in the preprocessed point cloud data to the approximate terrain surface, and classify each point in the preprocessed point cloud data based on the distance threshold.

[0097] Regarding the above-described technical solution of this embodiment, the process of applying the technical solution to the earthwork surveying area of ​​a specific construction project and the technical effects of the technical solution will be described in detail below with reference to Figures 1 to 4, so as to verify the applicability of the method of the present invention in actual engineering.

[0098] As shown in Figure 1, the survey area has a complex terrain environment, encompassing various terrain types such as flat areas, artificially piled mounds, and vegetated areas. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation described in this invention includes:

[0099] S1. Obtain raw point cloud data:

[0100] Based on the terrain undulations of the survey area, a UAV terrain-following flight path was planned, with a relative flight altitude of 80m, a forward overlap of 80%, a lateral overlap of 70%, and a gimbal tilt angle set to -45° to -90° for multi-angle acquisition to ensure complete acquisition of terrain texture.

[0101] Several image control points were set up in the survey area, and their high-precision coordinates were measured using RTK equipment.

[0102] The quality of the image data collected by the drone is checked, and invalid data with abnormal exposure, blurry images or missing location information is removed.

[0103] A first aerial triangulation was performed based on valid image data, and ties with reprojection errors greater than 1 pixel were removed according to the quality report. Image control points were then used to create punctures in the stereo model, followed by a second aerial triangulation for absolute orientation, ensuring a planar position accuracy better than 5 cm and an elevation accuracy better than 10 cm. Based on this, an irregular triangulation network was constructed, and raw point cloud data containing RGB color information was generated through dense matching of multi-view images.

[0104] S2. To address the inherent high-frequency noise and outlier characteristics of photogrammetric point clouds, preprocessing of the raw point cloud data is required:

[0105] Statistical filtering was used, with the number of neighborhood points k=50 and the standard deviation factor std=1.0, to remove floating noise points that were significantly deviated from the main point cloud.

[0106] Data thinning was performed using a voxel grid downsampling method, with a grid size of 0.1m × 0.1m × 0.1m, to reduce data redundancy while preserving terrain features.

[0107] S3. Based on the gradient rate of change, construct a KD Tree spatial index for initial terrain screening, calculate the gradient features of each point and its k-neighborhood, and construct a terrain screening constraint model:

[0108] Definition point Elevation gradient change rate G i The maximum slope between this point and all points in its neighborhood is calculated using the following formula:

[0109] ;

[0110] in, For the current point p i The three-dimensional coordinates Let k be the coordinates of the j-th point in its neighborhood, and k be the number of neighborhood points.

[0111] Set initial gradient threshold As shown in Figure 2, if If the point is determined to be in a high-steep area, it is marked as a non-ground point and needs to be removed; it will not participate in the collision calculation of subsequent cloth placement simulations. Mark it as a candidate ground point and proceed to step S4.

[0112] S4. Flip the candidate ground point set filtered by S3, i.e. This reverses the terrain's contours. Initialize the cloth grid above the flipped point cloud. Set the cloth resolution to 0.5m and the initial height slightly above the highest point of the site.

[0113] S5. As shown in Figure 3, a gradient adaptive adjustment mechanism is introduced during the iterative process of cloth simulation. Traditional cloth simulation methods typically use a globally fixed value for the stiffness of cloth particles, leading to a problem where residual noise on flat ground and smoothness on steep slopes are difficult to achieve simultaneously. Therefore, this embodiment establishes a gradient-stiffness mapping function:

[0114] In each iteration, the local average gradient of the original point cloud corresponding to the current position of the cloth particle is used. The hardness parameter of the particle is dynamically adjusted. :

[0115] ;

[0116] The fabric particles gradually fall under the influence of gravity, while simultaneously being constrained by internal forces imposed by the aforementioned dynamic stiffness parameter R. After 500 iterations, the fabric collides with the flipped point cloud and reaches a stable state, ultimately forming a model that approximates a realistic terrain surface.

[0117] S6. Flip the model of the approximate terrain surface back to the original coordinate system to generate an approximate terrain surface. Calculate the vertical distance from each point in the preprocessed point cloud data to the approximate terrain surface. Set the distance threshold h = 0.3m. If If the point is determined to be a ground point, then it is considered a non-ground point; otherwise, it is considered a non-ground point.

[0118] The method of this invention was compared and analyzed with the traditional CSF algorithm. As shown in Figure 4, the experimental results show that in flat areas, the present invention effectively reduces the impact of noise on the fabric simulation results by adaptively adjusting the fabric stiffness through gradient, avoiding false textures caused by high-frequency noise in photogrammetry, thereby improving the smoothness of the ground. In steep slope areas such as the edge of the soil mound, the fabric stiffness decreases accordingly with the increase of the terrain gradient, allowing the fabric to better conform to the real terrain and avoiding the peak-shaving and valley-filling phenomenon produced by the traditional algorithm. The soil mound volume calculation error was reduced from 3.9% to 2.8%. The final digital elevation model generated has an elevation accuracy controlled within 3 cm and an earthwork calculation error controlled within 3%, meeting the accuracy requirements of 1:500 large-scale topographic mapping.

Claims

1. A UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation, characterized in that, include: Obtain the original point cloud data of the survey area; The original point cloud data is preprocessed, and the spatial topology of the point cloud data is constructed based on the preprocessed point cloud data. Based on the spatial topological relationship, the gradient feature values ​​of each point in the preprocessed point cloud data are calculated, and a terrain initial screening constraint model is constructed. Candidate ground point sets are obtained by filtering based on the gradient feature values ​​of each point in the preprocessed point cloud data. The candidate ground point sets are flipped to obtain collision objects, and a traditional cloth simulation model is initialized above the collision objects. An improved cloth simulation iteration process is performed on the traditional cloth simulation model. During the iteration process, the hardness value of the cloth particles is adaptively adjusted according to the gradient feature values ​​of the points in the collision objects corresponding to the cloth particles, until the parameter combination of the traditional cloth simulation model reaches the optimal value, thus obtaining a stationary cloth model. An approximate terrain surface is generated based on the stationary cloth model. The distance from each point in the preprocessed point cloud data to the approximate terrain surface is calculated, and each point in the preprocessed point cloud data is classified based on a distance threshold.

2. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 1, characterized in that, The preprocessing includes statistical outlier removal filtering of the original point cloud and downsampling of the original point cloud data using the voxel grid method.

3. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 1, characterized in that, The construction of spatial topology relationships for point cloud data includes: using the KD Tree algorithm to construct a spatial index for the preprocessed point cloud data, determining the set of k nearest neighbors for each point in the preprocessed point cloud data, and establishing spatial topology relationships for the point cloud data.

4. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 3, characterized in that, The calculation of gradient feature values ​​for each point in the preprocessed point cloud data based on the spatial topological relationship includes: defining the elevation gradient change rate. The formula is: ;in, For the midpoint of the preprocessed point cloud data The three-dimensional coordinates For point The three-dimensional coordinates of the j-th point in the nearest neighbor set, where k is the point. The number of nearest neighbor sets; For point The largest elevation gradient change rate in the nearest neighbor set is also used as a point. The gradient eigenvalues.

5. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 4, characterized in that, The initial terrain screening constraint model includes: statistically analyzing the gradient feature values ​​of each point in the terrain of the survey area, and calculating the average gradient feature of the terrain of the survey area, using the following formula: Where N is the number of points in the topography of the survey area that are included in the statistics; an initial gradient threshold is set based on the average gradient characteristics of the topography of the survey area. The formula is: ;in, This is the adjustment coefficient; if Then the decision point Points in steep areas are marked as non-ground points; if Then the decision point For candidate ground points, add them to the candidate ground point set.

6. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 1, characterized in that, The step of adaptively adjusting the stiffness value of the cloth particles based on the gradient feature value of the midpoint of the collision object during the iteration process includes: in each iteration, setting the stiffness value of the cloth particles as a function of the gradient feature value, with the formula as follows: ;in, Let be the hardness value of the i-th fabric particle. Based on basic hardness, Let be the normalized gradient feature value of a point in the collision object corresponding to the i-th cloth particle. The position of the cloth particles is updated based on the Verlet integral method, and the next iteration is carried out until the cloth particles tend to be stationary. The cloth mesh resolution and the number of iterations are recorded at this time. By adjusting the cloth mesh resolution and the number of iterations, an optimal set of parameters is finally determined. Based on the optimal set of parameters, the stationary cloth model is obtained.

7. The UAV point cloud terrain filtering method based on gradient constraints and improved cloth simulation according to claim 6, characterized in that, The process of generating an approximate terrain surface based on the stationary cloth model, calculating the distance from each point in the preprocessed point cloud data to the approximate terrain surface, and classifying the points in the preprocessed point cloud data based on a distance threshold includes: flipping the stationary cloth model back to the original coordinate system to generate an approximate terrain surface; and calculating the distance from each point in the preprocessed point cloud data to the approximate terrain surface. , where i represents the i-th point, and a distance threshold is set. ,like If the i-th point in the original point cloud data is determined to be the final ground point, then it is determined to be a non-ground point.

8. A drone point cloud terrain filtering system based on gradient constraints and improved cloth simulation, characterized in that, include: The data acquisition module is used to acquire raw point cloud data of the survey area; The preprocessing module is used to preprocess the original point cloud data and construct the spatial topology relationship of the point cloud data based on the preprocessed point cloud data. The gradient calculation module is used to calculate the gradient feature values ​​of each point in the preprocessed point cloud data based on the spatial topological relationship, and to construct a terrain initial screening constraint model. Based on the gradient feature values ​​of each point in the preprocessed point cloud data, a candidate ground point set is obtained. The initial cloth model module is used to flip the candidate ground point set to obtain collision objects, and initialize a traditional cloth simulation model above the collision objects. The iterative optimization module is used to perform an improved cloth simulation iterative process on the traditional cloth simulation model. During the iteration process, the hardness value of the cloth particles is adaptively adjusted based on the gradient feature values ​​of the points in the corresponding collision objects, until the parameter combination of the traditional cloth simulation model reaches its optimal value, thus obtaining a stationary cloth model. The classification module is used to generate an approximate terrain surface based on the stationary cloth model; calculate the distance from each point in the preprocessed point cloud data to the approximate terrain surface; and classify each point in the preprocessed point cloud data based on a distance threshold.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images