Method and device for multi-scale point cloud feature extraction based on feature point neighborhood division

Abstract

The application is suitable for the technical field of point cloud data processing, and provides a multi-scale point cloud feature extraction method and device based on feature point division neighborhood. In the embodiment, for target point cloud data requiring feature extraction, the density change value and the density change gradient of the point cloud data are determined to determine the target feature points in the point cloud data. Then, the neighborhood feature point set of each target feature point is determined according to different radii, and then the feature points in the neighborhood feature point set are sampled and the feature vectors of each sampling feature point are calculated, and then the target feature vectors are obtained by merging the feature vectors. Finally, the multi-scale features of the target point cloud data are determined by a plurality of target feature vectors determined by different radii. The application solves the problems of poor capture of subtle geometric features, high calculation cost and low calculation efficiency of the existing point cloud feature extraction method.

Description

Technical Field

[0001] This application relates to the field of point cloud data processing technology, and in particular to a multi-scale point cloud feature extraction method and apparatus based on feature point neighborhood division. Background Technology

[0002] Feature extraction from point cloud data is a crucial task in 3D data processing, widely applied in fields such as autonomous driving and robotics. Its primary goal is to extract meaningful feature information from sparse and unordered 3D point cloud data. Methods for directly processing point cloud data have gained significant attention in recent years, with representative techniques including PointNet and its derivatives. PointNet was the first method to directly process unordered point cloud data, extracting features from each point individually and then integrating global features using symmetry functions (such as max-pooling). PointNet++ further improved this architecture by introducing a hierarchical sampling mechanism to capture local geometric features, thereby enhancing the model's ability to describe details. These methods do not require converting point cloud data into meshes or voxels, avoiding information loss and computational overhead associated with data transformation. Furthermore, other techniques, such as DGCNN, mine relationships between local points in point cloud data by constructing dynamic neighborhood graphs. These methods for directly processing point clouds offer good flexibility and scalability, making them suitable for handling complex 3D data.

[0003] While methods that directly process point cloud data (such as PointNet, PointNet++, and DGCNN) have made significant progress in 3D point cloud processing, some objective drawbacks remain. First, PointNet processes point clouds point by point, lacking characterization of local structures and failing to fully utilize neighborhood information, thus performing poorly in capturing subtle geometric features. Although PointNet++ improves this through hierarchical sampling and local feature extraction, its local region partitioning still relies on Euclidean distance, making it difficult to handle scenes with complex topological structures. Second, while DGCNN introduces dynamic neighborhood mapping, the dynamic changes in the neighborhood require high computational costs, potentially leading to low computational efficiency, especially when processing large-scale point cloud data. Although these techniques avoid information loss due to data transformation, there is still room for improvement in handling high-dimensional data and complex geometric structures. Summary of the Invention

[0004] In view of this, this application provides a multi-scale point cloud feature extraction method and apparatus based on feature point neighborhood division, in order to solve the problems of poor performance in capturing subtle geometric features, high computational cost and low computational efficiency of existing point cloud feature extraction methods.

[0005] The first aspect of this application provides a multi-scale point cloud feature extraction method based on feature point neighborhood division, the method comprising:

[0006] Acquire target point cloud data, determine the neighborhood feature point set of any feature point in the target point cloud data under different radii according to the preset radius set r, and determine the feature point density of the neighborhood feature point set;

[0007] The density change value of the feature points is determined based on the neighborhood feature point set, the radius set R, and the corresponding feature point density. and density change gradient , wherein the density change value The density variation of the neighborhood feature point set generated by the feature point at different radii, wherein the density variation gradient Characterizes the density difference between the feature point and other feature points in the neighborhood feature point set corresponding to the feature point;

[0008] According to the density change value and the density change gradient Determine the target feature point set For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points.

[0009] Several sampling feature points are determined from each target neighborhood feature point set. The feature vector of each sampling feature point is calculated, and these feature vectors are merged into a target feature vector. The number of feature points is adjusted several times according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

[0010] Optionally, the step of determining the density change value of the feature points based on the neighborhood feature point set, the radius set r, and the corresponding feature point density is further elaborated. include:

[0011] Through formula Determine the feature points density change value , wherein radius Lower feature point The feature point density of the corresponding neighborhood feature point set.

[0012] Optionally, the step of determining the density change gradient based on the neighborhood feature point set, the radius set r, and the corresponding feature point density is described. include:

[0013] Through formula Determine feature points With the feature points The density difference of other feature points within the neighborhood feature point set, wherein the For the neighborhood feature point set, the The number of feature points within the neighborhood feature point set. The feature point density of the neighborhood feature point set corresponding to the feature point is... This represents the feature point density of the neighborhood feature point set corresponding to other feature points.

[0014] Optionally, determining the target neighborhood feature point set of the target feature point based on a preset initial radius and a preset range of feature point numbers includes:

[0015] For any target feature point, an initial neighborhood feature point set is determined based on a preset initial radius, and the number of feature points in the initial neighborhood feature point set is compared with a preset range of feature point numbers.

[0016] When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the increased preset initial radius, until the number of feature points is within the feature point number range.

[0017] When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the reduced preset initial radius, until the number of feature points is within the feature point number range.

[0018] Optionally, determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector includes:

[0019] Through formula The target feature vectors, determined by the range of the number of feature points, are concatenated to obtain the multi-scale features, wherein... The initial radius is The target feature vector is determined at that time.

[0020] A second aspect of this application provides a multi-scale point cloud feature extraction device based on feature point neighborhood division, the device comprising:

[0021] The density determination unit is used to acquire target point cloud data, determine the neighborhood feature point set of any feature point in the target point cloud data under different radii according to a preset radius set r, and determine the feature point density of the neighborhood feature point set.

[0022] The density change parameter determination unit is used to determine the density change value of the feature points based on the neighborhood feature point set, the radius set R, and the corresponding feature point density. and density change gradient , wherein the density change value The density variation of the neighborhood feature point set generated by the feature point at different radii, wherein the density variation gradient Characterizes the density difference between the feature point and other feature points in the neighborhood feature point set corresponding to the feature point;

[0023] The target neighborhood feature point set determination unit is used to determine the density change value. and the density change gradient Determine the target feature point set For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points.

[0024] A multi-scale feature determination unit is used to determine several sampled feature points from each target neighborhood feature point set, calculate the feature vector of each sampled feature point, merge the feature vectors into a target feature vector, and perform several adjustments to the number range of the feature points according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

[0025] Optionally, the density change parameter determination unit determines the density change value of the feature point based on the neighborhood feature point set, the radius set r, and the corresponding feature point density. include:

[0026] Through formula Determine the feature points density change value , wherein radius Lower feature point The feature point density of the corresponding neighborhood feature point set.

[0027] Optionally, the density change parameter determination unit determines the density change gradient based on the neighborhood feature point set, the radius set r, and the corresponding feature point density. include:

[0028] Through formula Determine feature points With the feature points The density difference of other feature points within the neighborhood feature point set, wherein the For the neighborhood feature point set, the The number of feature points within the neighborhood feature point set. The feature point density of the neighborhood feature point set corresponding to the feature point is... This represents the feature point density of the neighborhood feature point set corresponding to other feature points.

[0029] Optionally, the target neighborhood feature point set determination unit, which determines the target feature point's target neighborhood feature point set based on a preset initial radius and a preset feature point quantity range, includes:

[0030] For any target feature point, an initial neighborhood feature point set is determined based on a preset initial radius, and the number of feature points in the initial neighborhood feature point set is compared with a preset range of feature point numbers.

[0031] When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the increased preset initial radius, until the number of feature points is within the feature point number range.

[0032] When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the reduced preset initial radius, until the number of feature points is within the feature point number range.

[0033] Optionally, the multi-scale feature determination unit, which determines the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector, includes:

[0034] Through formula The target feature vectors, determined by the range of the number of feature points, are concatenated to obtain the multi-scale features, wherein... The initial radius is The target feature vector is determined at that time.

[0035] In the embodiments provided in this application, for target point cloud data that requires feature extraction, the density change value and density change gradient of the point cloud data are first determined to identify the target feature points in the point cloud data. Then, a neighborhood feature point set for each target feature point is determined based on different radii. After sampling the feature points in the neighborhood feature point set and calculating the feature vector of each sampled feature point, they are merged into a target feature vector. Finally, the multi-scale features of the target point cloud data are determined through several target feature vectors determined by different radii.

[0036] The advantages of this application are:

[0037] 1) Adapting to the local topology of point clouds: This application selects feature points first and then divides the neighborhood, which can better adapt to the complex topology of point cloud data and overcome the shortcomings of the traditional average neighborhood division method that ignores the geometric characteristics of point clouds.

[0038] 2) Dynamically Adjust Neighborhood Radius: The neighborhood radius is dynamically adjusted based on the local density of feature points to ensure that each feature point has a reasonable number of feature points in its neighborhood, thereby improving the representation capability of neighborhood information. This helps avoid errors in sparse or overly dense regions and enhances the extraction effect of geometric structure information.

[0039] 3) Multi-scale feature extraction: Multi-scale density calculation and feature extraction are adopted to capture multi-level geometric information in point cloud data through neighborhoods of different radii, making the extracted features more complete and enhancing the expressive power of point cloud features. Attached Figure Description

[0040] Figure 1 A flowchart illustrating the method provided in this application embodiment;

[0041] Figure 2 This is a schematic diagram of neighborhood division provided for an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of feature extraction provided for an embodiment of this application;

[0043] Figure 4 This is a structural diagram of the device provided in the embodiments of this application;

[0044] Figure 5 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0045] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0046] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0047] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0048] This application provides a multi-scale point cloud feature extraction method and apparatus based on feature point neighborhood division, in order to solve the problems of poor performance in capturing subtle geometric features, high computational cost, and low computational efficiency of existing point cloud feature extraction methods.

[0049] The technical solutions of this application will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0050] like Figure 1 The diagram shows a flowchart of a multi-scale point cloud feature extraction method based on feature point neighborhood division provided in this application. The steps of this process include:

[0051] S101, acquire target point cloud data, determine the neighborhood feature point set of any feature point in the target point cloud data under different radii according to the preset radius set r, and determine the feature point density of the neighborhood feature point set.

[0052] In this embodiment, for each feature point in the target point cloud data From the set of radii r={ , ,…} Select a value as the radius of the neighborhood, find all feature points within this neighborhood, thus determining the neighborhood feature point set, and then... The density of the neighborhood feature point set is expressed by the formula. It is confirmed that, among them, This represents the number of feature points in the neighborhood feature point set. It is the volume of the neighborhood sphere, and the volume can be obtained through... To calculate the feature point density. Point The density of feature points distributed in its neighborhood.

[0053] Iterate through each radius in the radius set r and each feature point in the target point cloud data, and for each point... It can be in multiple different radii The feature point density of the neighborhood feature point set is calculated below. By observing density variations at different scales, changes in the local geometry of feature points can be captured.

[0054] S102, determine the density change value of the feature points based on the neighborhood feature point set, the radius set r, and the corresponding feature point density. and density change gradient , wherein the density change value The density variation of the neighborhood feature point set generated by the feature point at different radii, wherein the density variation gradient It represents the density difference between the feature point and other feature points in the neighborhood feature point set corresponding to the feature point.

[0055] In this embodiment, the density change value It can be done through formula Confirmed, in this formula Is it at different radii Lower point feature point The feature point density in the neighborhood feature point set. Density variation value. It can reflect the density differences of point clouds at different scales; a larger density change value means that the point cloud is denser than the point cloud. It is located in a prominent position in the local geometry, such as an edge or inflection point.

[0056] For density change gradient The density gradient of a feature point is obtained by comparing its density differences with those of other feature points in its neighborhood. The specific calculation formula is as follows: In the formula For the neighborhood feature point set, This represents the number of feature points within the neighborhood feature point set. The feature point density of this neighborhood feature point set, the This is the feature point density of the neighborhood feature point set corresponding to other feature points in this neighborhood feature point set. Determine the feature point density. and The method is the same as the one described above, and will not be repeated here. The density gradient can be used as a feature point. The direction of density variation in its neighborhood helps to determine the location of significant feature points.

[0057] S103, based on the density change value and the density change gradient Determine the target feature point set For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points.

[0058] In this embodiment, the density change value can be analyzed separately. and density change gradient Set a threshold for each, such as Corresponding density change value , Corresponding density change gradient When feature points This means that the region where the feature point is located has a complex geometric structure, such as a large change in curvature. Therefore, this feature point is defined as the target feature point. Add to the target feature point set Or when feature points When the density change direction of the feature point is relatively obvious, it means that the feature point may be located at the edge or inflection point of a local geometric structure. Therefore, this feature point is defined as the target feature point. Add to the target feature point set .

[0059] Determine the target feature point set Then, for each target feature point in the set By using the method in step S101, the radius of each target feature point is determined to be within a preset initial radius. The initial target neighborhood feature point set is then used, and the number of feature points in this neighborhood feature point set is compared with the preset feature point number range.

[0060] When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased. And based on the increased preset initial radius The number of feature points in the initial neighborhood feature point set of the target feature point is redetermined until the number of feature points is within the range of the number of feature points.

[0061] When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced. And based on the reduced preset initial radius The number of feature points in the initial neighborhood feature point set of the target feature point is redefined until the number of feature points falls within the specified range. The final neighborhood partitioning diagram is shown below. Figure 2 As shown.

[0062] Finally, the initial neighborhood feature point set whose number of feature points falls within the specified range is determined as the target neighborhood feature point set, denoted as . .

[0063] S104: Determine several sampling feature points from each target neighborhood feature point set, calculate the feature vector of each sampling feature point, merge the feature vectors into a target feature vector, and perform several adjustments to the feature point quantity range according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

[0064] In this embodiment, the target neighborhood feature point set generated for each target feature point can be denoted as: The PointNet method can be used to extract features from the feature point set of each target's neighborhood. The specific method is as follows:

[0065] 1. The FPS (Farthest Point Sampling) algorithm is used to determine sampling feature points from a set of feature points in the target's neighborhood. The sampling feature points selected by this method can ensure that the sampling points in each region are evenly distributed, thereby reducing computational complexity.

[0066] 2. Input each sampled feature point into the PointNet network, and apply a multilayer perceptron (MLP) to calculate the feature vector for each point: .in It is the coordinate information of the i-th sampling point. This is the local feature vector of that point. Then, the features of points in the neighborhood of each feature point are merged into a fixed-length target feature vector using global max pooling. .

[0067] After determining the target feature vector, in order to fully capture the multi-scale information of the point cloud, the number of feature points is first varied across different ranges. Then, the steps of grouping, sampling, and PointNet feature extraction are repeated to extract features at multiple scales. This generates target feature vectors with different initial radii. Then, the feature vectors of each target are concatenated to obtain the multi-scale features of the target point cloud data, which are represented as follows: A schematic diagram of specific feature extraction is shown below. Figure 3 As shown.

[0068] This embodiment dynamically adjusts the neighborhood radius based on the local density of feature points, ensuring that each feature point has a reasonable number of feature points in its neighborhood, thereby improving the representation capability of neighborhood information. This helps avoid errors in sparse or overly dense regions and enhances the extraction effect of geometric structure information.

[0069] Finally, a multi-scale feature fusion method was used to effectively capture the geometric structure of point clouds at different scales, thereby enhancing the feature representation capability.

[0070] This concludes the process. Figure 1 The process is shown below.

[0071] In this embodiment, for target point cloud data requiring feature extraction, the density variation value and density variation gradient of the point cloud data are first determined to identify the target feature points. Then, a neighborhood feature point set is determined for each target feature point based on different radii. The feature points in this neighborhood feature point set are then sampled, and the feature vector of each sampled feature point is calculated. These samples are then merged into a target feature vector. Finally, the multi-scale features of the target point cloud data are determined using several target feature vectors determined by different radii. This solves the problems of existing point cloud feature extraction methods, such as poor performance in capturing subtle geometric features, high computational cost, and low computational efficiency.

[0072] This application also provides a multi-scale point cloud feature extraction device based on feature point neighborhood division, such as... Figure 4 As shown, the device includes:

[0073] Density determination unit 401 is used to acquire target point cloud data, based on a preset radius set r={ , ,… } Determine the neighborhood feature point set of any feature point in the target point cloud data under different radii, and determine the feature point density of the neighborhood feature point set;

[0074] Density change parameter determination unit 402 is used to determine the density change value of the feature points based on the neighborhood feature point set, the radius set R, and the corresponding feature point density. and density change gradient , wherein the density change value The density variation of the neighborhood feature point set generated by the feature point at different radii, wherein the density variation gradient Characterizes the density difference between the feature point and other feature points in the neighborhood feature point set corresponding to the feature point;

[0075] The target neighborhood feature point set determination unit 403 is used to determine the density change value. and the density change gradient Determine the target feature point set For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points.

[0076] The multi-scale feature determination unit 404 is used to determine several sampled feature points from each target neighborhood feature point set, calculate the feature vector of each sampled feature point, merge the feature vectors into a target feature vector, and perform several adjustments to the number range of the feature points according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

[0077] In another embodiment, the density change parameter determination unit determines the density change value of the feature point based on the neighborhood feature point set, the radius set R, and the corresponding feature point density. include:

[0078] Through formula Determine the feature points density change value , wherein radius Lower feature point The feature point density of the corresponding neighborhood feature point set.

[0079] In another embodiment, the density change parameter determination unit determines the density change gradient based on the neighborhood feature point set, the radius set R, and the corresponding feature point density. include:

[0080] Through formula Determine feature points With the feature points The density difference of other feature points within the neighborhood feature point set, wherein the For the neighborhood feature point set, the The number of feature points within the neighborhood feature point set. The feature point density of the neighborhood feature point set corresponding to the feature point is... This represents the feature point density of the neighborhood feature point set corresponding to other feature points.

[0081] In another embodiment, the target neighborhood feature point set determination unit, which determines the target feature point's target neighborhood feature point set based on a preset initial radius and a preset feature point quantity range, includes:

[0082] For any target feature point, an initial neighborhood feature point set is determined based on a preset initial radius, and the number of feature points in the initial neighborhood feature point set is compared with a preset range of feature point numbers.

[0083] When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the increased preset initial radius, until the number of feature points is within the feature point number range.

[0084] When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the reduced preset initial radius, until the number of feature points is within the feature point number range.

[0085] In another embodiment, the multi-scale feature determination unit determines the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector, including:

[0086] Through formula The target feature vectors, determined by the range of the number of feature points, are concatenated to obtain the multi-scale features, wherein... The initial radius is The target feature vector is determined at that time.

[0087] The above embodiments of the present invention provide a multi-scale point cloud feature extraction method based on feature point neighborhood division, and a multi-scale point cloud feature extraction device based on feature point neighborhood division. The above method and device can solve the problems of poor performance in capturing subtle geometric features, high computational cost and low computational efficiency of existing point cloud feature extraction methods.

[0088] This embodiment also discloses a computer device, such as... Figure 5As shown, the computer device includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement any of the above-described multi-scale point cloud feature extraction methods based on feature point neighborhood division.

[0089] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A multi-scale point cloud feature extraction method based on feature point neighborhood division, characterized in that, The method includes: Acquire target point cloud data, and determine any feature point p in the target point cloud data according to a preset radius set r. i Develop neighborhood feature point sets with different radii and determine the feature point density of the neighborhood feature point sets; The feature point p is determined by the maximum and minimum values ​​of the feature point density. i The density change value Δρi is obtained through the formula Determine the feature point p i With the feature point p i The density difference of other feature points within the neighborhood feature point set, wherein the For the neighborhood feature point set, the ρ is the number of feature points in the neighborhood feature point set. i The feature point density of the neighborhood feature point set corresponding to the feature point, ρ j The feature point density of the neighborhood feature point set corresponding to other feature points; According to the density change value and the density change gradient Determine the target feature point set Specifically, this includes: the density change values ​​respectively. and density change gradient Set a threshold. Corresponding density change value , Corresponding density change gradient ,Will or The feature points are defined as the target feature points. And add to the target feature point set ; For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points. Several sampling feature points are determined from each target neighborhood feature point set. The feature vector of each sampling feature point is calculated, and these feature vectors are merged into a target feature vector. The number of feature points is adjusted several times according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

2. The method according to claim 1, characterized in that, The step of determining the density change value Δρi of feature point pi through the maximum and minimum values ​​of the feature point density includes: Through formula Determine the feature points density change value .

3. The method according to claim 1, characterized in that, The step of determining the target neighborhood feature point set of the target feature point based on a preset initial radius and a preset feature point quantity range includes: For any target feature point, an initial neighborhood feature point set is determined based on a preset initial radius, and the number of feature points in the initial neighborhood feature point set is compared with a preset range of feature point numbers. When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the increased preset initial radius, until the number of feature points is within the feature point number range. When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the reduced preset initial radius, until the number of feature points is within the feature point number range.

4. The method according to claim 1, characterized in that, The step of determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector includes: Through formula The target feature vectors, determined by the range of the number of feature points, are concatenated to obtain the multi-scale features, wherein... The initial radius is The target feature vector is determined at that time.

5. A multi-scale point cloud feature extraction device based on feature point neighborhood division, characterized in that, The device includes: The density determination unit is used to acquire target point cloud data and determine any feature point p in the target point cloud data according to a preset radius set r. i Develop neighborhood feature point sets with different radii and determine the feature point density of the neighborhood feature point sets; The density variation parameter determination unit is used to determine the feature point p by the maximum and minimum values ​​of the feature point density. i The density change value Δρi is obtained through the formula Determine the feature point p i With the feature point p i The density difference of other feature points within the neighborhood feature point set, wherein the For the neighborhood feature point set, the ρ is the number of feature points in the neighborhood feature point set. i The feature point density of the neighborhood feature point set corresponding to the feature point, ρ j The feature point density of the neighborhood feature point set corresponding to other feature points; The target neighborhood feature point set determination unit is used to determine the density change value. and the density change gradient Determine the target feature point set Specifically, this includes: the density change values ​​respectively. and density change gradient Set a threshold. Corresponding density change value , Corresponding density change gradient ,Will or The feature points are defined as the target feature points. And add to the target feature point set ; For the target feature point set Any target feature point within the range, based on a preset initial radius The target feature point set of the target feature point is determined by a preset range of the number of feature points. A multi-scale feature determination unit is used to determine several sampled feature points from each target neighborhood feature point set, calculate the feature vector of each sampled feature point, merge the feature vectors into a target feature vector, and perform several adjustments to the number range of the feature points according to the preset initial radius. The step of determining the target feature vector based on the adjusted range of the number of feature points, and determining the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector.

6. The apparatus according to claim 5, characterized in that, The step of determining the density change value Δρi of feature point pi through the maximum and minimum values ​​of the feature point density includes: Through formula Determine the feature points density change value .

7. The apparatus according to claim 5, characterized in that, The target neighborhood feature point set determination unit, which determines the target feature point based on a preset initial radius and a preset range of feature point numbers, includes: For any target feature point, an initial neighborhood feature point set is determined based on a preset initial radius, and the number of feature points in the initial neighborhood feature point set is compared with a preset range of feature point numbers. When the number of feature points is less than the minimum value of the feature point number range, the preset initial radius is increased, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the increased preset initial radius, until the number of feature points is within the feature point number range. When the number of feature points is greater than the maximum value of the feature point number range, the preset initial radius is reduced, and the number of feature points in the initial neighborhood feature point set of the target feature point is re-determined based on the reduced preset initial radius, until the number of feature points is within the feature point number range.

8. The apparatus according to claim 5, characterized in that, The multi-scale feature determination unit, which determines the multi-scale features of the target point cloud data based on the target feature vector and the adjusted target feature vector, includes: Through formula The target feature vectors, determined by the range of the number of feature points, are concatenated to obtain the multi-scale features, wherein... The initial radius is The target feature vector is determined at that time.

Images