Dynamic Gaussian filtering point cloud optimization method and system based on local features

Abstract

The invention discloses a dynamic Gaussian filtering point cloud optimization method and system based on local features, and relates to the technical field of point cloud data processing and three-dimensional computer vision, and the method comprises the steps: carrying out the preprocessing of original laser radar point cloud data, dynamically adjusting the sampling rate, constructing an index structure, and carrying out the dynamic Gaussian filtering based on the preprocessed point cloud data. The method comprises the steps of extracting geometric features of local neighborhoods, fusing the geometric features into weighted feature vectors, dynamically adjusting kernel function parameters of Gaussian filtering according to the weighted feature vectors, generating an adaptive filtering window, performing weighted Gaussian filtering processing on point cloud data by using the dynamically adjusted kernel function parameters, and outputting denoised point cloud data. According to the method, efficient point cloud retrieval is realized through dynamic sampling and index construction, multi-feature optimization weight is fused to generate an adaptive filtering window, de-noising and detail reservation are balanced, spatial distribution is recovered, missing points are interpolated and filled, a complete point cloud data set is generated, and data quality and processing precision are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of point cloud data processing and three-dimensional computer vision, and particularly relates to a dynamic Gaussian filtering point cloud optimization method and system based on local features. BACKGROUND

[0002] Point cloud data, as an important form of three-dimensional spatial data representation, has been widely used in many fields, including but not limited to robot navigation, autonomous driving, three-dimensional reconstruction, and cultural heritage protection. With the rapid development of LiDAR technology, the acquisition of point cloud data has become more efficient and accurate. However, the original point cloud data often contains a large amount of noise and redundant information, which brings great challenges to subsequent data processing and analysis. In order to improve the quality of point cloud data, researchers have proposed various filtering methods, among which Gaussian filtering has received extensive attention due to its simplicity and effectiveness. Traditional Gaussian filtering methods usually use a fixed size of filtering window and fixed kernel function parameters, which often fail to achieve good results when dealing with point cloud data with complex geometric features, leading to loss of detail features or excessive smoothing.

[0003] Point cloud filtering methods based on local features have gradually become a research hotspot. By extracting local geometric features of point cloud, the filtering parameters can be dynamically adjusted to achieve more refined filtering results. However, existing filtering methods based on local features still have some shortcomings. First, most methods use static neighborhood search strategy when extracting local features, which may lead to inaccurate feature extraction in areas with large changes in point cloud density. Second, existing methods often lack effective weight optimization mechanism when fusing multiple features, resulting in unsatisfactory filtering results. In addition, traditional filtering methods tend to over-smooth in high-curvature areas, losing important detail features. SUMMARY

[0004] In view of the above existing problems, the present application provides a dynamic Gaussian filtering point cloud optimization method and system based on local features to solve the problems of inaccurate feature extraction in areas with large changes in point cloud density, lack of effective weight optimization mechanism, resulting in unsatisfactory filtering results, and over-smoothing in high-curvature areas, losing important detail features.

[0005] To solve the above technical problems, a dynamic Gaussian filtering point cloud optimization method based on local features is proposed, which includes,

[0006] The original laser radar point cloud data is preprocessed, the sampling rate is dynamically adjusted, and an index structure is constructed; based on the preprocessed point cloud data, the geometric features of the local neighborhood are extracted and fused into a weighted feature vector, the kernel function parameters of the Gaussian filter are dynamically adjusted according to the weighted feature vector, and an adaptive filtering window is generated; the point cloud data is weighted and Gaussian filtered by using the dynamically adjusted kernel function parameters, and the denoised point cloud data is output.

[0007] As a preferred scheme of the local feature-based dynamic Gaussian filtering point cloud optimization method, the preprocessing includes dynamically setting the sampling rate according to the local density value of each point in the point cloud, dividing the three-dimensional point cloud space into uniform grid cells, mapping the coordinates of each point to a unique grid index, establishing a mapping relationship between the grid index and the point set, and retrieving the neighborhood point set of the target point.

[0008] The dynamic setting of the sampling rate includes reducing the sampling rate when the density is higher than a preset threshold and increasing the sampling rate when the density is lower than the preset threshold, and non-uniform compression of the data volume.

[0009] As a preferred scheme of the local feature-based dynamic Gaussian filtering point cloud optimization method, the extraction of the geometric features of the local neighborhood includes extracting the geometric features of the local neighborhood based on the preprocessed point cloud data by using the dynamic index structure.

[0010] The extraction of the geometric features of the local neighborhood further includes calculating the curvature by surface fitting, estimating the normal vector direction based on principal component analysis, and statistically analyzing the density of the neighborhood point distribution, and normalizing the curvature, the normal vector direction deviation and the density, and fusing the weighted feature vectors.

[0011] As a preferred scheme of the local feature-based dynamic Gaussian filtering point cloud optimization method, the fusion of the weighted feature vectors includes optimizing the weight coefficients of the curvature, the normal vector direction deviation and the density based on a training data set, minimizing the filtering error, linearly combining the normalized feature values according to the optimized weights to form a multi-dimensional feature vector, and dynamically adjusting the filtering parameters.

[0012] The dynamic index structure includes dividing the point cloud space along the X, Y and Z axes into cubic grid cells of equal size, setting the grid size according to the point cloud resolution, mapping the three-dimensional coordinates of the grid to one-dimensional index values, and storing the coordinate information of all points in each grid, and retrieving the neighborhood point set of the target point by retrieving the grid to which the target point belongs and the adjacent 26 grids.

[0013] As a preferred scheme of the local feature based dynamic Gaussian filtering point cloud optimization method, the dynamic adjustment of the kernel function parameters of the Gaussian filter includes dynamically adjusting the standard deviation parameter of the Gaussian filter kernel function according to the curvature value and the density value in the weighted feature vector, and generating an adaptive filtering window matched with the local geometric feature.

[0014] The generation of the adaptive filtering window matched with the local geometric feature includes setting the coverage range of the filtering window according to the normalized curvature value, and adjusting the kernel function weight coefficient in combination with the normalized density value.

[0015] The coverage range of the filtering window includes setting a curvature threshold, and when the local curvature exceeds the threshold, the standard deviation of the filtering window is reduced to 50% to 80% of the original value; a density threshold is set, and when the local density is lower than the threshold, the standard deviation of the filtering window is expanded to 120% to 150% of the original value.

[0016] As a preferred scheme of the local feature based dynamic Gaussian filtering point cloud optimization method, the weighted Gaussian filtering process includes weighted Gaussian filtering processing of the point cloud data by using the adaptive filtering window, and weighted average calculation of the neighborhood points to retain the detail features of the high curvature area.

[0017] The weighted average calculation includes weighted summation of the three-dimensional coordinates of the neighborhood points according to the Gaussian kernel function weight, and multiplying the weight by the attenuation coefficient for the neighborhood points in the high curvature area, i.e., the curvature is greater than the threshold.

[0018] The formula for generating the Gaussian filter kernel function is:

[0019]

[0020] Wherein, G(x, y) is the Gaussian filter kernel function, x and y are the coordinates of the kernel function center point, σ is the standard deviation of the Gaussian filter kernel function, exp is the natural exponential function, and π is the circular constant.

[0021] The formula for weighted Gaussian filtering processing is:

[0022]

[0023] Wherein, w j is the weight of the i-th neighborhood point, P i and P j are the coordinates of two points in the neighborhood, ||P i -P j | is the Euclidean distance between the point P i and the point P j , ρ j is the density value of the j-th point, and G is the Gaussian function. For the new position of the filtered point P i N is the total number of points in the neighborhood, and i and j are variable indices.

[0024] As a preferred scheme of the local feature-based dynamic Gaussian filtering point cloud optimization method, the output denoised point cloud data includes matching the filtered point coordinates with the grid mapping relationship in the original index structure, restoring the spatial distribution before downsampling, filling the missing points due to downsampling through an interpolation algorithm, and generating a final point cloud data set.

[0025] As a preferred scheme of the local feature-based dynamic Gaussian filtering point cloud optimization system, the system comprises a preprocessing module, a feature extraction module, a parameter dynamic adjustment module, and a filtering and optimization module.

[0026] The preprocessing module is configured to dynamically adjust the sampling rate according to the local density, divide the three-dimensional point cloud space into uniform grids, generate grid indexes through a hash function, and locate the neighborhood point set.

[0027] The feature extraction module is configured to fit a local surface based on the neighborhood point set, calculate a curvature value through a second derivative to represent the degree of local bending, calculate a covariance matrix through principal component analysis to calculate the ratio of the number of neighborhood points to the volume, calculate a local density value, and linearly combine the curvature, the direction deviation of the normal vector, and the density after normalization into a multi-dimensional feature vector according to an optimization weight.

[0028] The parameter dynamic adjustment module is configured to dynamically adjust the Gaussian filtering parameters according to the feature vector.

[0029] The filtering and optimization module is configured to perform weighted average calculation on the neighborhood points based on a dynamic kernel function, match the filtered point cloud with the original index, fill the missing points due to downsampling through interpolation, and generate a denoised data set.

[0030] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the local feature-based dynamic Gaussian filtering point cloud optimization method when executing the computer program.

[0031] A computer-readable storage medium stores a computer program, and the computer program implements the steps of the local feature-based dynamic Gaussian filtering point cloud optimization method when executed by a processor.

[0032] The application has the beneficial effects that: the application realizes non-uniform compression and efficient retrieval of point cloud data by preprocessing original laser radar point cloud data, dynamically adjusting a sampling rate, and constructing an index structure; based on the preprocessed point cloud data, geometric features of a local neighborhood are extracted and fused into a weighted feature vector, a kernel function parameter of a Gaussian filter is dynamically adjusted according to the weighted feature vector, and an adaptive filtering window is generated; the point cloud data is weighted and Gaussian filtered by using the dynamically adjusted kernel function parameter, and the point cloud data after denoising is output, the standard deviation parameter of the Gaussian filter kernel function is dynamically adjusted according to the curvature value and the density value in the weighted feature vector, and an adaptive filtering window matched with the local geometric features is generated; by setting a curvature threshold and a density threshold, the standard deviation of the filtering window can be reduced in a high-curvature area to retain more detailed features, and the standard deviation of the filtering window can be increased in a low-density area to enhance the denoising effect; by restoring the original spatial distribution and interpolating to fill in missing points, a complete point cloud data set can be generated, and the integrity and detailed features of the point cloud data are retained. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative effort.

[0034] Figure 1 The application provides a local feature-based dynamic Gaussian filtering point cloud optimization method.

[0035] Figure 2 The application provides a local feature-based dynamic Gaussian filtering point cloud optimization system. DETAILED DESCRIPTION

[0036] In order to make the above-mentioned purposes, features and advantages of the application more apparent and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative effort should be within the protection scope of the application.

[0037] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from the description, and those skilled in the art can make similar generalizations without departing from the concept of the application, therefore the application is not limited to the specific embodiments disclosed below.

[0038] Secondly, the "one embodiment" or "embodiment" referred to herein is intended to mean a specific feature, structure, or characteristic under at least one implementation of the application. The "in one embodiment" appearing in various places in the specification are not all referring to the same embodiment nor are they mutually exclusive of other embodiments.

[0039] The application is described in detail in conjunction with the schematic drawings. In the detailed description of the embodiments of the application, the sectional view of the device structure is partially enlarged without the general proportion for the convenience of illustration, and the schematic drawings are only examples which should not limit the scope of protection of the application herein. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in the actual manufacture.

[0040] Meanwhile, in the description of the application, it should be noted that the terms "upper, lower, inner and outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the application. In addition, the terms "first, second or third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0041] Unless otherwise specifically defined and limited, the terms "mounting, connecting, connection" in the application should be understood broadly, for example: it can be fixed connection, detachable connection or integral connection; it can also be mechanical connection, electrical connection or direct connection, it can also be indirectly connected through intermediate medium, or it can be the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0042] Embodiment 1, refer to Figure 1 As the first embodiment of the application, the embodiment provides a local feature-based dynamic Gaussian filtering point cloud optimization method, comprising:

[0043] S1: pre-processing the original laser radar point cloud data, dynamically adjusting the sampling rate, and constructing an index structure.

[0044] Further, the pre-processing includes performing a downsampling operation, and dynamically setting the sampling rate according to the local density value of each point in the point cloud, dividing the three-dimensional point cloud space into uniform grid cells, mapping the coordinates of each point to a unique grid index, establishing a mapping relationship between the grid index and the point set, and retrieving the neighborhood point set of the target point.

[0045] In the embodiment of the application, the dynamic adjustment of the sampling rate includes reducing the sampling rate when the density is higher than the preset threshold, increasing the sampling rate when the density is lower than the preset threshold, and performing non-uniform compression of the data volume.

[0046] In an alternative embodiment, the dynamic adjustment of the sampling rate comprises presetting a global density threshold, dividing the point cloud into high-density areas (density ≥ threshold) and low-density areas (density < threshold), using a fixed low sampling rate (such as 50%) for the high-density areas, randomly deleting some points, using a fixed high sampling rate for the low-density areas, using a KD tree, and accelerating the neighborhood search through the tree structure.

[0047] In another alternative embodiment, the dynamic adjustment of the sampling rate comprises dividing the point cloud space into a uniform cubic grid, each grid size being fixed (0.5m×0.5m×0.5m), randomly retaining one point in each grid, deleting the remaining points, storing the point cloud based on the grid division, and retrieving the points in the adjacent grid during neighborhood query.

[0048] It should be noted that the downsampling operation utilizes density-based adaptive downsampling and dynamically adjusts the sampling rate according to the density difference, which is expressed by the formula:

[0049]

[0050] where ρ i is the density value of the i-th point, P i and P j are the coordinates of two points in the neighborhood, ||P i -P j || is the Euclidean distance between point P i and point P j , N is the total number of points in the neighborhood, i and j are variable indices, τ is the bandwidth of the Gaussian kernel function, exp is the natural exponential function, ρ max is the maximum density value, s i is the sampling rate of the i-th point.

[0051] Furthermore, the construction of the index structure comprises dynamic index construction based on a hash table. The adaptive downsampling technique can dynamically adjust the sampling rate according to the local density, reduce the loss of key information, and improve the data quality. The dynamic index construction based on a hash table can significantly improve the efficiency of neighborhood query, especially for large-scale point cloud data. The formula is expressed as:

[0052]

[0053] where h(P i ) is the hash dynamic index of point P i , Δ is the size of the grid, a, b, and c are random hash coefficients, x', y', and z' are the coordinate vectors of point P i , and is the floor function.

[0054] S2: based on the pre-processed point cloud data, extracting the geometric features of the local neighborhood, and fusing into a weighted feature vector, dynamically adjusting the kernel function parameters of the Gaussian filter according to the weighted feature vector, and generating an adaptive filtering window.

[0055] Further, the extraction of the geometric features of the local neighborhood comprises, based on the pre-processed point cloud data, extracting the geometric features of the local neighborhood by using a dynamic index structure;

[0056] The extraction of the geometric features of the local neighborhood further comprises calculating the curvature by surface fitting, estimating the normal vector direction based on principal component analysis, and statistically calculating the density of the neighborhood point distribution, and normalizing the curvature, the normal vector direction deviation and the density, and fusing the weighted feature vector.

[0057] The K-neighborhood algorithm is used to calculate the local neighborhood to realize the local feature extraction of the points, wherein in order to realize efficient feature indexing, the selection of K value in the application adopts a density adaptive adjustment method, and the formula is as follows:

[0058]

[0059] Wherein, K i is the K value of the i-th point, K min is the minimum value of K, K max is the maximum value of K, ρ i is the density value of the i-th point, ρ max is the maximum density value.

[0060] The local geometric features of the point cloud are extracted for describing the local shape and structure of the point cloud, the least square method is used to fit the local plane, the principal component analysis based on the local neighborhood points is used to estimate the normal vector, the curvature is calculated, the covariance matrix of the local neighborhood is calculated, and the density of the points in the local neighborhood is calculated by using the density calculation. According to the extracted features, the curvature, the normal vector and the density are normalized, and are combined into a weighted feature vector, and the curvature calculation formula is as follows:

[0061]

[0062] Wherein, is the curvature of the i-th point, n is the normal vector, H is the Hessian matrix, and ||n|| is the length of the normal vector.

[0063] The covariance matrix of the local neighborhood is calculated as follows:

[0064]

[0065] Wherein, P j is the coordinate vector of the j-th point in the local neighborhood, is the centroid of the local neighborhood, N is the total number of points in the neighborhood, C is the covariance matrix of the local neighborhood, and T is the transpose;

[0066] The density of points in the local neighborhood is calculated as:

[0067]

[0068] Among them, ρ i is the density value of the i-th point, N is the total number of points in the neighborhood, V is the volume of the local neighborhood, and r is the neighborhood radius.

[0069] In an embodiment of the present application, calculating the curvature includes fitting a local surface based on a neighborhood point set and calculating the curvature through a second-order derivative.

[0070] In an optional embodiment, the curvature calculation includes selecting K nearest neighbor points of the target point, calculating the variance of the Z coordinate (height) as the curvature approximation, and mapping the height difference variance to the interval 0, 1 as a simplified curvature feature.

[0071] In another optional embodiment, the curvature calculation includes estimating the normal vector directions of the target point and the neighboring points through principal component analysis (PCA), calculating the average angle between the normal vector of the target point and the normal vector of the neighboring points, the larger the angle, the higher the curvature, and normalizing the average angle value into a curvature feature.

[0072] Furthermore, the fusion of weighted feature vectors includes optimizing the weight coefficients of curvature, normal vector direction deviation and density according to the training data set to minimize the filtering error, linearly combining the normalized feature values ​​according to the optimized weights to form a multidimensional feature vector, and dynamically adjusting the filtering parameters;

[0073] The dynamic index structure includes dividing the point cloud space into cubic grids of equal size along the X, Y, and Z axes. The grid size is set according to the point cloud resolution, mapping the three-dimensional coordinates of the grid into a one-dimensional index value, and storing the coordinate information of all points in each grid. By retrieving the grid to which the target point belongs and the 26 adjacent grids, the neighborhood point set of the target point is obtained.

[0074] It should be noted that the dynamic adjustment of the kernel function parameters of the Gaussian filter includes dynamically adjusting the standard deviation parameters of the Gaussian filter kernel function according to the curvature value and density value in the weighted feature vector to generate an adaptive filter window that matches the local geometric features;

[0075] In the embodiments of the present application, the adaptive filtering window matched with the local geometric features comprises: setting the coverage of the filtering window according to the normalized curvature value, and adjusting the kernel function weight coefficient in combination with the normalized density value; the coverage of the filtering window comprises: setting a curvature threshold, when the local curvature exceeds the threshold, reducing the standard deviation of the filtering window to 50% to 80% of the original value; setting a density threshold, when the local density is lower than the threshold, expanding the standard deviation of the filtering window to 120% to 150% of the original value.

[0076] In an optional embodiment, the adaptive filtering window matched with the local geometric features comprises: presetting a curvature threshold, determining a high curvature region higher than the threshold, reducing the filtering window to 70% of the original value in the high curvature region, expanding the filtering window to 130% of the original value in a low curvature region, and adjusting the window according to the curvature threshold in all regions.

[0077] The formula for dynamically calculating the standard deviation of the Gaussian filtering kernel function is:

[0078] σ=α·u+β·ρ

[0079] wherein, α and β are weight coefficients, u is the local curvature, ρ is the local density, and σ is the standard deviation of the Gaussian filtering kernel function.

[0080] A nonlinear weight adjustment method is introduced to improve the weight coefficients, and the formula is:

[0081]

[0082] wherein, α ’ and β ’ are improved weight coefficients, θ c is the curvature threshold, θ d is the density threshold, and k is a smoothing factor.

[0083] S3: performing weighted Gaussian filtering processing on the point cloud data by using the dynamically adjusted kernel function parameters, and outputting the denoised point cloud data.

[0084] Further, the weighted Gaussian filtering processing comprises: performing weighted Gaussian filtering processing on the point cloud data by using the adaptive filtering window, and performing weighted average calculation on the neighborhood points to retain the detail features of the high curvature region.

[0085] The weighted average calculation comprises: performing weighted summation on the three-dimensional coordinates of the neighborhood points according to the Gaussian kernel function weight, and multiplying the weight by a decay coefficient for the neighborhood points in the high curvature region, i.e., the curvature is greater than the threshold.

[0086] The formula for generating the Gaussian filtering kernel function is:

[0087]

[0088] wherein G(x, y) is a Gaussian filter kernel function, x and y are coordinates of the kernel function center, σ is a standard deviation of the Gaussian filter kernel function, exp is a natural exponential function, and π is a circular constant;

[0089] The weighted Gaussian filter processing formula is represented as:

[0090]

[0091] wherein w j is a weight of the i-th neighborhood point, P i and P j are coordinates of two points in the neighborhood, ||P i -P j || is a Euclidean distance between the point P i and the point P j , ρ j is a density value of the j-th point, G is a Gaussian function, is a new position of the filtered point P i , N is a total number of points in the neighborhood, and i and j are variable indices.

[0092] It should be noted that the output denoised point cloud data includes matching the filtered point coordinates with the grid mapping relationship in the original index structure, restoring the spatial distribution before downsampling, filling the missing points due to downsampling by an interpolation algorithm, and generating a final point cloud data set.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

[0094] Embodiment 2, referring to Figure 2 is a second embodiment of the present application, which provides a dynamic Gaussian filter point cloud optimization system based on local features, including a preprocessing module, a feature extraction module, a parameter dynamic adjustment module, and a filtering and optimization module.

[0095] The preprocessing module is used to dynamically adjust the sampling rate according to the local density, divide the three-dimensional point cloud space into uniform grids, generate grid indexes through a hash function, and locate the neighborhood point set.

[0096] The feature extraction module is configured to fit a local surface based on a neighborhood point set, calculate a curvature value by a second derivative, represent a local bending degree, calculate a local density value by principal component analysis and a covariance matrix, and a ratio of a neighborhood point number and a volume, and linearly combine the curvature, a normal vector direction deviation and the density into a multi-dimensional feature vector after normalization according to an optimization weight.

[0097] The parameter dynamic adjustment module is configured to dynamically adjust a Gaussian filter parameter according to the feature vector.

[0098] The filtering and optimization module is configured to perform a weighted average calculation on the neighborhood points based on a dynamic kernel function, match the filtered point cloud with an original index, fill in missing points caused by down-sampling by interpolation, and generate a denoised data set.

[0099] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

[0100] Embodiment 3, the third embodiment of the present application, which is different from the first two embodiments is:

[0101] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application which essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0102] The logic and / or steps represented in the flow diagrams and / or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. Just by way of example, a computer-readable medium can be any device or apparatus that can store and convey instructions for execution by the instruction execution system, apparatus, or device. With respect to the present description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0103] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer.

[0104] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies can be used: a combination of discrete logic circuits having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having logic gates, field programmable gate arrays (FPGA), or other components, in any suitable combination.

Claims

1. A dynamic Gaussian filtering point cloud optimization method based on local features, characterized by: include, Preprocess the raw lidar point cloud data, dynamically adjust the sampling rate, and build an index structure; Based on the preprocessed point cloud data, the geometric features of the local neighborhood are extracted and fused into a weighted feature vector. The kernel function parameters of the Gaussian filter are dynamically adjusted according to the weighted feature vector to generate an adaptive filtering window. The point cloud data is processed by weighted Gaussian filtering using the dynamically adjusted kernel function parameters, and the denoised point cloud data is output.

2. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 1, wherein: The preprocessing includes dynamically setting a sampling rate based on the local density value of each point in the point cloud, dividing the three-dimensional point cloud space into uniform grid cells, mapping the coordinates of each point to a unique grid index, establishing a mapping relationship between the grid index and the point set, and retrieving the neighborhood point set of the target point; Dynamically setting the sampling rate includes lowering the sampling rate when the density is higher than a preset threshold, and increasing the sampling rate when the density is lower than the preset threshold, thereby performing non-uniform compression of the data volume.

3. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 2, wherein: The extracting of the geometric features of the local neighborhood includes extracting the geometric features of the local neighborhood using a dynamic index structure based on the preprocessed point cloud data; The extraction of geometric features of the local neighborhood also includes calculating curvature through surface fitting, estimating normal vector direction based on principal component analysis, counting neighborhood point distribution density, normalizing curvature, normal vector direction deviation and density, and fusing weighted feature vectors.

4. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 3, wherein: The fusion of weighted feature vectors includes optimizing the weight coefficients of curvature, normal vector direction deviation and density according to the training data set, minimizing the filtering error, linearly combining the normalized feature values ​​according to the optimized weights to form a multidimensional feature vector, and dynamically adjusting the filtering parameters; The dynamic index structure includes dividing the point cloud space into cubic grids of equal size along the X, Y, and Z axes. The grid size is set according to the point cloud resolution, mapping the three-dimensional coordinates of the grid into a one-dimensional index value, and storing the coordinate information of all points in each grid. By retrieving the grid to which the target point belongs and the 26 adjacent grids, the neighborhood point set of the target point is obtained.

5. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 4, wherein: The dynamically adjusting the kernel function parameters of the Gaussian filter includes dynamically adjusting the standard deviation parameters of the Gaussian filter kernel function according to the curvature value and the density value in the weighted feature vector to generate an adaptive filter window that matches the local geometric features; Generating an adaptive filtering window that matches the local geometric features includes setting the coverage of the filtering window according to the normalized curvature value and adjusting the kernel function weight coefficient in combination with the normalized density value; The coverage of the filter window includes setting a curvature threshold. When the local curvature exceeds the threshold, the standard deviation of the filter window is reduced to 50% to 80% of the original value; Set the density threshold. When the local density is lower than the threshold, expand the standard deviation of the filter window to 120% to 150% of the original value.

6. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 5, wherein: The weighted Gaussian filtering process includes performing weighted Gaussian filtering on the point cloud data using an adaptive filtering window, calculating the weighted average of the neighborhood points, and retaining the detailed features of the high curvature area; The weighted average calculation includes weighted summation of the three-dimensional coordinates of the neighborhood points according to the weight of the Gaussian kernel function. For the neighborhood points in the high curvature area, i.e., the curvature is greater than the threshold, the weight is multiplied by the attenuation coefficient; The formula for generating the Gaussian filter kernel function is expressed as: Where G(x,y) is the Gaussian filter kernel function, x and y are the coordinates of the midpoint of the kernel function, σ is the standard deviation of the Gaussian filter kernel function, exp is the natural exponential function, and π is pi; The formula for weighted Gaussian filtering is expressed as: Among them, w j is the weight of the i-th neighborhood point, P i and P j are the coordinates of two points in the neighborhood, ||P i -P j || is point P i and point P j The Euclidean distance between j is the density value of the jth point, G is the Gaussian function, is the point P after filtering i The new position of , N is the total number of points in the neighborhood, and i and j are variable indices.

7. The method for optimizing point clouds using dynamic Gaussian filtering based on local features according to claim 6, wherein: The output of denoised point cloud data includes matching the filtered point coordinates with the grid mapping relationship in the original index structure, restoring the spatial distribution before downsampling, filling in the points missing due to downsampling through an interpolation algorithm, and generating a final point cloud dataset.

8. A system using the local feature-based dynamic Gaussian filtering point cloud optimization method according to any one of claims 1 to 7, characterized in that: It includes preprocessing module, feature extraction module, parameter dynamic adjustment module and filtering and optimization module; The preprocessing module is used to dynamically adjust the sampling rate according to the local density, divide the three-dimensional point cloud space into a uniform grid, generate a grid index through a hash function, and locate the neighborhood point set; The feature extraction module is used to fit the local surface based on the neighborhood point set, calculate the curvature value through the second-order derivative to characterize the local curvature, calculate the covariance matrix through principal component analysis, count the ratio of the number of neighborhood points to the volume, calculate the local density value, and normalize the curvature, normal vector direction deviation and density, and then linearly combine them into a multidimensional feature vector according to the optimized weight; The parameter dynamic adjustment module is used to dynamically adjust the Gaussian filter parameters according to the feature vector; The filtering and optimization module is used to perform weighted average calculation on neighborhood points based on a dynamic kernel function, match the filtered point cloud with the original index, fill in the downsampling missing points through interpolation, and generate a denoised data set.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the processor implements the steps of a dynamic Gaussian filtering point cloud optimization method based on local features according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a dynamic Gaussian filtering point cloud optimization method based on local features according to any one of claims 1 to 7 are implemented.

Images