Point Cloud Denoising Processing Method and Device Based on Morphological Recognition

Abstract

The present invention provides a point cloud noise reduction processing method and device based on morphological recognition, which relates to the technical field of point cloud data processing. The method includes obtaining point cloud data; sequentially performing preprocessing and preliminary filtering on the point cloud data to obtain preliminary filtered data; performing morphological recognition on the preliminary filtered data; marking potential noise points in the clustering analysis data based on neighborhood search and skewness calculation; performing vibration simulation on the potential noise points based on a preset vibration model, and filtering out the potential noise points with vibration amplitudes exceeding a preset amplitude threshold in the point cloud data to obtain noise-reduced point cloud data. Through a point cloud noise reduction processing method based on morphological recognition, the present invention realizes more accurate identification of noise in point cloud data, makes noise filtering more effective, retains the detailed information of the point cloud, and improves the processing efficiency of point cloud data.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing, and more specifically, to a method and device for point cloud noise reduction processing based on shape recognition. Background Art

[0002] Traditional point cloud noise reduction processing algorithms may have misjudgments in noise point recognition, especially when dealing with complex scenes and high-noise data. In addition, conventional noise filtering methods such as simple mean filtering or median filtering may not be able to effectively remove complex noise and may even lose details. Complex point cloud processing algorithms may have a large amount of calculation and a long processing time, making it difficult to meet the requirements of real-time or large-scale data processing. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, device, equipment and readable storage medium for point cloud noise reduction processing based on shape recognition to improve the above problems. To achieve the above purpose, the technical solutions adopted by the present invention are as follows:

[0004] In a first aspect, the present application provides a method for point cloud noise reduction processing based on shape recognition, which is characterized by including:

[0005] Obtain point cloud data;

[0006] Perform preprocessing and preliminary filtering on the point cloud data in sequence to obtain preliminarily filtered data;

[0007] Perform shape recognition on the preliminarily filtered data. The shape recognition includes calculating the initial normal vector of each point in the preliminarily filtered data to obtain point cloud data with initial normal vectors, extracting features of the point cloud data with initial normal vectors based on a multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain the main feature vectors of each point, performing reaction diffusion processing on the main feature vectors to obtain the shape feature vectors of each point, performing interpolation processing on the shape feature vectors based on the bilinear difference method to obtain the enhanced shape feature vectors of each point, extracting the sparse feature vectors of the enhanced shape feature vectors of each point through sparse representation and performing clustering analysis on the sparse feature vectors of each point to obtain clustering analysis data;

[0008] Mark potential noise points in the clustering analysis data based on neighborhood search and skewness calculation;

[0009] Perform vibration simulation on the potential noise points based on a preset vibration model, and filter out the potential noise points in the point cloud data whose vibration amplitude exceeds a preset amplitude threshold to obtain noise-reduced point cloud data.

[0010] Second aspect, the present application also provides a point cloud noise reduction processing device based on morphological recognition, including: an acquisition module, the acquisition module is used to acquire point cloud data;

[0011] A preliminary filtering module, the preliminary filtering module is used to perform preprocessing and preliminary filtering on the point cloud data in sequence to obtain preliminary filtered data;

[0012] A morphological recognition module, the morphological recognition module is used to perform morphological recognition on the preliminary filtered data. The morphological recognition includes calculating the initial normal vector of each point in the preliminary filtered data to obtain point cloud data with initial normal vectors, performing feature extraction on the point cloud data with initial normal vectors based on a multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain the main feature vector of each point, performing reaction diffusion processing on the main feature vector to obtain the morphological feature vector of each point, performing interpolation processing on the morphological feature vector based on the bilinear difference method to obtain the enhanced morphological feature vector of each point, extracting the sparse feature vector of the enhanced morphological feature vector of each point through sparse representation and performing clustering analysis on the sparse feature vector of each point to obtain clustering analysis data;

[0013] A marking module, the marking module is used to mark potential noise points in the clustering analysis data based on neighborhood search and skewness calculation;

[0014] A filtering module, the filtering module is used to perform vibration simulation on the potential noise points based on a preset vibration model and filter out the potential noise points in the point cloud data whose vibration amplitude exceeds a preset amplitude threshold to obtain noise-reduced point cloud data.

[0015] Third aspect, the present application also provides a point cloud noise reduction processing device based on morphological recognition, including:

[0016] A memory, used to store a computer program;

[0017] A processor, used to implement the steps of the point cloud noise reduction processing method based on morphological recognition when executing the computer program.

[0018] Fourth aspect, the present application also provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned point cloud noise reduction processing method based on morphological recognition are implemented.

[0019] The beneficial effects of the present invention are as follows: Through a point cloud noise reduction processing method based on morphology recognition, the present invention realizes more accurate recognition of the noise in point cloud data, making the noise filtering more effective, while retaining the detailed information of the point cloud and improving the processing efficiency of the point cloud data.

[0020] Other features and advantages of the present invention will be described in the subsequent specification, and, in part, will become apparent from the specification or be understood by implementing the embodiments of the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification, claims, and drawings. Brief Description of the Drawings

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other relevant drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic flowchart of the point cloud noise reduction processing method based on morphology recognition described in the embodiments of the present invention;

[0023] Figure 2 It is a schematic logic diagram of the point cloud noise reduction processing method based on morphology recognition described in the embodiments of the present invention;

[0024] Figure 3 It is a schematic structural diagram of the point cloud noise reduction processing device based on morphology recognition described in the embodiments of the present invention.

[0025] Reference numerals in the figure: 800, point cloud noise reduction processing device based on morphology recognition; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. Detailed Embodiments

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0027] It should be noted that similar reference numerals and letters denote similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0028] Embodiment 1:

[0029] This embodiment provides a point cloud noise reduction processing method based on morphological recognition.

[0030] See Figure 1 and Figure 2 , in which the figures show that this method includes step S100, step S200, step S300, step S400, and step S500.

[0031] Step S100: Obtain point cloud data;

[0032] The point cloud data can be the following types of data:

[0033] LiDAR scan data: used for applications such as environmental perception, autonomous driving, and terrain mapping;

[0034] Depth camera data: for example, depth images collected using devices such as Kinect and Intel RealSense, converted into three-dimensional point cloud data;

[0035] Structured light scan data: three-dimensional point cloud data obtained using structured light technology, commonly used in industrial inspection, 3D modeling, and reverse engineering;

[0036] Stereo vision data: point cloud data generated by taking images from multiple perspectives through a stereo camera system and using a stereo matching algorithm, commonly used in robot navigation, object recognition, and 3D reconstruction;

[0037] Sonar data: three-dimensional point cloud data for underwater environmental perception. Sonar point cloud data usually contains three-dimensional coordinates and intensity information, used for applications such as marine mapping and underwater archaeology.

[0038] Step S200: Perform preprocessing and preliminary filtering on the point cloud data in sequence to obtain preliminarily filtered data;

[0039] Performing preprocessing on the point cloud data includes:

[0040] Performing normalization processing on the point cloud data to obtain the normalized point cloud data;

[0041] Downsample the normalized point cloud data based on voxel filtering to obtain the downsampled point cloud data;

[0042] Calculate the bounding box of the downsampled point cloud data, and scale the coordinates of each point in the downsampled point cloud data to a preset scale range according to the bounding box.

[0043] Perform preliminary filtering on the point cloud data in sequence, including:

[0044] Calculate the average distance from each point in the point cloud data to other points, and filter out the points in the point cloud data where the average distance is greater than the preset distance threshold to obtain the first point cloud data;

[0045] Calculate the normal vector angle between each point in the first point cloud data and other points, and filter out the points in the first point cloud data where the normal vector angle is greater than the preset angle threshold to obtain the second point cloud data;

[0046] Calculate the neighborhood density of each point in the second point cloud data according to the preset neighborhood range, and filter out the points in the second point cloud data where the neighborhood density is less than the preset density threshold to obtain the third point cloud data;

[0047] Calculate the color difference between each point in the third point cloud data and other points, and filter out the points in the third point cloud data where the color difference is greater than the preset color difference threshold to obtain the preliminary filtered data.

[0048] Step S300: Perform morphological recognition on the preliminary filtered data. The morphological recognition includes calculating the initial normal vector of each point in the preliminary filtered data to obtain the point cloud data with initial normal vectors, extracting features of the point cloud data with initial normal vectors based on the multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain the main feature vectors of each point, performing reaction diffusion processing on the main feature vectors to obtain the morphological feature vectors of each point, performing interpolation processing on the morphological feature vectors based on the bilinear interpolation method to obtain the enhanced morphological feature vectors of each point, extracting the sparse feature vectors of the enhanced morphological feature vectors of each point through sparse representation and performing clustering analysis on the sparse feature vectors of each point to obtain the clustering analysis data;

[0049] In the step S300:

[0050] For each point p i , find its nearest neighbor points and calculate the initial normal vector of each point p i through the PCA method;

[0051] Extract features from the point cloud data with initial normal vectors based on the multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, including:

[0052] Preset multiple scales, where the scale is the neighborhood radius;

[0053] Preset multiple scales r1, r2,..., r n 。

[0054] Calculate the curvature and the current normal vector of each point in the point cloud data with initial normal vectors at each scale;

[0055] For each point p i , find the set of its neighborhood points at each scale r j where p is the neighborhood point. k

[0056]

[0057] Calculate the covariance matrix of the neighborhood point set and calculate the eigenvalues and eigenvectors of the covariance matrix. The eigenvector corresponding to the smallest eigenvalue is the current normal vector

[0058] The curvature is calculated as follows:

[0059]

[0060] where λ min , λ mid , λ max represent the smallest eigenvalue, the middle eigenvalue, and the largest eigenvalue of the covariance matrix, respectively.

[0061] At each scale, calculate the angle between the current normal vector and the initial normal vector;

[0062] The angle is calculated as follows:

[0063]

[0064] Aggregate the curvature and the angle of each point at each scale into a feature vector to obtain the multi-scale geometric feature f(p i ) of each point.

[0065]

[0066] ​They are the scales r1, r2,......, r in sequence n The corresponding curvature They are the scales r1, r2,......, r in sequence n The corresponding included angle

[0067] The present invention introduces a multi-scale geometric analysis method to extract morphological features at different scales and enhance the ability to capture details of point cloud data

[0068] Perform curvature flow processing on the multi-scale geometric features to obtain the main feature vectors, including

[0069] Construct a multi-scale geometric feature curvature flow equation according to the curvature, the included angle and a preset control parameter

[0070]

[0071] Wherein And Are the gradients of the curvature and the included angle at different scales respectively, and the α j And β j Are both control parameters, representing the influence weights of the curvature and the included angle on the flow at different scales, and t represents time

[0072] Calculate the first change amount Δf(p of the multi-scale geometric features at each first time step Δt i );

[0073] The calculation formula of the first change amount Δf(p i ) is as follows

[0074]

[0075] The formula for updating the multi-scale geometric feature vector is as follows

[0076] f(p i , t+Δt) = f(p i , t) + Δf(p i )

[0077] f(p i , t+Δt) is the multi-scale geometric feature of point p i At the moment of t+Δt, and f(p i , t) is the multi-scale geometric feature of point P i At the moment of t

[0078] Calculate the norm of the first variation, and determine whether the norm of the first variation satisfies the first preset convergence criterion. If so, stop the iteration and update the multi-scale geometric feature according to the current first variation to obtain the main eigenvector. If not, continue the iteration until the first maximum number of iterations is reached.

[0079] Perform reaction-diffusion processing on the main eigenvector, including:

[0080] Construct a reaction-diffusion equation based on the diffusion coefficient, Laplacian operator, and reaction term function. The Laplacian operator is calculated from the main eigenvectors of the current point and its neighboring points, and the reaction term function is constructed based on the initialized main eigenvector g(p i ), the main eigenvector of the current point, and the reaction coefficient;

[0081] The formula for the reaction-diffusion equation is as follows:

[0082]

[0083] where g(p i ) is the main eigenvector, is the Laplacian operator, R(g(p i )) is the reaction term function, and D is the diffusion coefficient.

[0084] The calculation formula for the Laplacian operator is as follows:

[0085]

[0086] where g(p k ) is the main eigenvector of the neighboring point p k ;

[0087] The formula for the reaction term function is as follows:

[0088] R(g(p i )) = λg(p i )(1 - g(p i ))

[0089] where λ is a constant parameter used to control the intensity or rate of the reaction term.

[0090] Calculate the second variation Δg(p i ) of the main eigenvector at each second time step Δt';

[0091] The calculation formula for the second variation is as follows:

[0092]

[0093] The update of the main eigenvector is as follows:

[0094]

[0095] where g(p i , t′ + Δt′) is the main eigenvector of point p i at time t′ + Δt′, and g(p i , t′) is the main eigenvector of point p i at time t′.

[0096] Calculate the norm of the second variation, and determine whether the norm of the second variation satisfies the second preset convergence criterion. If so, stop the iteration, and update the main eigenvector according to the current second variation to obtain the morphological eigenvector h(pi). If not, continue the iteration until the second maximum number of iterations is reached.

[0097] The present invention combines a reaction-diffusion equation and a curvature flow equation for smoothing, which can not only remove noise but also retain important morphological features of the point cloud data.

[0098] Interpolate the morphological eigenvector based on the bilinear interpolation method to obtain the enhanced morphological eigenvector of each point, including:

[0099] Obtain the position coordinates of each point in the point cloud data;

[0100] Project the position coordinates of each point onto a two-dimensional plane to obtain the two-dimensional point coordinates of each point;

[0101] Construct a regular grid on the two-dimensional plane according to the two-dimensional point coordinates of all points in the point cloud data, and determine the grid cell where each point is located in the regular grid. Each grid cell has four vertices;

[0102] Record the coordinates of each vertex;

[0103] Calculate the distance from the vertex to the point according to the two-dimensional point coordinates and the coordinates of the vertex, and calculate the interpolation weight of the vertex according to the inverse ratio of the distance;

[0104] Calculate the eigenvector of each vertex according to the morphological eigenvector of the point and the interpolation weight;

[0105] The enhanced morphological eigenvector is the sum of the eigenvectors of the four vertices.

[0106] For example, there is a point cloud data set as follows:

[0107] {(2.1, 3.5), (4.2, 5.3), (3.0, 4.8), (5.1, 6.1)}。

[0108] First, project {(2.1, 3.5), (4.2, 5.3), (3.0, 4.8), (5.1, 6.1)} onto a two-dimensional plane. Determine the value range of the abscissa of the point cloud data set as [2, 6] and the value range of the ordinate as [3, 7] according to the projected two-dimensional coordinates, and preset the side length of the grid cells of the regular network to 1;

[0109] Construct a regular network as follows:

[0110] (2, 3), (3, 3), (4, 3), (5, 3), (6, 3)

[0111] (2, 4), (3, 4), (4, 4), (5, 4), (6, 4)

[0112] (2, 5), (3, 5), (4, 5), (5, 5), (6, 5)

[0113] (2, 6), (3, 6), (4, 6), (5, 6), (6, 6)

[0114] (2, 7), (3, 7), (4, 7), (5, 7), (6, 7)

[0115] For example, the point (2.1, 3.5) falls within the grid cells with vertex coordinates (2, 3), (3, 3), (2, 4), and (3, 4).

[0116] For example, for the grid vertex (2, 3), the calculation formula for its interpolation weight w is:

[0117]

[0118] where d((2, 3), (2.1, 3.5)) is the distance from the grid vertex (2, 3) to the point (2.1, 3.5);

[0119] Then, the feature vector f((2, 3)) of the grid vertex (2, 3) is:

[0120] f((2, 3)) = w · h((2.1, 3.5))

[0121] where f((2, 3)) is the feature vector of the grid vertex (2, 3), and h((2.1, 3.5)) is the morphological feature vector of the point (2.1, 3.5)

[0122] Similarly, the feature vectors of other grid vertices can be calculated, and the sum of the feature vectors of the four vertices can be calculated to obtain the enhanced morphological feature vector Z(pi).

[0123] The present invention can refine and enhance morphological features through the double-line difference method.

[0124] Extracting the sparse feature vectors of the enhanced morphological feature vectors through sparse representation includes:

[0125] Randomly select some enhanced morphological feature vectors from all the enhanced morphological feature vectors of the points to construct a dictionary;

[0126] Based on the dictionary and the enhanced morphological feature vectors of each point, calculate the sparse vector of each point based on the LASSO algorithm;

[0127] Extract features from the sparse vectors of each point to obtain the sparse feature vectors of each point.

[0128] For example, construct a dictionary D through some randomly selected enhanced morphological feature vectors;

[0129] The function for calculating the sparse vector of each point based on the dictionary D and the enhanced morphological feature vector Z(pi) of each point based on the LASSO algorithm is as follows:

[0130]

[0131] where a i is the sparse vector, λ is the sparsity control parameter used to balance the reconstruction error and sparsity. By solving the above optimization problem, the sparse vector a i , ||a i ||1 represents the regularization term of a i , and the 1 in ||a i ||1 represents the l1 norm.

[0132] The present invention extracts the sparse features of the point cloud through sparse representation, further enhancing the discrimination ability of the morphological features and making the main morphological features more prominent.

[0133] Step S400: Mark the potential noise points in the clustering analysis data based on neighborhood search and skewness calculation;

[0134] Marking the potential noise points in the clustering analysis data based on neighborhood search and skewness calculation includes:

[0135] Search for the local neighborhood of each point in each cluster of the clustering analysis data based on neighborhood search. The local neighborhood is all the points in the sphere with the point as the center within a preset radius;

[0136] Calculate the skewness of each point based on the local neighborhood, the sparse feature vectors of each point in the local neighborhood, and the local neighborhood;

[0137] Mark the points with skewness exceeding a preset skewness threshold as the potential noise points.

[0138] Mark the potential noise points in the clustering analysis data based on neighborhood search and skewness calculation, including:

[0139] Calculate the mean vector μ of the points in the local neighborhood, and the calculation formula is:

[0140]

[0141] N i ={p m |||p m -pi||≤r}

[0142] Calculate the skewness of the points in the local neighborhood, and the calculation formula is:

[0143]

[0144] where p i is the point in the cluster, P m is the point in the local neighborhood, μ is the mean vector, N i is the local neighborhood, X(P m ) is the sparse feature vector of point P m , r is the preset radius, S i is the skewness, σ i is the standard deviation of the local neighborhood, N i (p i ) is the set of neighborhood points in the local neighborhood.

[0145] The present invention improves the accuracy of noise detection by combining clustering analysis and local skewness analysis.

[0146] Step S500: Perform vibration simulation on the potential noise points based on a preset vibration model, and filter out the potential noise points with vibration amplitude exceeding a preset amplitude threshold in the point cloud data to obtain denoised point cloud data.

[0147] Perform vibration simulation on the potential noise points based on a preset vibration model, and filter out the potential noise points with vibration amplitude exceeding a preset amplitude threshold in the point cloud data, including:

[0148] Initialize vibration simulation parameters, and the vibration simulation parameters include a vibration amplitude threshold, a simulation time step, and a maximum simulation time;

[0149] Apply a perturbation to each of the potential noise points;

[0150] The applied perturbation can be expressed as:

[0151] p i p(t"+Δt") = p i (t") + Δp i

[0152] where t‘’ represents the moment, Δt‘’ represents the time increment, and p i p(t‘’) represents the position of the potential noise point p at time t i p i p(t‘’ + Δt‘’) represents the position of the potential noise point P at time t‘’ + Δt‘’ i p, and Δp i represents the change in the position of point P within the time increment Δt‘’ i .

[0153] Calculate the unit vibration amplitude of the potential noise point within each of the simulation time steps, and calculate the cumulative vibration amplitude based on the maximum simulation time and the unit vibration amplitude;

[0154] Then, within each time step, the calculation formula for the unit vibration amplitude A i (t‘’) is as follows:

[0155] A i (t‘’) = ‖p i (t‘’ + Δt‘’) - p i (t‘’)‖

[0156] For the maximum simulation time T max , the calculation formula for the cumulative vibration amplitude A i is as follows:

[0157]

[0158] Determine whether the cumulative vibration amplitude is greater than a preset amplitude threshold. If so, filter out the potential noise points in the point cloud data whose vibration amplitude exceeds the preset amplitude threshold.

[0159] The present invention further removes noise points and improves the quality of point cloud data through physical simulation vibration filtering, in combination with vibration parameters and a judgment threshold.

[0160] The present invention improves the noise reduction quality and accuracy of point cloud data through the above-mentioned multi-scale geometric feature extraction, curvature flow, reaction diffusion, bilinear interpolation, sparse representation, clustering analysis, and vibration simulation.

[0161] Embodiment 2:

[0162] This embodiment provides a point cloud noise reduction processing device based on morphological recognition. The device includes:

[0163] An acquisition module for acquiring point cloud data;

[0164] A preliminary filtering module for preprocessing and preliminarily filtering the point cloud data in sequence to obtain preliminarily filtered data;

[0165] A morphology recognition module for performing morphology recognition on the preliminarily filtered data. The morphology recognition includes calculating an initial normal vector of each point in the preliminarily filtered data to obtain point cloud data with initial normal vectors, extracting features of the point cloud data with initial normal vectors based on a multi-scale geometric analysis algorithm to obtain multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain a main feature vector of each point, performing reaction diffusion processing on the main feature vector to obtain a morphology feature vector of each point, performing interpolation processing on the morphology feature vector based on a bilinear interpolation method to obtain an enhanced morphology feature vector of each point, extracting a sparse feature vector of the enhanced morphology feature vector of each point through sparse representation and performing clustering analysis on the sparse feature vector of each point to obtain clustering analysis data;

[0166] A marking module for marking potential noise points in the clustering analysis data based on neighborhood search and skewness calculation;

[0167] A filtering module for performing vibration simulation on the potential noise points based on a preset vibration model and filtering out the potential noise points in the point cloud data whose vibration amplitude exceeds a preset amplitude threshold to obtain denoised point cloud data.

[0168] It should be noted that regarding the device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0169] Embodiment 3:

[0170] Corresponding to the above method embodiment, in this embodiment, a point cloud denoising processing device based on morphology recognition is further provided. A point cloud denoising processing device based on morphology recognition described below can be correspondingly referred to with a point cloud denoising processing method based on morphology recognition described above.

[0171] Figure 3 is a block diagram of a point cloud denoising processing device 800 based on morphology recognition shown according to an exemplary embodiment. As Figure 3As shown, the point cloud noise reduction processing device 800 based on morphological recognition may include: a processor 801 and a memory 802. The point cloud noise reduction processing device 800 based on morphological recognition may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0172] Among them, the processor 801 is used to control the overall operation of the point cloud noise reduction processing device 800 based on morphological recognition to complete all or part of the steps in the above-mentioned point cloud noise reduction processing method based on morphological recognition. The memory 802 is used to store various types of data to support the operation of the point cloud noise reduction processing device 800 based on morphological recognition. These data may include, for example, instructions for any application or method operating on the point cloud noise reduction processing device 800 based on morphological recognition, as well as application-related data, such as contact data, sent and received messages, pictures, audio, video, and so on. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. The multimedia component 803 may include a screen and an audio component. Among them, the screen may be a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone, and the microphone is used to receive external audio signals. The received audio signals may be further stored in the memory 802 or sent through the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above other interface modules may be a keyboard, a mouse, buttons, etc. These buttons may be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the point cloud noise reduction processing device 800 based on morphological recognition and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination of one or more of them. Therefore, the corresponding communication component 805 may include: a Wi-Fi module, a Bluetooth module, and an NFC module.

[0173] In an exemplary embodiment, the point cloud noise reduction processing device 800 based on morphological recognition may be implemented by one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components, and is used to execute the above-mentioned point cloud noise reduction processing method based on morphological recognition.

[0174] In another exemplary embodiment, there is also provided a computer-readable storage medium including program instructions, and when the program instructions are executed by a processor, the steps of the above-mentioned point cloud noise reduction processing method based on morphological recognition are implemented. For example, the computer-readable storage medium may be the above-mentioned memory 802 including program instructions, and the above-mentioned program instructions may be executed by the processor 801 of the point cloud noise reduction processing device 800 based on morphological recognition to complete the above-mentioned point cloud noise reduction processing method.

[0175] Embodiment 4:

[0176] Corresponding to the above method embodiment, in this embodiment, there is also provided a readable storage medium, and a readable storage medium described below can be correspondingly referred to with a point cloud noise reduction processing method based on morphological recognition described above.

[0177] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the point cloud noise reduction processing method based on morphological recognition in the above method embodiment are implemented.

[0178] Specifically, the readable storage medium may be various readable storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc that can store program codes.

[0179] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0180] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A point cloud noise reduction processing method based on morphological recognition, characterized in that, Including: Obtain point cloud data; Successively perform preprocessing and preliminary filtering on the point cloud data to obtain preliminarily filtered data; Perform morphological recognition on the preliminarily filtered data. The morphological recognition includes calculating the initial normal vector of each point in the preliminarily filtered data to obtain point cloud data with initial normal vectors, extracting features of the point cloud data with initial normal vectors based on a multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain the main feature vector of each point, performing reaction-diffusion processing on the main feature vector to obtain the morphological feature vector of each point, performing interpolation processing on the morphological feature vector based on bilinear interpolation to obtain the enhanced morphological feature vector of each point, extracting the sparse feature vector of the enhanced morphological feature vector of each point through sparse representation and performing clustering analysis on the sparse feature vector of each point to obtain clustering analysis data; Mark potential noise points in the clustering analysis data based on neighborhood search and skewness calculation; Perform vibration simulation on the potential noise points based on a preset vibration model, and filter out the potential noise points in the point cloud data whose vibration amplitude exceeds a preset amplitude threshold to obtain denoised point cloud data; Among them, performing interpolation processing on the morphological feature vector based on bilinear interpolation to obtain the enhanced morphological feature vector of each point includes: Obtain the position coordinates of each point in the point cloud data; Project the position coordinates of each point onto a two-dimensional plane to obtain the two-dimensional point coordinates of each point; Construct a regular grid on the two-dimensional plane according to the two-dimensional point coordinates of all points in the point cloud data, and determine the grid cell where each point is located in the regular grid. Each grid cell has four vertices; Record the coordinates of each vertex; Calculate the distance from the vertex to the point according to the two-dimensional point coordinates and the coordinates of the vertex, and calculate the interpolation weight of the vertex according to the inverse ratio of the distance; Calculate the feature vector of each vertex according to the morphological feature vector of the point and the interpolation weight; The enhanced morphological feature vector is the sum of the feature vectors of the four vertices.

2. The method for denoising point cloud based on morphology recognition according to claim 1, wherein , performing preprocessing on the point cloud data includes: Perform normalization processing on the point cloud data to obtain the normalized point cloud data; Perform downsampling on the normalized point cloud data based on voxel filtering to obtain the downsampled point cloud data; Calculate the bounding box of the downsampled point cloud data, and scale the coordinates of each point in the downsampled point cloud data to a preset scale range according to the bounding box.

3. The method for denoising point cloud based on morphology recognition according to claim 1, wherein , successively performing preliminary filtering on the point cloud data includes: Calculate the average distance from a point in the point cloud data to other points, and filter out the points in the point cloud data whose average distance is greater than a preset distance threshold to obtain first point cloud data; Calculate the normal vector angle between a point and other points in the first point cloud data, and filter out the points in the first point cloud data whose normal vector angle is greater than a preset angle threshold to obtain second point cloud data; Calculate the neighborhood density of each point in the second point cloud data according to a preset neighborhood range, and filter out the points in the second point cloud data with a neighborhood density less than a preset density threshold to obtain third point cloud data; Calculate the color difference between the points in the third point cloud data and other points, and filter out the points in the third point cloud data with a color difference greater than a preset color difference threshold to obtain the preliminary filtered data.

4. The method for denoising point cloud based on morphological recognition according to claim 1, wherein , Based on a multi-scale geometric analysis algorithm, perform feature extraction on the point cloud data with initial normal vectors to obtain the multi-scale geometric features of each point, including: Preset multiple scales, where the scale is the neighborhood radius; Calculate the curvature and the current normal vector of each point in the point cloud data with initial normal vectors at each scale; At each scale, calculate the angle between the current normal vector and the initial normal vector; Aggregate the curvature and the angle of each point at each scale into a feature vector to obtain the multi-scale geometric feature of each point.

5. The method for point cloud noise reduction processing based on morphological recognition according to claim 4, characterized in that , Perform curvature flow processing on the multi-scale geometric features to obtain the main feature vector, including: Construct a multi-scale geometric feature curvature flow equation according to the curvature, the angle, and a preset control parameter; Calculate the first change amount of the multi-scale geometric features at each first time step according to the multi-scale geometric feature curvature flow equation; Calculate the norm of the first change amount, and determine whether the norm of the first change amount satisfies a first preset convergence criterion. If so, stop the iteration, and update the multi-scale geometric features according to the current first change amount to obtain the main feature vector. If not, continue the iteration until the first maximum number of iterations is reached.

6. The method for point cloud noise reduction processing based on morphological recognition according to claim 1, wherein , Perform reaction-diffusion processing on the main feature vector, including: Construct a reaction-diffusion equation according to the diffusion coefficient, the Laplace operator, and the reaction term function. The Laplace operator is calculated through the main feature vectors of the current point and its neighborhood points, and the reaction term function is constructed according to the initialized main feature vectors, the main feature vector of the current point, and the reaction coefficient; Calculate the second change amount of the main feature vector at each second time step according to the reaction-diffusion equation; Calculate the norm of the second change amount, and determine whether the norm of the second change amount satisfies a second preset convergence criterion. If so, stop the iteration, and update the main feature vector according to the current second change amount to obtain the morphological feature vector. If not, continue the iteration until the second maximum number of iterations is reached.

7. The method for denoising point cloud based on morphological recognition according to claim 1, wherein , Extract the sparse feature vector of the enhanced morphological feature vector through sparse representation, including: Randomly select some enhanced morphological feature vectors from the enhanced morphological feature vectors of all points to construct a dictionary; Based on the dictionary and the enhanced morphological feature vector of each point, calculate the sparse vector of each point based on the LASSO algorithm; Perform feature extraction on the sparse vector of each point to obtain the sparse feature vector of each point.

8. The method for denoising point cloud based on morphological recognition according to claim 2, wherein , Mark the potential noise points in the clustering analysis data based on neighborhood search and skewness calculation, including: Search for the local neighborhood of each point in each cluster of the clustering analysis data based on neighborhood search. The local neighborhood is all the points in a sphere with the point as the center within a preset radius. Calculate the skewness of each point according to the local neighborhood, the sparse feature vector of each point in the local neighborhood, and the standard deviation of the local neighborhood. Mark the points with skewness exceeding the preset skewness threshold as the potential noise points.

9. A point cloud noise reduction processing device based on morphological recognition, characterized in that Include: An acquisition module for acquiring point cloud data. A preliminary filtering module for preprocessing and preliminarily filtering the point cloud data in sequence to obtain preliminarily filtered data. A morphology recognition module for performing morphology recognition on the preliminarily filtered data. The morphology recognition includes calculating the initial normal vector of each point in the preliminarily filtered data to obtain point cloud data with initial normal vectors, extracting features of the point cloud data with initial normal vectors based on a multi-scale geometric analysis algorithm to obtain the multi-scale geometric features of each point, performing curvature flow processing on the multi-scale geometric features to obtain the main feature vector of each point, performing reaction-diffusion processing on the main feature vector to obtain the morphology feature vector of each point, performing interpolation processing on the morphology feature vector based on bilinear interpolation to obtain the enhanced morphology feature vector of each point, and performing clustering analysis on the sparse feature vectors of the enhanced morphology feature vectors of each point through sparse representation to obtain clustering analysis data. A marking module for marking potential noise points in the clustering analysis data based on neighborhood search and skewness calculation. A filtering module for performing vibration simulation on the potential noise points based on a preset vibration model and filtering out the potential noise points with vibration amplitudes exceeding the preset amplitude threshold in the point cloud data to obtain denoised point cloud data. Among them, performing interpolation processing on the morphology feature vector based on bilinear interpolation to obtain the enhanced morphology feature vector of each point includes: Obtain the position coordinates of each point in the point cloud data. Project the position coordinates of each point onto a two-dimensional plane to obtain the two-dimensional point coordinates of each point. Construct a regular grid on the two-dimensional plane according to the two-dimensional point coordinates of all the points in the point cloud data and determine the grid cell where each point is located in the regular grid. Each grid cell has four vertices. Record the coordinates of each vertex. Calculate the distance from the vertex to the point according to the two-dimensional point coordinates and the coordinates of the vertex, and calculate the interpolation weight of the vertex according to the inverse of the distance. Calculate the feature vector of each vertex according to the morphology feature vector of the point and the interpolation weight. The enhanced morphology feature vector is the sum of the feature vectors of the four vertices.

Images