Adaptive point cloud sampling adjustment method and system

Abstract

The invention discloses a self-adaptive point cloud sampling adjustment method. The method comprises the following steps: training a feature analysis model; sampling a workpiece at an initial sampling rate to obtain point cloud data of the workpiece, inputting the point cloud data into the feature analysis model to obtain regional features, and dividing a local region based on the regional features; and adjusting the sampling rate of the local region based on the region features. The local structure of the point cloud can be reflected more finely based on the division of the regional features, so that the subsequent sampling rate adjustment is more accurate. The refined division mode is beneficial for reserving important features such as edges and angular points in the point cloud, removing redundant information at the same time, and providing a more valuable data basis for subsequent point cloud processing tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of data acquisition, and in particular to a method and system for adaptive point cloud sampling adjustment. Background Art

[0002] Traditional point cloud downsampling methods often rely on a fixed sampling rate for data processing, which may face significant limitations when processing point cloud data with complex geometric features and high variability. Specifically, a fixed sampling rate may lead to over-downsampling in some areas, which means that key structures and detail information may be unnecessarily removed in these areas, affecting the accuracy of subsequent analysis and processing. Conversely, in other areas, if the sampling rate is fixed, too many points may be retained, which not only increases the redundancy of the data, but also may lead to unnecessary computational and storage overhead, reducing the overall processing efficiency.

[0003] In addition, traditional downsampling methods usually use fixed region division strategies, which lack sufficient flexibility to adapt to the complex local features in point cloud data. The inflexibility of region division may result in an inability to effectively balance sampling accuracy and computational efficiency when processing point clouds with irregular shapes or highly non-uniform distributions. For example, in areas with dense point clouds, fixed region division may not be able to fully capture detailed information, while in areas with sparse point clouds, it may introduce too much noise or error due to the large grid.

[0004] Therefore, how to better balance sampling accuracy, computational efficiency, and storage overhead is an urgent problem to be solved in the current field of point cloud processing. Summary of the invention

[0005] In view of the above-mentioned defects, the purpose of the present invention is to propose a method and system for adaptive point cloud sampling adjustment to solve the problems of inflexible area division and uneven sampling rate distribution in the prior art.

[0006] To achieve this purpose, the present invention adopts the following technical solution: a method for adaptive point cloud sampling adjustment, comprising the following steps:

[0007] Train feature analysis models;

[0008] Sampling the workpiece at an initial sampling rate to obtain point cloud data of the workpiece, inputting the point cloud data into the feature analysis model to obtain regional features, and dividing the local area based on the regional features;

[0009] The sampling rate of the local area is adjusted based on the regional characteristics.

[0010] Preferably, the training process of the feature analysis model is as follows:

[0011] Step A: input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point;

[0012] Step B: For each point, find the K nearest neighbors in its local neighborhood, and learn a K×K transformation matrix X through a small neural network to rearrange the points in the local neighborhood;

[0013] Applying a standard convolution operation to the rearranged points to extract local features, and after extracting the local features, aggregating the local features to form global features;

[0014] Step C: constructing a plurality of said small neural networks;

[0015] The output global features of the previous small neural network are used as the input of the next small neural network;

[0016] Step D: After multiple small neural networks, the final global features are obtained;

[0017] Passing the final global features to the fully connected layer for classification or regression tasks;

[0018] Step E: Train the model using the standard back-propagation algorithm and optimizer; and optimize the weight parameters of the small neural network by minimizing the loss function.

[0019] Preferably, the regional characteristics include density and / or curvature of the local area.

[0020] Preferably, the initial sampling rate of the local area is adjusted as follows:

[0021] If the density is lower than the density threshold, reducing the sampling rate;

[0022] If the density is higher than the density threshold, increasing the sampling rate;

[0023] If the curvature is higher than the curvature threshold, increasing the sampling rate;

[0024] If the curvature is below a curvature threshold, the sampling rate is reduced.

[0025] Preferably, if the adjustment instruction includes both an increase instruction and a decrease instruction, the sampling rate is increased by linear interpolation or spline interpolation, and then the sampling rate is decreased by filtering.

[0026] A system for adaptive point cloud sampling and adjustment, using the method for adaptive point cloud sampling and adjustment, comprising a training module, a collection module and an adjustment module;

[0027] The training module is used to train the feature analysis model;

[0028] The acquisition module is used to sample the workpiece at an initial sampling rate, obtain point cloud data of the workpiece, input the point cloud data into the feature analysis model to obtain regional features, and divide the local area based on the regional features;

[0029] The adjustment module is used to adjust the sampling rate of the local area based on the regional characteristics.

[0030] Preferably, the training module includes a first submodule, a second submodule, a third submodule, a fourth submodule and a fifth submodule;

[0031] The first submodule is used to input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point;

[0032] The second submodule is used to find the K nearest neighboring points in the local neighborhood of each point, learn a K×K transformation matrix X through a small neural network, and rearrange the points in the local neighborhood; apply standard convolution operations to the rearranged points to extract local features, and after extracting the local features, aggregate the local features to form global features;

[0033] The third submodule is used to construct a plurality of said small neural networks;

[0034] The output global features of the previous small neural network are used as the input of the next small neural network;

[0035] The fourth submodule is used to obtain the final global features after passing through multiple small neural networks;

[0036] Passing the final global features to the fully connected layer for classification or regression tasks;

[0037] The fifth submodule is used to train the model using the standard back-propagation algorithm and optimizer; and optimize the weight parameters of the small neural network by minimizing the loss function.

[0038] Preferably, the regulating module includes a first regulating submodule and a second regulating submodule;

[0039] The first regulating submodule is used to reduce the sampling rate when the density is lower than the density threshold, or reduce the sampling rate when the curvature is lower than the curvature threshold;

[0040] The second regulating submodule is used to increase the sampling rate when the density is higher than the density threshold, or to increase the sampling rate when the curvature is higher than the curvature threshold.

[0041] Preferably, the regulating module further includes a third regulating submodule;

[0042] The third regulating submodule is used to increase the sampling rate by linear interpolation or spline interpolation when the first regulating submodule and the second regulating submodule are called simultaneously, and then reduce the sampling rate by filtering.

[0043] One of the above technical solutions has the following advantages or beneficial effects: 1. The division based on regional features can more finely reflect the local structure of the point cloud, making the subsequent sampling rate adjustment more accurate. This refined division method helps to retain important features in the point cloud, such as edges, corners, etc., while removing redundant information, providing a more valuable data basis for subsequent point cloud processing tasks.

[0044] 2. By calculating the density and curvature of the local area, the sampling rate is dynamically adjusted, thereby reducing the sampling rate in dense areas to reduce redundant information and increasing the sampling rate in sparse areas to retain important features. This adaptive sampling rate adjustment method enables the downsampled point cloud to better reflect the structure and characteristics of the original data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of an embodiment of the present invention.

[0046] Figure 2 It is a schematic diagram of the structure of an embodiment of the system of the present invention. DETAILED DESCRIPTION

[0047] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0048] In the description of the embodiments of the present invention, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0049] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "plurality" is two or more. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0050] like Figures 1-2 As shown, a method for adaptive point cloud sampling adjustment includes the following steps:

[0051] Train feature analysis models;

[0052] Sampling the workpiece at an initial sampling rate to obtain point cloud data of the workpiece, inputting the point cloud data into the feature analysis model to obtain regional features, and dividing the local area based on the regional features;

[0053] The sampling rate of the local area is adjusted based on the regional characteristics.

[0054] In the present invention, the workpiece is first sampled at an initial sampling rate to obtain point cloud data of the workpiece, and then the point cloud data is input into a trained feature analysis model. The point cloud data is analyzed by the training feature analysis model. The training feature analysis model is based on the density and / or curvature of the point cloud data, and then on whether the density or curvature of adjacent point cloud data is in an interval to divide the corresponding local areas. The local areas divided by this method have similar regional features, and therefore can be adopted at the same sampling rate, thereby achieving flexible division of the sampling areas.

[0055] The acquired regional features reflect the characteristics of the region. The division based on regional features can more finely reflect the local structure of the point cloud, making the subsequent sampling rate adjustment more accurate. This refined division method helps to retain important features in the point cloud, such as edges and corners, while removing redundant information, providing a more valuable data foundation for subsequent point cloud processing tasks.

[0056] In addition, the traditional point cloud downsampling method usually adopts a fixed sampling rate, which is difficult to adapt to the density and complexity changes in different areas of the point cloud. The present invention dynamically adjusts the sampling rate by calculating the density and curvature of the local area, thereby reducing the sampling rate in dense areas to reduce redundant information and increasing the sampling rate in sparse areas to retain important features. This adaptive sampling rate adjustment method enables the downsampled point cloud to better reflect the structure and characteristics of the original data.

[0057] Preferably, the training process of the feature analysis model is as follows:

[0058] Step A: Input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point; it allows processing of point cloud datasets with different numbers and feature dimensions, thereby improving the applicability of the model.

[0059] Step B: For each point, find the K nearest neighbors in its local neighborhood, and use a small neural network to learn a K×K transformation matrix X to rearrange the points in the local neighborhood; by finding the K nearest neighbors of each point, the model can focus on the local context information of each point, which is crucial for capturing the local structure in the point cloud. In order to further capture the accuracy of regional features, it is also necessary to learn the transformation matrix so that the model can adaptively adjust the arrangement of points.

[0060] Applying a standard convolution operation to the rearranged points to extract local features, and after extracting the local features, aggregating the local features to form global features;

[0061] Step C: constructing a plurality of said small neural networks;

[0062] The output global features of the previous small neural network are used as the input of the next small neural network;

[0063] In the present invention, a plurality of small neural networks are stacked to form a corresponding layered structure. Through gradual extraction, the model of the present invention can extract deeper information, which is convenient for subsequent point cloud data analysis to obtain accurate regional features.

[0064] Step D: After multiple small neural networks, the final global features are obtained;

[0065] Passing the final global features to the fully connected layer for classification or regression tasks;

[0066] Step E: Use the standard back-propagation algorithm and optimizer (such as Adam, SGD) to train the model; and optimize the weight parameters of the small neural network by minimizing the loss function.

[0067] Preferably, the regional characteristics include density and / or curvature of the local area.

[0068] Preferably, the initial sampling rate of the local area is adjusted as follows:

[0069] If the density is lower than the density threshold, reducing the sampling rate;

[0070] If the density is higher than the density threshold, increasing the sampling rate;

[0071] If the curvature is higher than the curvature threshold, increasing the sampling rate;

[0072] If the curvature is below a curvature threshold, the sampling rate is reduced.

[0073] The present invention can automatically adjust the sampling rate according to the density and curvature of the local area. This means that in areas with dense data or high curvature, the sampling rate will be increased accordingly to more accurately capture the details of these areas. On the contrary, in areas with sparse data or low curvature, the sampling rate will be reduced to reduce unnecessary calculations, thereby improving overall efficiency. Especially when processing large-scale data sets, this adaptive adjustment can significantly reduce computing time and resource consumption.

[0074] Preferably, if the adjustment instruction includes both an increase instruction and a decrease instruction, the sampling rate is increased by linear interpolation or spline interpolation, and then the sampling rate is decreased by filtering.

[0075] By first using linear interpolation or spline interpolation methods to increase the sampling rate, this process can ensure that while increasing the density of point cloud data, the continuity and smoothness of the point cloud data are maintained, effectively avoiding data jumps or distortion, and laying a solid foundation for subsequent precise processing. Next, the sampling rate is reduced using filtering methods, which can not only intelligently remove redundant data according to actual needs, reduce the computational burden, and improve processing efficiency, but also cleverly retain key feature information and prevent the loss of details caused by excessive downsampling, thereby ensuring data simplification while also ensuring data quality and integrity. This solution achieves fine control and efficient processing of sampling rate adjustment, providing strong support for the optimized management and in-depth analysis of point cloud data.

[0076] A system for adaptive point cloud sampling and adjustment, using the method for adaptive point cloud sampling and adjustment, comprising a training module, a collection module and an adjustment module;

[0077] The training module is used to train the feature analysis model;

[0078] The acquisition module is used to sample the workpiece at an initial sampling rate, obtain point cloud data of the workpiece, input the point cloud data into the feature analysis model to obtain regional features, and divide the local area based on the regional features;

[0079] The adjustment module is used to adjust the sampling rate of the local area based on the regional characteristics.

[0080] Preferably, the training module includes a first submodule, a second submodule, a third submodule, a fourth submodule and a fifth submodule;

[0081] The first submodule is used to input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point;

[0082] The second submodule is used to find the K nearest neighboring points in the local neighborhood of each point, learn a K×K transformation matrix X through a small neural network, and rearrange the points in the local neighborhood; apply standard convolution operations to the rearranged points to extract local features, and after extracting the local features, aggregate the local features to form global features;

[0083] The third submodule is used to construct a plurality of said small neural networks;

[0084] The output global features of the previous small neural network are used as the input of the next small neural network;

[0085] The fourth submodule is used to obtain the final global features after passing through multiple small neural networks;

[0086] Passing the final global features to the fully connected layer for classification or regression tasks;

[0087] The fifth submodule is used to train the model using a standard back-propagation algorithm and optimizer (such as Adam, SGD); and optimize the weight parameters of the small neural network by minimizing the loss function.

[0088] Preferably, the regulating module includes a first regulating submodule and a second regulating submodule;

[0089] The first regulating submodule is used to reduce the sampling rate when the density is lower than the density threshold, or reduce the sampling rate when the curvature is lower than the curvature threshold;

[0090] The second regulating submodule is used to increase the sampling rate when the density is higher than the density threshold, or to increase the sampling rate when the curvature is higher than the curvature threshold.

[0091] Preferably, the regulating module further includes a third regulating submodule;

[0092] The third regulating submodule is used to increase the sampling rate by linear interpolation or spline interpolation when the first regulating submodule and the second regulating submodule are called simultaneously, and then reduce the sampling rate by filtering.

[0093] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0094] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A method for adaptive point cloud sampling adjustment, characterized in that: The steps include: Train feature analysis models; Sampling the workpiece at an initial sampling rate to obtain point cloud data of the workpiece, inputting the point cloud data into the feature analysis model to obtain regional features, and dividing the local area based on the regional features; The sampling rate of the local area is adjusted based on the regional characteristics.

2. The method for adaptive point cloud sampling adjustment according to claim 1, characterized in that: The training process of the feature analysis model is as follows: Step A: input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point; Step B: For each point, find the K nearest neighbors in its local neighborhood, and learn a K×K transformation matrix X through a small neural network to rearrange the points in the local neighborhood; Applying a standard convolution operation to the rearranged points to extract local features, and after extracting the local features, aggregating the local features to form global features; Step C: constructing a plurality of said small neural networks; The output global features of the previous small neural network are used as the input of the next small neural network; Step D: After multiple small neural networks, the final global features are obtained; Passing the final global features to the fully connected layer for classification or regression tasks; Step E: Train the model using the standard back-propagation algorithm and optimizer; and optimize the weight parameters of the small neural network by minimizing the loss function.

3. The method for adaptive point cloud sampling adjustment according to claim 1, characterized in that: The regional characteristics include the density and / or curvature of the local region.

4. The method for adaptive point cloud sampling adjustment according to claim 3, characterized in that: The initial sampling rate of the local area is adjusted as follows: If the density is lower than the density threshold, reducing the sampling rate; If the density is higher than the density threshold, increasing the sampling rate; If the curvature is higher than the curvature threshold, increasing the sampling rate; If the curvature is below a curvature threshold, the sampling rate is reduced.

5. The method for adaptive point cloud sampling adjustment according to claim 4, characterized in that: If the adjustment instruction includes both an increase instruction and a decrease instruction, the sampling rate is increased by linear interpolation or spline interpolation, and then the sampling rate is reduced by filtering.

6. A system for adaptive point cloud sampling and adjustment, characterized in that: A method for adaptive point cloud sampling and adjustment using any one of claims 1 to 5, comprising a training module, a collection module and an adjustment module; The training module is used to train the feature analysis model; The acquisition module is used to sample the workpiece at an initial sampling rate, obtain point cloud data of the workpiece, input the point cloud data into the feature analysis model to obtain regional features, and divide the local area based on the regional features; The adjustment module is used to adjust the sampling rate of the local area based on the regional characteristics.

7. The system for adaptive point cloud sampling and adjustment according to claim 6, characterized in that: The training module includes a first submodule, a second submodule, a third submodule, a fourth submodule and a fifth submodule; The first submodule is used to input a training point cloud dataset, wherein the training point cloud dataset is an N×D matrix, where N is the number of points and D is the feature dimension of each point; The second submodule is used to find the K nearest neighboring points in the local neighborhood of each point, learn a K×K transformation matrix X through a small neural network, and rearrange the points in the local neighborhood; apply standard convolution operations to the rearranged points to extract local features, and after extracting the local features, aggregate the local features to form global features; The third submodule is used to construct a plurality of said small neural networks; The output global features of the previous small neural network are used as the input of the next small neural network; The fourth submodule is used to obtain the final global features after passing through multiple small neural networks; Passing the final global features to the fully connected layer for classification or regression tasks; The fifth submodule is used to train the model using the standard back-propagation algorithm and optimizer; and optimize the weight parameters of the small neural network by minimizing the loss function.

8. The system for adaptive point cloud sampling and adjustment according to claim 6, characterized in that: The regulating module includes a first regulating submodule and a second regulating submodule; The first regulating submodule is used to reduce the sampling rate when the density is lower than the density threshold, or reduce the sampling rate when the curvature is lower than the curvature threshold; The second regulating submodule is used to increase the sampling rate when the density is higher than the density threshold, or to increase the sampling rate when the curvature is higher than the curvature threshold.

9. The system for adaptive point cloud sampling and adjustment according to claim 6, characterized in that: The regulating module also includes a third regulating submodule; The third regulating submodule is used to increase the sampling rate by linear interpolation or spline interpolation when the first regulating submodule and the second regulating submodule are called simultaneously, and then reduce the sampling rate by filtering.

Images