A bridge point cloud extraction method based on evolutionary gradient search

Abstract

The present invention discloses a bridge point cloud extraction method based on evolutionary gradient search, which relates to the field of data processing technology. The bridge deck point cloud is denoised by using a clustering algorithm; the denoised bridge deck point cloud is initially extracted based on multi-threading and minimum heap data structure, and vertical density constraints are set to further extract the bridge deck point cloud; the gradient-based optimization algorithm is combined with the evolutionary strategy, and incorporated into a unified framework as a complementary algorithm, and the extracted bridge deck point cloud is optimized by alternately using gradient update and evolutionary update methods, thereby obtaining the final extracted bridge deck point cloud. The present invention improves the processing accuracy of the bridge deck point cloud, and by introducing the evolutionary gradient search strategy, it can more effectively identify and separate the main structure of the bridge in complex point cloud data, while significantly reducing the interference of noise and redundant information on the final result.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a bridge point cloud extraction method based on evolutionary gradient search. Background Art

[0002] With the rapid development of 3D laser scanning technology, its application in bridge engineering, urban planning, traffic management and other fields is becoming more and more extensive. 3D laser scanning technology can efficiently obtain 3D point cloud data of bridges, which is of great value for the design, construction, maintenance and traffic flow analysis of bridges. However, in practical applications, the point cloud data of bridges often contains a lot of noise and redundant information, such as non-bridge deck elements such as vehicles and pedestrians, as well as errors caused by scanning equipment and environmental factors. These noise and redundant information will seriously affect the accuracy and reliability of point cloud data, and thus affect the subsequent bridge engineering design and traffic management decisions.

[0003] The current mainstream point cloud filtering algorithms have shown certain effects when processing bridge point cloud data, but they also have their own limitations. The filtering algorithm based on slope fitting can better identify the slope changes of the bridge deck in the processing of bridge point clouds, but it may encounter difficulties when encountering complex bridge structures or mutation points (such as piers, supports, etc.), resulting in poor filtering effects. The filtering algorithm based on mathematical morphology extracts key features from bridge point clouds by setting specific structural elements, but the selection of structural elements often depends on experience, and is more sensitive to noise in bridge point clouds, and it is easy to mistakenly delete or retain erroneous point cloud data. The irregular triangulated network filtering algorithm simulates the morphology of the bridge surface by constructing a triangulated network. Although it can handle irregular data distribution in bridge point clouds, it may produce errors when processing bridge fractures, deformations or mutation areas, affecting the accuracy of filtering. The cloth simulation filtering algorithm simulates the natural sagging process of cloth under the action of gravity to distinguish the bridge body from the surrounding non-bridge point cloud. However, this method has a large amount of calculation and low processing efficiency for large bridge point cloud data. It may not be effective for certain specific scenes (such as high-hanging bridges, steep bridge towers, etc.), and it is difficult to accurately distinguish the bridge body from background noise.

[0004] In summary, traditional filtering algorithms face the challenge of being unable to accurately identify the true structural features of bridges when processing bridge point cloud data rich in noise and redundant information. The mixture of noise and redundant information will seriously interfere with the normal operation of the algorithm, resulting in inaccurate or biased extracted bridge structure information. Summary of the invention

[0005] In order to overcome the defects in the above-mentioned prior art, the present invention provides a bridge point cloud extraction method based on evolutionary gradient search, which improves the efficiency and accuracy of bridge deck point cloud processing and provides more reliable data support for bridge engineering design and traffic management.

[0006] To achieve the above object, the present invention adopts the following technical solutions, including:

[0007] A bridge point cloud extraction method based on evolutionary gradient search, comprising:

[0008] S1, denoising the bridge deck point cloud;

[0009] S2, based on multi-threading and minimum heap data structure, performs preliminary extraction of the denoised bridge deck point cloud, and sets vertical density constraints to further extract the bridge deck point cloud;

[0010] S3, combines the gradient-based optimization algorithm with the evolutionary strategy to optimize the extracted bridge deck point cloud and obtain the final extracted bridge deck point cloud.

[0011] Preferably, the specific process of step S3 is as follows:

[0012] S31, set algorithm parameters: maximum number of iterations, initial step size and gradient estimation step size;

[0013] S32, define the fitness function F(C):

[0014] ;

[0015] Among them, the point cloud C contains n points, Z i is the elevation of the i-th point, i=1,2,...,n; Zv is the average elevation of all points in the point cloud; the optimization goal is to minimize F(C);

[0016] S33, population initialization: the initial population is the bridge deck point cloud extracted in step S2, and each individual is represented by the coordinates of each point;

[0017] S34: Estimate gradient: Use the finite difference method to perturb the elevation of each point and calculate the fitness difference before and after the perturbation to estimate the gradient;

[0018] S35: Update point cloud: update the elevation of each point along the negative gradient direction;

[0019] S36, update step size: if the fitness after the update is better than the fitness before the update, increase the step size of the point cloud update; otherwise, reduce the step size of the point cloud update;

[0020] S37: Check termination conditions: If the step size is less than the set value, or the fitness does not improve after several consecutive iterations, or the maximum number of iterations is reached, the iteration is stopped and the final extracted bridge deck point cloud is output.

[0021] Preferably, in step S34, the gradient calculation formula is:

[0022] ;

[0023] in, F(Z i t ) is the gradient of the i-th point in the t-th iteration; ε is the set gradient estimation step size, ε>0; Z i t is the elevation of the i-th point in the t-th iteration.

[0024] Preferably, in step S35, the elevation update formula of the point is:

[0025] ;

[0026] in, F(Z i t ) is the gradient of the i-th point in the t-th iteration; t+1 is the step size updated along the negative gradient direction in the t+1th iteration; Z i t is the elevation of the i-th point in the t-th iteration; Z i t+1 is the elevation of the i-th point in the t+1-th iteration.

[0027] Preferably, in step S36, the step length update formula is:

[0028] ;

[0029] in, t+1 is the step size updated along the negative gradient direction in the t+1th iteration; t is the step size updated along the negative gradient direction in the tth iteration; 1 is the initial step size set; Z i t is the elevation of the i-th point in the t-th iteration; ξ is the step update coefficient, ξ>1; F(Z i t ) is the gradient of the i-th point in the t-th iteration.

[0030] Preferably, in step S2, the denoised bridge deck point cloud is initially extracted based on multi-threading and minimum heap data structure, as shown below:

[0031] S21, construct a minimum heap priority queue according to the elevation of the point;

[0032] S22, using KD tree to perform neighbor search and obtain a neighborhood point set;

[0033] S23, expand the local minimum method to initially extract the bridge deck point cloud:

[0034] Find the lowest elevation point in the neighborhood, and determine whether the elevation difference between the lowest elevation point and the adjacent points is greater than the preset elevation threshold. If it is greater than the elevation threshold, delete the lowest elevation point, and define the second lowest elevation point as the new lowest elevation point, and continue to determine whether the elevation difference between the new lowest elevation point and the adjacent points is greater than the elevation threshold; if it is less than or equal to the elevation threshold, retain the lowest elevation point, and determine the next point, i.e. the second lowest elevation point, and continue to determine whether the elevation difference between the next point and the adjacent points is greater than the preset elevation threshold; repeat the iteration to preliminarily extract the bridge deck point cloud.

[0035] Preferably, in step S2, vertical density constraints are set to further extract the bridge deck point cloud, as shown below:

[0036] S24, setting a verticality density constraint as an additional condition, deleting points whose verticality is less than a verticality threshold, thereby further extracting the bridge deck point cloud; wherein the verticality of a point is the absolute value of the projection of the normal vector of the point on the Z axis.

[0037] Preferably, in step S1, a clustering algorithm is used to denoise the bridge deck point cloud; if the number of neighboring points of a point in the point cloud within the neighborhood radius is less than the minimum number of neighboring points, the point is regarded as a noise point and deleted.

[0038] Preferably, the neighborhood radius and the minimum number of neighboring points are dynamically adjusted by an adaptive method.

[0039] The present invention also provides a computer program product, which includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the above-mentioned bridge point cloud extraction method based on evolutionary gradient search is implemented.

[0040] The advantages of the present invention are:

[0041] (1) The present invention improves the processing accuracy of bridge deck point clouds. By introducing an evolutionary gradient search strategy, it can more effectively identify and separate the main structure of the bridge in complex point cloud data, while significantly reducing the interference of noise and redundant information on the final result.

[0042] (2) This paper proposes a unified optimization framework that combines the gradient optimization algorithm and the evolutionary strategy, aiming to extract bridge point cloud data efficiently and with high precision. This framework cleverly combines the fast local convergence characteristics of the gradient optimization algorithm and the global search advantages of the evolutionary strategy, thereby overcoming the limitations of traditional methods in processing complex point cloud data. Through this framework, not only can the key features in the point cloud data be quickly located and optimized, but it can also effectively avoid falling into the local optimal solution, ensuring that the final extracted bridge point cloud data is both accurate and reliable.

[0043] (3) The present invention introduces an evolutionary strategy to further optimize the point cloud data after preliminary extraction. The evolutionary strategy simulates biological evolution processes such as natural selection and genetic variation to perform a global search on the point cloud data to find potential optimal solutions. The specific workflow is: in each iteration, the parameter vector (elevation) of the population individual is randomly perturbed, and the objective function (fitness function) is evaluated for the new population individual. For individuals with higher objective function values, their parameter vectors are recombined to generate parameter vectors for the next generation of population individuals. The above steps are repeated multiple times until the target optimization is completed. The evolutionary strategy has a strong global search capability and can jump out of the local optimal solution. The present invention can achieve efficient and high-precision extraction of bridge point cloud data, improve the efficiency of point cloud data processing, and provide more reliable data support for bridge engineering design and traffic management.

[0044] (4) Search for neighboring points through the KD tree, build a minimum heap priority queue, determine whether the elevation difference between the current point and the adjacent point exceeds the preset elevation threshold, and iteratively adjust the lowest elevation point until the condition is met. At the same time, the vertical density constraint is introduced to optimize the selection of points to reflect the local elevation characteristics and avoid misleading caused by drastic changes in terrain.

[0045] (5) The present invention provides strong technical support for fields such as bridge engineering design and traffic management, and has broad application prospects and important practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of a bridge point cloud extraction method based on evolutionary gradient search according to an embodiment of the present invention.

[0047] Figure 2 This is an example diagram of the original bridge point cloud according to an embodiment of the present invention.

[0048] Figure 3 This is an example diagram of the bridge deck point cloud finally extracted by the embodiment of the present invention.

[0049] Figure 4 This is an example diagram of a non-bridge deck point cloud according to an embodiment of the present invention.

[0050] Figure 5 2 is an example diagram of bridge modeling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Depend on Figure 1 As shown, a bridge point cloud extraction method based on evolutionary gradient search includes the following steps:

[0053] S1, the DBSCAN clustering algorithm is used to denoise the bridge deck point cloud obtained by the lidar to obtain purer bridge point cloud data.

[0054] In step S1, Visual Studio software is used to run the DBSCAN clustering algorithm, and the neighborhood radius and the minimum number of neighboring points are dynamically adjusted through an adaptive method. In this embodiment, after a large number of attempts at different combinations of neighborhood radius and minimum number of neighboring points, the neighborhood radius is finally determined to be 0.5 and the minimum number of neighboring points is 5. This combination can achieve the best effect of point cloud denoising. If the number of neighboring points of a point in the point cloud within the neighborhood radius is less than the minimum number of neighboring points, the point is regarded as a noise point and deleted.

[0055] S2, based on multi-threading and minimum heap data structure, performs preliminary extraction of bridge deck point cloud, and sets vertical density constraints to further extract bridge deck point cloud.

[0056] The specific process of step S2 is as follows:

[0057] S21, construct a minimum heap priority queue according to the elevation of the points, and sort the points from small to large according to their elevation.

[0058] S22, use the KD tree to perform a neighbor search to obtain a neighborhood point set.

[0059] S23, expand the local minimum method to initially extract the bridge deck point cloud:

[0060] Find the lowest elevation point in the neighborhood, and determine whether the elevation difference between the lowest elevation point and the adjacent points is greater than the preset elevation threshold. If it is greater than the elevation threshold, delete the lowest elevation point, and define the second lowest elevation point as the new lowest elevation point, and continue to determine whether the elevation difference between the new lowest elevation point and the adjacent points is greater than the preset elevation threshold; if it is less than or equal to the elevation threshold, retain the lowest elevation point, and determine the next point, i.e. the second lowest elevation point, and continue to determine whether the elevation difference between the next point and the adjacent points is greater than the preset elevation threshold; repeat the iteration to preliminarily extract the bridge deck point cloud.

[0061] S24, setting the verticality density constraint as an additional condition, deleting points whose verticality is less than the verticality threshold, so as to further extract the bridge deck point cloud.

[0062] The verticality Nv of a point is the absolute value of the projection of the normal vector of the point on the Z axis. This value is marked as Nv, and the value range of Nv is limited to between 0 and 1, that is, Nv∈(0,1).

[0063] Ideally, if the terrain represented by the point cloud is completely flat, the normal vectors of all points point in the direction of the Z axis or in the opposite direction, and Nv is close to 1 or -1. On the contrary, in areas with drastic slope changes, the directions of the normal vectors are more dispersed, resulting in larger changes in the value of Nv. That is, when the vertical difference between adjacent points is small, it indicates that the terrain slope changes gently in this area; and when the vertical difference between adjacent points is large, it indicates that the terrain slope changes more drastically.

[0064] In this embodiment, the PCL library is used to set the verticality threshold, and the points whose verticality Nv is less than the verticality threshold are deleted. The verticality threshold is set between 0.5 and 0.7.

[0065] Through the KD tree, we search for neighboring points, build a minimum heap priority queue, determine whether the elevation difference between the current point and the adjacent point exceeds the preset elevation threshold, and iteratively adjust the lowest elevation point until the condition is met. At the same time, we introduce the vertical density constraint, that is, Nv∈(0,1), to optimize the selection of points, reflect the local elevation characteristics, and avoid misleading caused by drastic changes in terrain.

[0066] S3, combines the gradient-based optimization algorithm with the evolutionary strategy as complementary algorithms into a unified framework, and optimizes the bridge deck point cloud extracted in step S2 by alternating between gradient updating and evolutionary updating methods, thereby obtaining the final extracted bridge deck point cloud.

[0067] The specific process of step S3 is as follows:

[0068] S31, setting algorithm parameters, including:

[0069] Maximum number of iterations: controls the maximum number of steps of the algorithm. The maximum number of iterations determines the running time of the algorithm and the accuracy of the final result. The initial value is set to 200.

[0070] Initial step size: The initial step size determines the magnitude of the update along the gradient direction in the first iteration, and the initial value is 0.1.

[0071] Gradient estimation step size: Small step size used in estimating the gradient using the finite difference method.

[0072] S32, define a fitness function F(C).

[0073] Given a point cloud C, containing n points, the height value of each point is Z i , the subscript i represents the i-th point, i=1,2,...,n; the fitness function F(C) is the mean square error (MSE) of the elevation of all points, and the elevation Z of all points in the point cloud is calculated i The average of the sum of squares of the differences from the mean elevation Zv:

[0074] ;

[0075] The smaller the value of the fitness function F(C), the more concentrated (i.e., flatter) the height distribution of the point cloud is, so the optimization goal is to minimize F(C).

[0076] S33, population initialization: the initial population is the bridge deck point cloud initially extracted in step S2, each point in the bridge deck point cloud is regarded as an individual, and each individual is represented as the coordinates of the point.

[0077] S34, estimate gradient: use the finite difference method to perturb the elevation of each point, and calculate the fitness difference before and after the perturbation to estimate the gradient, where the gradient is calculated only in the z direction.

[0078] S35, update point cloud: update the elevation of each point along the negative gradient direction to reduce the value of the fitness function.

[0079] S36, update step size: if the fitness after update is better than the fitness before update, increase the step size to speed up convergence; otherwise, reduce the step size.

[0080] S37: Check termination conditions: If the step length is less than the set value (1×10 in this embodiment), -6 ), or if the fitness does not improve after several consecutive iterations, or if the maximum number of iterations is reached, the iteration is stopped and the final extracted bridge deck point cloud is output.

[0081] The fastest descent method is a specific implementation of the gradient method. Its idea is to determine the gradient direction in each iteration and then choose a suitable step size so that the fitness function value can be minimized.

[0082] The formula for updating the elevation of a point is:

[0083] ;

[0084] Among them, Z i t is the elevation of the i-th point in the t-th iteration; Z i t+1 is the elevation of the i-th point in the t+1-th iteration;

[0085] t+1 is the step size updated along the negative gradient direction in the t+1th iteration; F(Z i t ) is the gradient of the i-th point in the t-th iteration;

[0086] F(Z i t ) indicates that Z i t Substitute into the fitness function for calculation; Indicates derivative.

[0087] The gradient is approximated using the forward difference formula instead of a strict limit definition. The gradient calculation formula is:

[0088] ;

[0089] Among them, ε is the gradient estimation step size set, ε is a small positive number, set to 1×10 -5 .

[0090] A simple line search method is used to obtain the step size for updating along the negative gradient direction:

[0091] ;

[0092] Where ξ is the step size update coefficient, ξ>1; 1 is the initial step size set; t is the step size for updating along the negative gradient direction in the tth iteration.

[0093] Given an optimization problem, the gradient evolution strategy can work with a large number of candidate solutions as an individual. Each individual represents a set of model parameters to be optimized by the optimizer. In each generation of evolution, the evolution strategy and stochastic gradient descent are used alternately for optimization in a staged manner. Considering each individual in the population as a species, each species evolves independently in the gradient update method during any generation, and then interacts with each other in the evolution update method. The gradient evolution strategy integrates the evolution strategy and stochastic gradient descent into a complementary optimization strategy.

[0094] In this embodiment, in order to quantitatively evaluate the accuracy of bridge deck point cloud extraction, the elevations of the actually measured control points and the bridge deck points extracted by the algorithm are compared.

[0095] The measured bridge surface points are compared with the bridge surface points extracted by the algorithm, and the accuracy of the extracted bridge surface points is analyzed. The specific process is as follows:

[0096] The mean elevation error and mean absolute elevation error are calculated according to the relevant technical requirements in the national standard "Quality Inspection and Acceptance of Digital Surveying and Mapping Results" (GB / T 181316-2008) and the surveying and mapping industry standard "Technical Specifications for Production of Basic Geographic Information Digital Results 1:500, 1:1000, 1:2000 (Part 2): Digital Elevation Model" (CH / T 9020.2-2013).

[0097] Since manual calculation is time-consuming, elevation error calculation is performed based on AutoCAD secondary development. IDW (Inverse Distance Weighted) is a commonly used and simple spatial interpolation method, which uses the distance between the interpolation point and the sample point as the weight for weighted averaging. The closer the sample point is to the interpolation point, the greater the weight is.

[0098] On the basis of ensuring elevation accuracy, the IDW spatial interpolation algorithm is used to generate the bridge deck point cloud DEM (digital elevation model). The specific process of bridge deck point cloud DEM modeling is as follows:

[0099] First, data preprocessing is performed. Before DEM modeling, the extracted bridge deck point cloud needs to be preprocessed. This includes removing noise points, filling data holes, and performing point cloud filtering to ensure the accuracy and integrity of the point cloud data. Through the preprocessing steps, errors in the subsequent modeling process can be reduced and the accuracy of the DEM can be improved.

[0100] Then the point cloud is gridded to convert the preprocessed bridge deck point cloud data into regular grid data. This is usually achieved through the inverse distance weighted interpolation method.

[0101] Finally, the bridge deck point cloud DEM is constructed, and the initial triangulated network is constructed using the obtained bridge deck seed points. The judgment criteria for unclassified points in the iterative encryption process are the same as those in the classical triangulated network algorithm. The angle θ and distance d from the unclassified point to the corresponding triangulated network are calculated. When θ and d are less than the angle threshold and distance threshold, respectively, the point is added to the bridge deck point set and the triangulated network is encrypted.

[0102] in, Figure 2 This is an example diagram of the original bridge point cloud according to an embodiment of the present invention. Figure 3 This is an example diagram of the bridge deck point cloud finally extracted by the embodiment of the present invention. Figure 4 This is an example diagram of a non-bridge deck point cloud according to an embodiment of the present invention. Figure 5 2 is an example diagram of bridge modeling according to an embodiment of the present invention.

[0103] In this embodiment, a unified optimization framework combining a gradient optimization algorithm and an evolutionary strategy is proposed, aiming at extracting bridge point cloud data efficiently and with high precision. This framework cleverly combines the fast local convergence characteristics of the gradient optimization algorithm and the global search advantages of the evolutionary strategy, thereby overcoming the limitations faced by traditional methods in processing complex point cloud data. Through this framework, not only can the key features in the point cloud data be quickly located and optimized, but it can also effectively avoid falling into the local optimal solution, ensuring that the final extracted bridge point cloud data is both accurate and reliable. In summary, the framework proposed in this embodiment provides strong technical support for fields such as bridge engineering design and traffic management, and has broad application prospects and important practical significance.

[0104] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A bridge point cloud extraction method based on evolutionary gradient search, characterized in that: include: S1, denoising the bridge deck point cloud; S2, based on multi-threading and minimum heap data structure, performs preliminary extraction of the denoised bridge deck point cloud, and sets vertical density constraints to further extract the bridge deck point cloud; S3, combining the gradient-based optimization algorithm with the evolutionary strategy to optimize the extracted bridge deck point cloud to obtain the final extracted bridge deck point cloud; The specific process of step S3 is as follows: S31, set algorithm parameters: maximum number of iterations, initial step size and gradient estimation step size; S32, define fitness function F(C); ； Among them, the point cloud C contains n points, Z i is the elevation of the i-th point, i=1,2,...,n; Zv is the average elevation of all points in the point cloud; the optimization goal is to minimize F(C); S33, population initialization: the initial population is the bridge deck point cloud extracted in step S2, and each individual is represented by the coordinates of each point; S34: Estimate gradient: Use the finite difference method to perturb the elevation of each point and calculate the fitness difference before and after the perturbation to estimate the gradient; S35: Update point cloud: update the elevation of each point along the negative gradient direction; S36, update step size: if the fitness after the update is better than the fitness before the update, increase the step size of the point cloud update; otherwise, reduce the step size of the point cloud update; S37: Check termination conditions: If the step size is less than the set value, or the fitness does not improve after several consecutive iterations, or the maximum number of iterations is reached, the iteration is stopped and the final extracted bridge deck point cloud is output.

2. The bridge point cloud extraction method based on evolutionary gradient search according to claim 1 is characterized in that: In step S34, the gradient calculation formula is: ； Among them, ∇F(Z i t ) is the gradient of the i-th point in the t-th iteration; ε is the set gradient estimation step size, ε>0; Z i t is the elevation of the i-th point in the t-th iteration.

3. The bridge point cloud extraction method based on evolutionary gradient search according to claim 1 is characterized in that: In step S35, the elevation update formula of the point is: ； Among them, ∇F(Z i t ) is the gradient of the i-th point in the t-th iteration; t+1 is the step size updated along the negative gradient direction in the t+1th iteration; Z i t is the elevation of the i-th point in the t-th iteration; Z i t+1 is the elevation of the i-th point in the t+1-th iteration.

4. The bridge point cloud extraction method based on evolutionary gradient search according to claim 1 is characterized in that: In step S36, the step length update formula is: ； Among them, t+1 is the step size updated along the negative gradient direction in the t+1th iteration; t is the step size updated along the negative gradient direction in the tth iteration; 1 is the initial step size set; Z i t is the elevation of the ith point in the tth iteration; ξ is the step update coefficient, ξ>1; ∇F(Z i t ) is the gradient of the i-th point in the t-th iteration.

5. The bridge point cloud extraction method based on evolutionary gradient search according to claim 1 is characterized in that: In step S2, the denoised bridge deck point cloud is initially extracted based on multi-threading and minimum heap data structure, as shown below: S21, construct a minimum heap priority queue according to the elevation of the point; S22, using KD tree to perform neighbor search and obtain a neighborhood point set; S23, expand the local minimum method to initially extract the bridge deck point cloud: Find the lowest elevation point in the neighborhood, and determine whether the elevation difference between the lowest elevation point and the adjacent point is greater than the preset elevation threshold. If it is greater than the elevation threshold, delete the lowest elevation point, and define the second lowest elevation point as the new lowest elevation point, and continue to determine whether the elevation difference between the new lowest elevation point and the adjacent point is greater than the elevation threshold; if it is less than or equal to the elevation threshold, retain the lowest elevation point, and determine the next point, i.e. the second lowest elevation point, and continue to determine whether the elevation difference between the next point and the adjacent point is greater than the preset elevation threshold; Repeat the iterations to preliminarily extract the bridge deck point cloud.

6. The bridge point cloud extraction method based on evolutionary gradient search according to claim 5 is characterized in that: In step S2, vertical density constraints are set to further extract the bridge deck point cloud, as shown below: S24, setting a verticality density constraint as an additional condition, deleting points whose verticality is less than a verticality threshold, thereby further extracting the bridge deck point cloud; wherein the verticality of a point is the absolute value of the projection of the normal vector of the point on the Z axis.

7. The bridge point cloud extraction method based on evolutionary gradient search according to claim 1 is characterized in that: In step S1, a clustering algorithm is used to denoise the bridge deck point cloud; if the number of neighboring points of a point in the point cloud within the neighborhood radius is less than the minimum number of neighboring points, the point is regarded as a noise point and deleted.

8. The bridge point cloud extraction method based on evolutionary gradient search according to claim 7 is characterized in that: The neighborhood radius and the minimum number of neighboring points are dynamically adjusted through an adaptive method.

9. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements a bridge point cloud extraction method based on evolutionary gradient search as described in any one of claims 1 to 8.

Images