A Point Cloud Boundary Extraction Method Based on Virtual Support Release Disturbance Response

Abstract

This invention discloses a point cloud boundary extraction method based on virtual support release perturbation response, belonging to the field of 3D point cloud data processing technology. It is used for boundary extraction of 3D point cloud data, including acquiring raw point cloud data, constructing local structural units and the structural support relationships between them; applying virtual support perturbation release to each local structural unit, calculating the support release response information of each local structural unit based on the perturbation process changes, constructing a structural instability response field, and calculating the structural instability transition degree between adjacent local structural units; constructing an objective function, locating the structural boundary based on the structural instability transition degree, filtering local structural units located at the structural boundary through the objective function, constructing a set of candidate local structural units for the boundary, and outputting the point cloud boundary extraction result. This invention introduces a structural recovery step count and constructs a response discrimination mechanism, improving the accuracy, stability, and interpretability of point cloud boundary extraction.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional point cloud processing technology, and more particularly to the field of three-dimensional point cloud boundary extraction technology, specifically a point cloud boundary extraction method based on virtual support release disturbance response. Background Technology

[0002] Point cloud boundary extraction, as a fundamental processing step in point cloud understanding, aims to identify object contour boundaries, structural boundaries, topological edges, or surface abrupt changes from the original point cloud, so as to provide a basis for subsequent processing such as target recognition, contour localization, structural analysis, defect detection, segmentation constraints, or model reconstruction.

[0003] Currently, point cloud boundary extraction methods can be mainly divided into the following three categories: The first category consists of methods based on local geometric features, such as using curvature, normal vectors, distance between points, and surface smoothness for region growing, clustering, or boundary extraction. These methods are relatively straightforward to implement, but they are sensitive to noise, local missing features, and non-uniform sampling. In complex connected regions, slender structural regions, occluded boundary regions, and thin sheet edge regions, boundary omissions, boundary drift, or local breaks are prone to occur.

[0004] The second category comprises graph model-based methods, such as boundary location, region separation, or structural partitioning based on connectivity, similarity, or cost relationships between adjacent points. While these methods can utilize inter-point relationships to some extent, they typically rely on pre-defined connection weights, similarity functions, or partitioning criteria. The representation of structural boundaries remains primarily based on static relationships, making it difficult to effectively characterize the dynamic changes in local structural stability during disturbance propagation. Existing graph model methods also lack a technical approach that indirectly characterizes structural boundaries by altering the contribution of local structural units to the overall structural stability.

[0005] The third category consists of deep learning-based methods, such as boundary extraction through point feature learning, boundary point classification, contour saliency estimation, or edge response prediction. These methods perform well when labeled data is plentiful, but they rely on training samples, limiting model transferability and interpretability. Furthermore, the stability of boundary extraction may decrease when sampling density, sensor type, or target structure changes.

[0006] Most existing technologies rely on the geometric features, adjacency relationships, or statistical similarity of point clouds under the current observation state for boundary determination. A common drawback is that they analyze only the current static state of the structure, lacking modeling of the response behavior of local structural units after disturbance. For geometrically close locations with true structural boundaries, static features alone are insufficient for differentiation, easily leading to boundary omissions, drift, or breakage. Specifically, existing technologies typically only consider "which points are close to each other" or "which points appear similar in the current state," without addressing the mechanism of altering the contribution of local structural units to the overall structural stability, or indirectly characterizing structural boundaries based on the response propagation and recovery behavior after disturbance. Therefore, existing technologies struggle to accurately identify true boundaries from the perspective of structural response.

[0007] Therefore, there is an urgent need for a method to extract point cloud structural boundaries by simulating the response process after the support of local structural units on the stability of surrounding structures, in order to improve the accuracy, stability and interpretability of point cloud boundary extraction in complex scenes. Summary of the Invention

[0008] The purpose of this invention is to provide a point cloud boundary extraction method based on virtual support release disturbance response, so as to solve the problem that the existing technology only relies on static geometric features or static relationships for boundary judgment, lacks modeling of the response behavior after disturbance, and causes point cloud boundary extraction to be easily affected by noise, uneven sampling and complex structure, resulting in omission, drift or breakage.

[0009] To address the above objectives, this invention provides a point cloud boundary extraction method based on virtual support release perturbation response, comprising: S1. Obtain raw point cloud data, select neighboring points from the raw point cloud data using a neighborhood search algorithm, and construct multiple point cloud neighborhoods; S2. Establish the directed structural support relationship between any two points in the point cloud neighborhood, construct the local structural units corresponding to each point cloud neighborhood, and construct the structural support relationship between the local structural units based on the support response function between the local structural units. S3. Apply virtual support disturbance release to each local structural unit and obtain the disturbance process changes of each local structural unit. The disturbance process changes of the local structural unit include changes in propagation state, changes in connectivity state, and changes in surface continuity. S4. Based on the changes in the disturbance process of each local structural unit, calculate the support release response information of each local structural unit. The support release response information includes the output of the support release response function, the number of structural recovery steps, and the main direction of disturbance propagation. S5. Based on the support release response information of each local structural unit, construct the structural instability response field and calculate the structural instability transition degree between adjacent local structural units; S6. Construct an objective function, locate the structural boundary based on the structural instability transition degree, filter local structural elements located at the structural boundary through the objective function, construct a set of candidate local structural elements for the boundary, and output the point cloud boundary extraction results.

[0010] In S2, the support response function between local structural elements is determined based on the spatial distance, normal difference, curvature transition degree, and surface extension direction consistency between them. ; In the formula, Represents local structural units Pointing to local structural units The support response function value, Represents local structural units With local structural units The distance correlation term is used to characterize the spatial distance between local structural units. Represents local structural units With local structural units The normal consistency term is used to characterize the normal difference between local structural units. Represents local structural units With local structural units The curvature transition term is used to characterize the degree of curvature transition between local structural units. Represents local structural units With local structural units The surface extension consistency term is used to characterize the consistency of the surface extension direction between local structural units. , , , They represent , , , coefficient, and This is an index for local structural units.

[0011] In S3, virtual support disturbance release is applied to each local structural unit, including adjusting the support response function between local structural units through the support release disturbance operator, and changing the support contribution participation mode in the structural propagation process so that the disturbance effect spreads in the local structure in the form of propagation. The support release perturbation operator is: ; In the formula, Indicates to The support response function value after applying virtual support to release the disturbance. Represents local structural units The support release coefficient.

[0012] The characteristic feature is that the output of the support release response function in S4 is: ; In the formula, Represents local structural units The support release response function output, Represents local structural units The change in propagation path before and after applying virtual support to release the disturbance. Represents local structural units The change in connectivity stability before and after applying virtual support to release the disturbance. Represents local structural units The change in surface continuity before and after applying virtual support to release the disturbance. Indicates the difference. , , They represent , , The coefficient.

[0013] The structural recovery step count is used to characterize the total number of iterations required for a local structural element to recover from its state after the virtual support release disturbance is applied to a stable state. During the iteration process, when the local structural unit The local structural unit is determined when the response state satisfies the iterative stability criterion for m consecutive times. Reaching a stable state; The iterative stability criterion is: ; In the formula, Represents local structural units In the The response state vector during round propagation. Represents local structural units In the The response state vector during round propagation. Represents the norm, This represents the stability threshold. Indicates the iteration round index; For a local structural unit that has reached a stable state, the number of structural recovery steps is the iteration cycle index of the last iteration when the iterative stability criterion is met for m consecutive iterations; for a local structural unit that has not reached a stable state within the preset maximum number of iterations, the number of structural recovery steps is the maximum number of iterations.

[0014] The structural instability transition degree is: ; In the formula, Represents local structural units With local structural units The degree of structural instability transitions between them Represents local structural units The support release response function output, Represents local structural units The main direction of disturbance propagation, Represents local structural units The main direction of disturbance propagation, Represents local structural units The number of structural recovery steps, Represents local structural units The number of structural recovery steps, This represents the perturbation direction coefficient. This represents the structural recovery step coefficient.

[0015] The objective function of S6 is: ; In the formula, Describes the evaluation function for the boundary candidate set. The function representing the joint degree of discreteness, Represents the set of candidate local structural elements at the boundary. Indicates the index of the boundary connected subset. Represents a set The total number of boundary-connected subsets, Indicates the candidate boundary weights. Indicates local structural units With local structural units The adjacent local structural unit pairs formed Represents a set The set of connection relationships between adjacent local structural unit pairs.

[0016] Compared with the prior art, the present invention has the following advantages: Based on the geometric similarity or spatial proximity of point clouds in the current state, this invention establishes a directional structural support relationship. Based on the response transmission mechanism, it simulates the response process after the support of local structural units is released by applying virtual supports to release disturbances. The point cloud boundary is judged from the level of response behavior, which improves the accuracy of point cloud boundary extraction. This invention introduces the structural recovery step count as a temporal response parameter. Within the internal regions of a structure, abundant support paths allow disturbances to dissipate quickly through multiple paths, resulting in a smaller recovery step count. However, in boundary or elongated connection regions, limited support paths hinder disturbance propagation, prolonging the recovery process and significantly increasing the recovery step count. This parameter not only describes the response state but also forms the core of the structural discrimination mechanism, significantly improving the ability to identify point cloud boundaries in complex connection regions, elongated structural regions, and unstable boundary regions. This invention does not rely on labeled samples for supervised training, has good interpretability and generalization ability, and can be stably applied to various 3D perception scenarios such as autonomous driving point clouds, industrial scanning point clouds, and indoor reconstruction point clouds. It effectively solves the problems of boundary omission, boundary drift or local breakage that are prone to occur in existing technologies under conditions such as noise, non-uniform sampling and occlusion boundaries, and significantly improves the accuracy, stability and interpretability of point cloud boundary extraction. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall process of the point cloud boundary extraction method provided by the present invention;

[0018] Figure 2 The original point cloud reference data map provided for this invention;

[0019] Figure 3 The application method A provided by the present invention is for... Figure 2 The result of boundary extraction;

[0020] Figure 4 The application method B provided by the present invention is for... Figure 2 The result of boundary extraction;

[0021] Figure 5 The application method C provided by the present invention is for... Figure 2 The result of boundary extraction;

[0022] Figure 6 To apply the point cloud boundary extraction method provided by this invention to... Figure 2 The result of boundary extraction. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] Example 1 A point cloud boundary extraction method based on virtual support release perturbation response includes: S1. Obtain raw point cloud data, select neighboring points from the raw point cloud data using a neighborhood search algorithm, and construct multiple point cloud neighborhoods; S2. Establish the directed structural support relationship between any two points in the point cloud neighborhood, construct the local structural units corresponding to each point cloud neighborhood, and construct the structural support relationship between the local structural units based on the support response function between the local structural units. S3. Apply virtual support disturbance release to each local structural unit and obtain the disturbance process changes of each local structural unit. The disturbance process changes of the local structural unit include changes in propagation state, changes in connectivity state, and changes in surface continuity. S4. Based on the changes in the disturbance process of each local structural unit, calculate the support release response information of each local structural unit. The support release response information includes the output of the support release response function, the number of structural recovery steps, and the main direction of disturbance propagation. S5. Based on the support release response information of each local structural unit, construct the structural instability response field and calculate the structural instability transition degree between adjacent local structural units; S6. Construct an objective function, locate the structural boundary based on the structural instability transition degree, filter local structural elements located at the structural boundary through the objective function, construct a set of candidate local structural elements for the boundary, and output the point cloud boundary extraction results.

[0025] In S2, the support response function between local structural elements is determined based on the spatial distance, normal difference, curvature transition degree, and surface extension direction consistency between them. ; In the formula, Represents local structural units Pointing to local structural units The support response function value, Represents local structural units With local structural units The distance correlation term is used to characterize the spatial distance between local structural units. Represents local structural units With local structural units The normal consistency term is used to characterize the normal difference between local structural units. Represents local structural units With local structural units The curvature transition term is used to characterize the degree of curvature transition between local structural units. Represents local structural units With local structural units The surface extension consistency term is used to characterize the consistency of the surface extension direction between local structural units. , , , They represent , , , coefficient, and For local structural unit indexes, and They are not equal.

[0026] In S3, virtual support disturbance release is applied to each local structural unit, including adjusting the support response function between local structural units through the support release disturbance operator, and changing the support contribution participation mode in the structural propagation process so that the disturbance effect spreads in the local structure in the form of propagation. The support release perturbation operator is: ; In the formula, Indicates to The support response function value after applying virtual support to release the disturbance. Represents local structural units The support release coefficient.

[0027] The characteristic feature is that the output of the support release response function in S4 is: ; In the formula, Represents local structural units The support release response function output, Represents local structural units The change in propagation path before and after applying virtual support to release the disturbance. Represents local structural units The change in connectivity stability before and after applying virtual support to release the disturbance. Represents local structural units The change in surface continuity before and after applying virtual support to release the disturbance. Indicates the difference. , , They represent , , The coefficient.

[0028] The structural recovery step count is used to characterize the total number of iterations required for a local structural element to recover from its state after the virtual support release disturbance is applied to a stable state. During the iteration process, when the local structural unit The local structural unit is determined when the response state satisfies the iterative stability criterion for m consecutive times. Reaching a stable state; The iterative stability criterion is: ; In the formula, Represents local structural units In the The response state vector during round propagation. Represents local structural units In the The response state vector during round propagation. Represents the norm, This represents the stability threshold. Indicates the iteration round index; For a local structural unit that has reached a stable state, the number of structural recovery steps is the iteration cycle index of the last iteration when the iterative stability criterion is met for m consecutive iterations; for a local structural unit that has not reached a stable state within the preset maximum number of iterations, the number of structural recovery steps is the maximum number of iterations.

[0029] The structural instability transition degree is: ; In the formula, Represents local structural units With local structural units The degree of structural instability transitions between them Represents local structural units The support release response function output, Represents local structural units The main direction of disturbance propagation, Represents local structural units The main direction of disturbance propagation, Represents local structural units The number of structural recovery steps, Represents local structural units The number of structural recovery steps, This represents the perturbation direction coefficient. This represents the structural recovery step coefficient.

[0030] The objective function of S6 is: ; In the formula, Describes the evaluation function for the boundary candidate set. The function representing the joint degree of discreteness, Represents the set of candidate local structural elements at the boundary. Indicates the index of the boundary connected subset. Represents a set The total number of boundary-connected subsets, Indicates the candidate boundary weights. Indicates local structural units With local structural units The adjacent local structural unit pairs formed Represents a set The set of connection relationships between adjacent local structural unit pairs.

[0031] In S2, a local structural unit is a local geometric structure formed by its neighboring points, centered on a specific point in the original point cloud data. It is equivalent to a point cloud neighborhood with a support relationship. Existing neighborhood search algorithms are used to obtain neighboring points from the original point cloud data and establish connections and support relationships between them, thus forming local structural units. Neighborhood search algorithms include K-nearest neighbor search and radius-based neighborhood search. The neighborhood distribution of data points located within the original point cloud data is usually relatively uniform. However, data points located at the boundaries of the original point cloud data exhibit a significant asymmetric structure due to the lack of neighbors on one side. The perturbation response of the corresponding local structural units reflects this asymmetric characteristic, thus allowing the identification of boundary regions through the perturbation response of the local structural units.

[0032] The structural support relationships between local structural units constructed in S2 differ from static proximity relationships established solely based on nearest neighbors, radius neighborhoods, or spatial distance. These structural support relationships participate in path selection, propagation intensity allocation, and response attenuation calculation during disturbance propagation. The structural support relationships between local structural units include determining the support response function from the source local structural unit to the target local structural unit based on at least one of the following: spatial distance between local structural units, normal difference, curvature transition degree, and consistency of surface extension direction. The coefficients of each term in the support response function are non-negative. , , , The value of is not less than zero. The coefficients of each term in the support response function are used to characterize the contribution of the distance correlation term, normal uniformity term, curvature transition term, and surface extension uniformity term to the support response function, satisfying: Based on the characteristics of point cloud data, the coefficients of each term in the support response function are determined using either an adaptive or structure-aware approach based on feature distribution, and weights are assigned to each feature term using these coefficients. It is inversely proportional to the point cloud density of local structural units, which can make To obtain the normalized point cloud density value, subtract the normalized point cloud density value from 1. First, calculate the point cloud density of all local structural units, then normalize it so that its value is between 0 and 1, thus obtaining the normalized point cloud density value. Then calculate... ; It is positively correlated with the local curvature change rate and is obtained by calculating the normalized curvature change rate using existing adaptive algorithms; it is also determined based on the curvature difference between local structural elements and neighboring elements using existing adaptive algorithms. It is obtained from the mean curvature; The average angle between the principal directions of adjacent local structural units can be calculated using an adaptive algorithm. If the adaptive algorithm is not used, a preset angle can be used. .

[0033] The structural support relationship is a directed structural support relationship. Starting with a source element (any local structural element selected as the source element), multiple directed support vectors are established to surrounding neighboring points to describe the local geometric distribution and structural support state. Arrows represent the directed structural support relationships, with different directions corresponding to different local support strengths and geometric changes. The support vectors not only reflect the positional relationships between neighboring points but also participate in disturbance propagation and dynamic response analysis.

[0034] The support response function is used to calculate the response strength of a local structure under disturbance. The function comprehensively considers curvature changes, area changes, neighborhood geometric offset, and dynamic propagation terms to obtain the support response value of the local structure. In the figure, thick solid lines represent strong support; thin lines represent weak support. The internal region usually has a uniform and stable strong support structure, while the boundary region, due to missing neighborhoods, exhibits weakened support, directional imbalance, and enhanced response. Therefore, the point cloud boundary can be identified by changes in the support response. A complete support network is formed between local structural units, and stable support relationships are established between neighboring points through directed connections, allowing disturbances to propagate and spread within the network. At this point, the local structure exhibits strong stability, and its support response function value is denoted as... , representing local structural units With local structural units The original supporting contributions between them.

[0035] In S3, virtual support release perturbation adjusts the support response function of local structural units through a perturbation operator, rather than physically deleting sampled points in the original point cloud or breaking static adjacency relationships. Virtual support release perturbation alters the way support contributions participate in the structural propagation process, allowing the perturbation effect to spread throughout the local structure in a propagating manner. Applying "virtual non-physical deletion" to some connections in the local structure simulates structural instability and boundary response processes by changing local support weights; that is, it doesn't actually delete points, but weakens the support contribution of certain connections during propagation. After perturbation, a support release coefficient is introduced to adjust the degree of local support attenuation, yielding a new perturbation response value, indicating that support strength decreases after perturbation. In the internal region, due to the relatively intact support relationships, the overall structure remains relatively stable even with some weakened connections; however, the boundary region itself has insufficient support and is more prone to significant response changes after perturbation. Therefore, boundary points can be identified through the difference in response before and after perturbation.

[0036] In S4, the output of the support release response function can also be obtained through nonlinear combination, adaptive function, or piecewise response function. The structural recovery step count characterizes the number of steps required for a local structural unit to recover from its state after the virtual support release perturbation is applied to a stable state. It is determined by the number of propagation rounds, iteration rounds, or levels. If determined by the number of propagation rounds, each round counts as one step; for example, if there are 3 rounds of propagation, the structural recovery step count is 3, indicating that the information expands outward by a maximum of 3 hops / 3 layers of neighborhood. If determined by the number of iteration rounds, each iteration counts as one step; for example, if the algorithm is set to recover 10 iterations, the structural recovery step count is 10. If there is a convergence condition, the structural recovery step count is "the number of iterations at convergence," and the structural recovery step count cannot exceed the maximum number of iteration rounds. If determined by the number of levels, each level recovered counts as one step. It is necessary to distinguish whether the initial layer is given. If starting from an empty structure, all layers must be recovered. If the root / initial layer is known, only the layers after the root / initial layer are recovered. If all three parameters appear simultaneously, the following rules apply: ; In the formula, For the number of structural recovery steps, For the number of rounds of propagation, For the number of iteration rounds, This represents the number of levels. If the model is subject to all three constraints simultaneously, to ensure sufficient coverage during the recovery process, we can take: , It is the number of effective hierarchical recovery steps obtained according to the rules above. The function is used to find the maximum value. The number of structural recovery steps is not set independently, but is equal to the "actual number of rounds executed" in the selected recovery mechanism. For propagation recovery, the number of rounds propagation is the number of steps; for iterative recovery, the number of iterations is the number of steps; and for hierarchical recovery, the number of layers recovered is the number of steps.

[0037] In the iterative stability determination condition, the vector norm is used to calculate the difference vector of the response states between two adjacent iterations. The stability determination threshold is a preset value, which is usually greater than zero. m is the preset number of consecutive stability determinations, and m is a positive integer. Its value is adjusted according to actual factors such as point cloud density, noise level, and structural complexity. Local structural unit In the The response state vector during round propagation is: ; In the formula, Represents the transpose of a matrix or vector. Represents local structural units In the The output of the support-release response function during wheel propagation is equivalent to the local structural unit. In the The support release response function value during wheel propagation. Representing local structural units respectively In the During the propagation of the wheel , , .

[0038] The structure recovery step count is the value of the iteration cycle corresponding to the first time that the above-mentioned iterative stability judgment condition is met for m consecutive iterations after the perturbation is applied. If the iteration terminates after the local structural unit is determined to have reached a stable state, the structure recovery step count is the value of the ending iteration cycle. If the condition is not met within the preset maximum iteration cycle, the maximum iteration cycle is taken as the structure recovery step count. The maximum iteration cycle is set according to the number of neighborhoods of the original point cloud data. For sparse point clouds or small-scale structures, the maximum iteration cycle is usually 50 to 100; for dense point clouds or large-scale structures, it is usually 200 to 500.

[0039] The main direction of disturbance propagation is determined based on the timing of response state changes in each local structural unit, the gradient of structural recovery steps, or the propagation consistency of the response state vector.

[0040] ; In the formula, Local structural unit The main direction of disturbance propagation; To be related to local structural units A set of adjacent local structural units that have a structural support relationship; Local structural unit For local structural units The contribution weight of the direction of perturbation propagation; For local structural units Pointing to local structural units The unit direction vector; This is a regularization term used to avoid a denominator of zero. Specifically, the denominator contains... The norm of a vector.

[0041] The disturbance propagates along the local structural network in a directional manner and generates local structural units. response function output With structural recovery steps .in, Indicates the intensity of local structural response; This represents the number of steps required to restore stability. In the internal region, due to its intact structure and uniform support, disturbances decay rapidly, resulting in a weaker response and faster recovery. In the boundary region, due to missing neighborhoods and unbalanced support, disturbances propagate for a longer time, exhibiting a stronger response and a larger number of recovery steps. When the change in adjacent iterations is less than a threshold, the structure stabilizes. The difference in recovery characteristics between the internal and boundary regions is utilized to achieve boundary localization and extraction.

[0042] In S5, within the local structural interior region, since the local structural unit has many redundant support paths, the disturbance effect can be dispersed and gradually dissipated along multiple support paths, thus the number of structural recovery steps is small; in the structural boundary region, slender connection region, or topological edge region, since the local structural unit has fewer redundant support paths, the disturbance effect is difficult to distribute through multiple paths, thus the number of structural recovery steps is large; the number of structural recovery steps is one of the structural discrimination criteria for distinguishing between the local structural interior region and the structural boundary region.

[0043] The structural instability transition degree between adjacent local structural units is used to characterize the discontinuity of the response behavior of adjacent local structural units under support release perturbation. Based on the boundary connectivity and final extraction process of the structural response field, it is divided into three stages: "response field and candidate boundary pairs", "connectivity expansion process", and "extraction results and final output". First, candidate boundary regions are extracted based on the structural instability response field. Positions with response transition values ​​exceeding a threshold are considered candidate boundary points. The boundary continuity and structural constraints are optimized through an objective function. By continuously expanding along the response enhancement region, the discrete candidate boundaries are gradually connected to form a continuous boundary front zone.

[0044] In S6, the structural boundary is located based on the structural instability transition degree, including: S6.1, determine adjacent local structural unit pairs whose structural instability transition degree exceeds the neighborhood consistency judgment value as candidate boundary pairs, and construct a set of candidate boundary pairs; S6.2 extends the candidate boundary pair set from S6.1 to form a structural boundary, including the structural boundary leading edge zone, boundary curve, and boundary surface. S6.3 By solving the objective function, the local structural elements located in the leading edge zone of the structural boundary are identified as boundary candidate local structural elements. A set of boundary candidate local structural elements is constructed based on the boundary candidate local structural elements, and the boundary candidate local structural elements are the elements of the set.

[0045] In S6.1, the neighborhood consistency judgment value can be set as the mean of the transition degrees of all adjacent units plus one standard deviation, or it can be set directly as a fixed empirical value, such as 0.6.

[0046] In S6.2, a connectivity expansion algorithm (such as region growing and graph connected component extraction) is used to expand the candidate boundary pair set to construct a structural boundary front band (or curve / surface). The candidate boundary pair set from S6.1 is expanded to form the structural boundary front band. This involves treating the candidate boundary pairs as edges of a feature graph composed of the candidate boundary pair set and its subgraph edge set. All connected components of the feature graph are extracted, and the boundary front band is formed by connecting each connected component to a one-hop unit. The boundary candidate local structural unit set consists of local structural units located within or associated with the structural boundary front band. The candidate boundary weights are non-negative weight coefficients. The boundary candidate local structural unit set is determined based on the structural boundary front band, and boundary points corresponding to the boundary candidate local structural unit set are extracted from the original point cloud. At least one of boundary curves and boundary surfaces is output to obtain the point cloud boundary extraction result. Let the candidate boundary pair set constructed in S6.1 be... The induced subgraph edge set is , characterization The set of adjacency relations of all local structural units, and the feature graph formed by the set of candidate boundary pairs and their subgraph edge sets are: The diagram It is decomposed into several connected components, and the vertex set of each connected component is a connected subset. Different connected subsets are connected by... There are no connected edges in the middle. Objective function middle The term pervades all connected subsets, used to calculate the joint discreteness of each connected subset, and then sums the discretenesses of each connected subset, with different subsets independent of each other. The term traverses all adjacent pairs of elements within it. All edges (i.e. The sum of the structural instability transitions of all adjacent elements within the structure is calculated, and the candidate boundary weights represent the encouragement... It contains more edges with higher transition degrees. Each connected subset is A non-empty subset is composed of and its internal adjacency relationship The only decision.

[0047] In the support release perturbation operator, local structural elements support release coefficient It is an integrated perturbation factor, essentially a single-dimensional perturbation intensity input to the support response function. Its function is to quantify the applied perturbation effect on the local structural units. All support response function values ​​of the departure Perform proportional attenuation. The larger the value, the stronger the attenuation, indicating a local structural unit. The more the support provided to the outside world is "released" (weakened), the more it is affected by the local structural units. The local geometric and topological features are jointly determined, including local normal variance, curvature variation, and neighborhood overlap defect degree. Propagation state change, connectivity state change, and surface continuity change are three observable responses reflecting changes in support contribution during structural propagation after a perturbation is applied to the support response function. A virtual support release perturbation is applied to the local structural unit using a support release perturbation operator. The propagation state change, connectivity state change, and surface continuity change caused by the propagation and evolution of this perturbation in the structural support composed of the support response function are observed. The boundary is then identified based on the comprehensive performance of these perturbation process changes.

[0048] The propagation state change refers to the change in the ability of perturbation energy originating from the corresponding local structural unit to propagate to a distant location after a perturbation is applied through a perturbation operator. The larger the size, the stronger the attenuation, and the weaker the ability of the disturbance energy to propagate to distant places. The change in propagation state is determined by comparing the local structural units before and after the disturbance. Quantified by the number or depth of the propagation paths originating from the source, it is equivalent to the change in the propagation paths that support the output of the release response function. .

[0049] Changes in connectivity reflect changes in the magnitude of connectivity components of adjacent structural units before and after a disturbance. The connectivity stability of adjacent structural units before and after a disturbance is quantified and is equivalent to the change in connectivity stability output by the support release response function. .when This makes certain local structural units When the support edge of a connected structural support relationship breaks, it relates to the local structural unit. The connected components may also break.

[0050] The change in surface continuity reflects the change in the support response value along the surface direction before and after the disturbance. It is quantified by the first-order difference change in the support response value along the surface direction before and after the disturbance, and is equivalent to a local structural unit. Changes in surface continuity before and after applying virtual support to release disturbance .

[0051] Before and after applying virtual supports to release the disturbance, the propagation state, connectivity state, and surface continuity state of the local structural unit are recorded respectively. The change in its disturbance process is expressed as: ; ; ; In the formula, , These represent the propagation states before and after the disturbance, respectively, with the propagation state being equivalent to the propagation path length; , These represent the connectivity states before and after the disturbance, respectively. The connectivity state is equivalent to the connectivity stability. Indicates the change in connectivity; , These represent the surface continuity state before and after the disturbance, respectively.

[0052] Support the coefficients in the output of the release response function. , , All are non-negative weighting coefficients, used to characterize the contribution of changes in propagation path, connectivity stability, and surface continuity to the output of the support-release response function, and satisfying the following: The result is obtained by combining statistical characteristics before and after the perturbation using existing adaptive algorithms. If an adaptive algorithm is not used, a preset value can be used for sharp boundaries. , , For smooth boundaries, preset values ​​can be used. , , For the boundaries of the holes, a preset value can be used. , , .

[0053] The final boundary extraction results include boundary points, boundary curves, and boundary surfaces. During the boundary extraction process, internal regions are automatically faded due to their weaker response, while boundary regions are expanded through connectivity to form a complete and continuous boundary structure, thus achieving the final localization and extraction of the point cloud boundaries.

[0054] Example 2 A point cloud boundary extraction device based on virtual support release disturbance response, comprising: The acquisition and support relationship construction module is used to acquire the original point cloud data of the target object or target scene, and construct the structural support relationship between local structural units based on the original point cloud data. The structural support relationship is used to characterize the response transmission capability of the source local structural unit to the target local structural unit in the disturbance propagation process. The virtual support release module is used to apply virtual support release perturbation to each local structural unit in the original point cloud, so as to change the support contribution participation mode of the corresponding local structural unit in the subsequent perturbation propagation process. The response calculation module is used to calculate the support release response information corresponding to each local structural unit based on the changes in the propagation state, connectivity state, and surface continuity of the local structure before and after the virtual support release disturbance. The support release response information includes at least the output of the support release response function and the number of structural recovery steps. The structural instability response field construction module is used to construct the structural instability response field based on the support release response information of each local structural unit. The leading edge positioning module is used to determine the structural instability transition degree between adjacent local structural units based on the structural instability response field, and to locate the leading edge of the structural boundary according to the structural instability transition degree; The boundary extraction output module is used to determine the set of candidate local structural units based on the leading edge of the structural boundary, and to extract boundary points from the original point cloud to obtain the point cloud boundary extraction result.

[0055] To verify the effectiveness and applicability of the point cloud boundary extraction method based on virtual support release disturbance response, an experimental evaluation was conducted on Example 1. Real-world scene point cloud data and industrial scanned point cloud data were selected as test objects, and compared with existing boundary extraction methods. The test data selected for the experiment included several real-world scene point cloud samples and several industrial scanned point cloud samples.

[0056] Real-world point cloud data includes typical complex structural scenes such as vehicle-background boundary areas, roadside edge areas, rod-shaped structure areas, and partially occluded areas; industrial scan point cloud data includes thin-sheet edge areas, acute-angle connectors, step edges, and areas with uneven sampling density. Boundary reference ground truth is constructed through manual annotation combined with consistency rule verification for subsequent quantitative evaluation.

[0057] The point cloud boundary extraction method based on virtual support release perturbation response in Example 1 is designated as Method D. Three existing point cloud boundary extraction methods, Method A, Method B, and Method C, are selected for performance comparison with Method D. Method A is a static geometric boundary extraction method based on curvature and normal features; Method B is a boundary localization method based on local adjacency graphs and static topological relationships; and Method C is a boundary classification method based on point cloud deep feature learning. Method C is trained on an independently labeled training dataset and then evaluated on test data. Evaluation metrics include precision, recall, F1 score, boundary integrity, and processing time. The average number of structural recovery steps is used as an auxiliary analysis metric for the point cloud boundary extraction method based on virtual support release perturbation response, characterizing the average number of propagation rounds required for a local structural unit to recover to a stable state after applying a virtual support release perturbation.

[0058] The experimental results of three existing point cloud boundary extraction methods and a point cloud boundary extraction method based on virtual support release disturbance response on real scene point clouds are shown in Table 1. Table 1. Performance comparison of point cloud boundary extraction in real-world scenarios; .

[0059] As shown in Table 1, the point cloud boundary extraction method based on virtual support release disturbance response has achieved stable results in terms of precision, recall, F1 score and boundary integrity compared to methods A, B and C.

[0060] Compared with method A, method D reduces false detections in complex connected regions and occluded boundary regions; compared with method B, method D shows better control over cross-boundary false connectivity; compared with method C, method D maintains similar boundary extraction accuracy and has structural interpretability without relying on training samples.

[0061] Furthermore, the average number of structural recovery steps for method D is 4.3, indicating that the local structure can recover to a stable state within a limited number of propagation rounds after the disturbance is released by applying virtual supports. At the same time, the boundary region and the internal region show differences in recovery behavior, which can be used as one of the criteria for structural discrimination.

[0062] The experimental results of three existing point cloud boundary extraction methods and a point cloud boundary extraction method based on virtual support release disturbance response in industrial scanning and non-uniform sampling scenarios are shown in Table 2. Table 2 Performance comparison in industrial scanning and non-uniform sampling scenarios; .

[0063] As shown in Table 2, under industrial scanning conditions, Method D demonstrates stable performance in terms of F1 score and boundary integrity. In areas with thin sheet edges, acute-angle connections, and fewer local support paths, it is evident that modeling the disturbance propagation and recovery process can identify locations of structural changes.

[0064] An ablation experiment was conducted on the point cloud boundary extraction method based on virtual support release disturbance response, and the results are shown in Table 3. Table 3. Ablation experiment results; .

[0065] As shown in Table 3, model performance degrades when structural recovery steps or the main direction of perturbation propagation are removed; the performance degradation is even greater when static proximity relationships are used instead of structural support relationships, or when virtual support release perturbations are not applied. These results indicate that structural support relationships and virtual support release perturbations play a role in boundary identification.

[0066] As shown in Tables 1 to 3, the point cloud boundary extraction method based on virtual support release perturbation response achieves stable boundary extraction results under different scenario conditions by constructing structural support relationships, applying virtual support release perturbations, and combining the number of structural recovery steps for boundary discrimination. This method demonstrates good adaptability in complex connected regions, slender structural regions, and regions with varying sampling density.

[0067] Figure 1 This is a schematic diagram illustrating the overall process of a point cloud boundary extraction method according to one embodiment of the present invention, demonstrating a point cloud boundary extraction process based on local structural response. First, local structural units are constructed from the original point cloud data. Directed support relationships are established between neighboring points to form a local geometric structure. Then, virtual supports are applied to release perturbations, the response information of the local structure under perturbation is calculated, and a structural instability response field is constructed. Because the boundary region has missing neighbors and unbalanced supports, its response changes are usually more pronounced. Therefore, the boundary front can be located through structural instability transitions, ultimately achieving the extraction of point cloud boundaries, boundary curves, or boundary surfaces.

[0068] Figure 2 It displays the original point cloud benchmark containing approximately 70,000 points. Figure 3 , Figure 4 , Figure 5 The extraction results of method A (geometry), method B (graphical model), and method C (deep learning) are shown respectively. Figures 3 to 5 The blue dashed line represents the physical boundaries predicted by the corresponding method, including the physical contours of the vehicle and the physical boundaries of the walls. Figure 6The results of the point cloud boundary extraction method based on virtual support release disturbance response are shown. In the figure, the red box represents the physical outline of the car, and the green dashed line represents the physical boundary of the wall. Among the three existing point cloud boundary extraction methods, the highest F1 score is only 0.859. In contrast, the boundary curve (red solid line) of the point cloud boundary extraction method based on virtual support release disturbance response not only visually matches the real outline better, but also performs best in quantitative indicators, demonstrating the algorithm's high-precision extraction capability in complex scenes.

[0069] like Figure 3 The drawbacks of method A shown are: severe boundary drift, numerous burrs, and confusion between the curb and vehicle boundaries; such as Figure 4 As shown, the drawback of method B is the obvious boundary breakage, especially the appearance of voids at the shading points; for example... Figure 5 As shown, the drawbacks of method C are: it performs well overall but misses slender structures, has discontinuous boundaries, and poor smoothness; for example... Figures 3 to 6 As shown, the advantages of the point cloud boundary extraction method based on virtual support release disturbance response are: the boundary curve is completely continuous, smooth, and closely follows the real structure, without drift or breakage, perfectly capturing the two-sided boundary at the intersection of vehicle and curb.

[0070] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A point cloud boundary extraction method based on virtual support release disturbance response, characterized in that, include: S1. Obtain raw point cloud data, select neighboring points from the raw point cloud data using a neighborhood search algorithm, and construct multiple point cloud neighborhoods; S2. Establish the directed structural support relationship between any two points in the point cloud neighborhood, construct the local structural units corresponding to each point cloud neighborhood, and construct the structural support relationship between the local structural units based on the support response function between the local structural units. S3. Apply virtual support disturbance release to each local structural unit and obtain the disturbance process changes of each local structural unit. The disturbance process changes of the local structural unit include changes in propagation state, changes in connectivity state, and changes in surface continuity. S4. Based on the changes in the disturbance process of each local structural unit, calculate the support release response information of each local structural unit. The support release response information includes the output of the support release response function, the number of structural recovery steps, and the main direction of disturbance propagation. S5. Based on the support release response information of each local structural unit, construct the structural instability response field and calculate the structural instability transition degree between adjacent local structural units; S6. Construct an objective function, locate the structural boundary based on the structural instability transition degree, filter local structural elements located at the structural boundary through the objective function, construct a set of candidate local structural elements for the boundary, and output the point cloud boundary extraction results; In S3, virtual support disturbance release is applied to each local structural unit, including adjusting the support response function between local structural units through the support release disturbance operator, and changing the support contribution participation mode in the structural propagation process so that the disturbance effect spreads in the local structure in the form of propagation. The support release perturbation operator is: ； In the formula, Indicates to The support response function value after applying virtual support to release the disturbance. Represents local structural units The support release coefficient; The output of the support release response function in S4 is: ； In the formula, Represents local structural units The support release response function output, Represents local structural units The change in propagation path before and after applying virtual support to release the disturbance. Represents local structural units The change in connectivity stability before and after applying virtual support to release the disturbance. Represents local structural units The change in surface continuity before and after applying virtual support to release the disturbance. Indicates the difference. , , They represent , , The coefficient.

2. The point cloud boundary extraction method based on virtual support release disturbance response according to claim 1, characterized in that, In S2, the support response function between local structural elements is determined based on the spatial distance, normal difference, curvature transition degree, and surface extension direction consistency between them. ； In the formula, Represents local structural units Pointing to local structural units The support response function value, Represents local structural units With local structural units The distance correlation term is used to characterize the spatial distance between local structural units. Represents local structural units With local structural units The normal consistency term is used to characterize the normal difference between local structural units. Represents local structural units With local structural units The curvature transition term is used to characterize the degree of curvature transition between local structural units. Represents local structural units With local structural units The surface extension consistency term is used to characterize the consistency of the surface extension direction between local structural units. , , , They represent , , , coefficient, and This is an index for local structural units.

3. The point cloud boundary extraction method based on virtual support release perturbation response according to claim 2, characterized in that, The structural recovery step count is used to characterize the total number of iterations required for a local structural unit to recover from the state after the virtual support release disturbance is applied to the stable state. During the iteration process, when the local structural unit The local structural unit is determined when the response state satisfies the iterative stability criterion for m consecutive times. Reaching a stable state; The iterative stability criterion is: ； In the formula, Represents local structural units In the The response state vector during round propagation, Represents local structural units In the The response state vector during round propagation, Represents the norm, This represents the stability threshold. Indicates the iteration round index; For a local structural unit that has reached a stable state, the number of structural recovery steps is the iteration cycle index of the last iteration when the iterative stability criterion is met for m consecutive iterations; for a local structural unit that has not reached a stable state within the preset maximum number of iterations, the number of structural recovery steps is the maximum number of iterations.

4. The point cloud boundary extraction method based on virtual support release disturbance response according to claim 3, characterized in that, The structural instability transition degree is: ； In the formula, Represents local structural units With local structural units The degree of structural instability transitions between them Represents local structural units The support release response function output, Represents local structural units The main direction of disturbance propagation, Represents local structural units The main direction of disturbance propagation, Represents local structural units The number of structural recovery steps, Represents local structural units The number of structural recovery steps, This represents the perturbation direction coefficient. This represents the structural recovery step coefficient.

5. The point cloud boundary extraction method based on virtual support release disturbance response according to claim 4, characterized in that, The objective function of S6 is: ； In the formula, Describes the evaluation function for the boundary candidate set. The function representing the joint degree of discreteness, Represents the set of candidate local structural elements at the boundary. Indicates the index of the boundary connected subset. Represents a set The total number of boundary-connected subsets, Indicates the candidate boundary weights. Indicates local structural units With local structural units The adjacent local structural unit pairs formed Represents a set The set of connection relationships between adjacent local structural unit pairs.

Images