A shipborne laser radar feature transformation and denoising method under a harbor navigation scene

Abstract

This invention provides a method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios, belonging to the field of ship berthing perception and ship autonomous positioning technology. The method includes: loading raw data from the shipborne lidar and raw data from the shipborne inertial measurement unit; integrating the ship's acceleration and angular velocity to calculate the ship's velocity increment, displacement increment, and rotation increment; using a ship motion compensation algorithm based on spherical interpolation to correct point cloud rotation distortion; using a ship motion compensation algorithm based on bilinear interpolation to correct point cloud displacement distortion; converting the motion-compensated 3D point cloud into a 2D feature matrix based on the shipborne lidar laser emission sequence, including depth features, angle features, and intensity features; dividing the noise reduction range into near-field, mid-field, and far-field based on the point cloud depth features of the port environment; and using a point cloud noise reduction algorithm based on spatial intensity and depth constraints to filter out low-frequency noise and isolated noise in the near-field, mid-field, and far-field.

Description

Technical Field

[0001] This invention relates to the field of ship berthing perception and ship autonomous positioning technology, and more particularly to a method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios. Background Technology

[0002] The performance of a ship's intelligent navigation decision-making system is directly related to the measurement accuracy of its onboard sensing equipment. During navigation in port waters, ships are subject to the coupled influence of multiple factors, including their own inertia, instantaneous speed changes, and wave disturbances, resulting in complex six-degree-of-freedom motion. This dynamic motion characteristic leads to significant spatial positional deviations in the 3D point cloud data acquired by the shipborne lidar, making it difficult to accurately reflect the true geometric features of targets. Simultaneously, the unique port environment, including water mist scattering, wave reflection, and specular reflection from the sea surface, introduces spatially correlated non-Gaussian noise into the original point cloud. This noise exhibits a non-uniform spatial distribution, being dense in the near field and discrete in the far field, and displays significant non-stationary characteristics in the temporal dimension, severely impacting the accuracy of the sensing system's distance measurement of obstacles and the point cloud segmentation effect.

[0003] Current mainstream point cloud motion compensation and filtering algorithms are primarily designed for land-based intelligent transportation scenarios, based on the core assumption that the platform has a relatively stable motion state and limited environmental noise. However, ships navigating in harbors experience greater roll and pitch movements, and the dynamic noise generated on the sea surface exhibits significant spatial non-uniformity and temporal abrupt changes. Existing Kalman filter-based motion compensation methods struggle to accurately model the complex kinematic characteristics of ships, while traditional statistical filtering algorithms are prone to over-filtering out effective target point clouds when dealing with dynamic sea surface noise. Furthermore, while deep learning-based point cloud processing methods improve noise suppression capabilities to some extent, they suffer from high computational complexity and insufficient real-time performance, making it difficult to meet the real-time requirements of ship autonomous navigation systems for processing perceived data. Summary of the Invention

[0004] To address the technical problems of point cloud distortion, insufficient noise processing, and unacceptable raw data quality in existing technologies, this invention provides a feature transformation and noise reduction method for shipborne lidar in port navigation scenarios. This invention primarily utilizes a ship motion compensation algorithm based on spherical interpolation and bilinear interpolation to correct point cloud rotation and displacement distortion. It then converts the 3D point cloud into a 2D feature matrix using a laser emission sequence. Based on the depth features of the port environment, it divides the area into near-field, mid-field, and far-field noise reduction regions. A regional adaptive multi-constraint point cloud noise reduction algorithm is employed to filter out low-frequency and isolated noise in each region, thereby improving the data quality of shipborne lidar, enhancing anti-interference capabilities, and increasing the accuracy of target distance measurement and clustering segmentation in the ship's perception system.

[0005] The technical means employed in this invention are as follows:

[0006] A method for feature transformation and noise reduction of shipborne lidar in a harbor navigation scenario includes: S1. Load the original data from the shipborne lidar and the original data from the shipborne inertial measurement unit, integrate the ship's acceleration and angular velocity, and calculate the ship's velocity increment, displacement increment, and rotation increment. S2. A ship motion compensation algorithm based on spherical interpolation is used to correct point cloud rotation distortion. S3. A ship motion compensation algorithm based on bilinear interpolation is used to correct point cloud displacement distortion. S4. Based on the laser emission sequence of the shipborne lidar, the motion-compensated 3D point cloud is converted into a 2D feature matrix, including depth features, angle features, and intensity features. S5. Based on the point cloud depth features of the port environment, the noise reduction range is divided into near field, mid field and far field. S6. A point cloud noise reduction algorithm based on spatial intensity and depth constraints is adopted to filter out low-frequency noise and isolated noise in the near field, mid field and far field.

[0007] Further, step S1 includes: S11. Initialize the ship's current attitude, read the raw data and start and end timestamps of the shipborne lidar, read the raw data and start and end timestamps of the shipborne inertial measurement unit, traverse all shipborne inertial measurement unit timestamps within the time period of the shipborne lidar start and end timestamps, and store the shipborne inertial measurement unit data at the corresponding time. S12. Load and integrate the ship acceleration data from the shipborne inertial measurement unit to obtain the ship velocity data, and integrate the ship velocity data to obtain the ship displacement data. S13. Load and integrate the ship's angular velocity data from the shipborne inertial measurement unit to obtain the ship's rotation increment.

[0008] Further, step S2 includes: S21. Based on the principle of spherical interpolation, construct the rotation compensation matrix as follows:

[0009] in, Indicates the increment of ship rotation. The rotation matrix representing the current attitude of the ship. The rotation matrix representing the ship's attitude at the current moment. The rotation matrix representing the ship's attitude at the next moment. Indicates the current sampling timestamp. Indicates the next sampling timestamp. Indicates the sampling timestamp. Represents matrix exponentiation. Represents the matrix logarithm operation; S22. Transform the rotation compensation matrix to the shipborne lidar coordinate system for the original point cloud. Perform rotational compensation:

[0010] in, This represents the extrinsic parameter from the shipborne inertial measurement unit coordinate system to the shipborne lidar coordinate system. This represents the extrinsic parameters from the shipborne lidar coordinate system to the shipborne inertial measurement unit coordinate system. This represents the point cloud after rotation compensation.

[0011] Further, step S3 includes: S31. Based on the current displacement matrix and displacement increment, calculate the displacement matrix for the next moment:

[0012] in, The displacement matrix represents the current attitude of the ship. It represents the change in the ship's displacement within a sampling period; S32. Calculate the bilinear interpolation coefficients, including the forward interpolation ratio and the backward interpolation ratio, using the following formula:

[0013]

[0014] in, Indicates the forward interpolation ratio, Indicates the backward interpolation ratio; S33. Based on the principle of bilinear interpolation, construct the displacement compensation matrix as follows:

[0015] in, Represents the displacement compensation matrix; S34. Transform the displacement compensation matrix to the shipborne lidar coordinate system and process the point cloud after rotation compensation. Perform displacement compensation:

[0016] in, This represents a 3D point cloud after motion compensation.

[0017] Further, step S4 includes: S41. Read the number of laser beams of the shipborne lidar. and the number of emission points per harness ; Obtain 3D point clouds after complete motion compensation ; S42, Generate 3 sizes. The 2D feature matrices are the depth feature matrices. Angular feature matrix Intensity feature matrix ; S43. Traverse the 3D point cloud after complete motion compensation. Extract the depth value of each point Angle value and intensity value Then, normalization is performed, and the laser emission order of the shipborne lidar is stored in the depth feature matrix. Angular feature matrix Intensity feature matrix The corresponding position.

[0018] Further, step S5 includes: S51. Generate three index arrays, namely the near-field index array. Midfield index array and far-field index array ; S52, Traversing the Depth Feature Matrix If the current point index depth value If it is a near-field point, it is stored in the near-field index array. ;like If it is a midfield point, it is determined to be a midfield point and stored in the midfield index array. ;like If it is a far-field point, it is stored in the index array far-field. .

[0019] Further, step S6 includes: S61. Construct a labeling function for regional adaptive point cloud denoising. Near-field points are subject to a single intensity constraint, mid-field points are subject to a dual constraint of intensity and spatial abrupt change, and far-field points are subject to a dual constraint of intensity and depth spatial continuity. The noise determination rule is as follows:

[0020] in, Indicates the tagging function, Indicates the intensity threshold. Indicates the mutation threshold. Indicates index Mean intensity within the neighborhood Indicates index Depth values ​​at each point within the neighborhood; S62, Traverse the near-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. If so, then mark that point as a noise point; S63, Traverse the midfield index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And the current point intensity value is related to its position in the background. The absolute value of the difference between the mean intensity values ​​within the middle neighborhood is greater than or equal to the mutation threshold. If so, then mark that point as a noise point; S64, Traverse the far-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And its depth feature matrix If the depth value of each point in the middle neighborhood is 0, then the point is marked as a noise point; S65. Filter out the point cloud data marked as noise points and output the denoised depth feature matrix, angle feature matrix and intensity feature matrix.

[0021] Compared with the prior art, the present invention has the following advantages: 1. This invention provides a feature transformation and noise reduction method for shipborne lidar in port navigation scenarios, which solves the problems of point cloud rotation and displacement distortion caused by complex ship movements in the port, as well as the interference of non-uniform noise in the port environment. It improves the data quality and anti-interference capability of shipborne lidar, enhances the accuracy of target distance measurement and clustering segmentation of the ship perception system, and ensures the real-time performance of the algorithm to meet the application requirements of ship autonomous navigation.

[0022] 2. This invention combines spherical interpolation and bilinear interpolation algorithms to achieve point cloud motion compensation, specifically correcting rotational and displacement distortions caused by ship motion. Compared with the traditional Kalman filtering method, it can more accurately model the complex six-degree-of-freedom motion characteristics of ships in the harbor, with higher point cloud distortion correction accuracy and effectively restoring the true geometric features of the target.

[0023] 3. This invention converts 3D point clouds into 2D feature matrices arranged according to the laser emission sequence, preserving the original acquisition features of the lidar. Compared with unordered 3D point cloud processing, it reduces the computational complexity of subsequent noise reduction algorithms, improves the real-time performance of data processing, and meets the real-time requirements of ship autonomous navigation.

[0024] 4. Based on depth features, this invention divides point clouds into near, mid, and far fields, and adopts a regional adaptive noise reduction strategy with spatial, intensity, and depth constraints. Differentiated judgment rules are formulated for the noise distribution characteristics of different regions, which solves the problem of excessive loss of effective point clouds when filtering non-uniform noise in port environments. While filtering out low-frequency noise and isolated noise, it retains effective target point clouds to the greatest extent, improving the distance measurement accuracy of targets and the accuracy of point cloud clustering and segmentation of the ship perception system.

[0025] 5. The algorithm of this invention is based on geometric interpolation and feature constraints. It does not require complex deep learning model training and a large amount of computing power. It is easy to implement in engineering and can be directly deployed in shipborne embedded sensing systems, which has good engineering application value.

[0026] In summary, this invention effectively solves the problems of point cloud distortion, poor non-uniform noise processing, and insufficient real-time performance of existing ship-borne lidar in harbors, significantly improving the perception performance of shipborne lidar and providing reliable perception data support for intelligent navigation and autonomous berthing of ships in harbors. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the method of the present invention.

[0029] Figure 2 This is a schematic diagram for calculating the motion state of a ship, provided as an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of feature transformation provided in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the point cloud noise reduction range division provided in an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0033] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0034] like Figure 1 As shown, this invention provides a method for feature transformation and noise reduction of shipborne lidar in a harbor navigation scenario, including: S1. Load the original data from the shipborne lidar and the original data from the shipborne inertial measurement unit, integrate the ship's acceleration and angular velocity, and calculate the ship's velocity increment, displacement increment, and rotation increment. S2. A ship motion compensation algorithm based on spherical interpolation is used to correct point cloud rotation distortion. S3. A ship motion compensation algorithm based on bilinear interpolation is used to correct point cloud displacement distortion. S4. Based on the laser emission sequence of the shipborne lidar, the motion-compensated 3D point cloud is converted into a 2D feature matrix, including depth features, angle features, and intensity features. S5. Based on the point cloud depth features of the port environment, the noise reduction range is divided into near field, mid field and far field. S6. A point cloud noise reduction algorithm based on spatial intensity and depth constraints is adopted to filter out low-frequency noise and isolated noise in the near field, mid field and far field.

[0035] In specific implementation, as a preferred embodiment of the present invention, such as Figure 2 As shown, step S1 includes: S11. Initialize the ship's current attitude Read raw data from shipborne lidar and start and end timestamps Read raw data from the shipborne inertial measurement unit with start and end timestamps Traverse the start and end timestamps of the shipborne lidar to Timestamps of all shipborne inertial measurement units within the time period Store the shipborne inertial measurement unit data at the corresponding time. S12. Load and integrate the ship acceleration data from the shipborne inertial measurement unit. Obtain ship speed data Ship displacement data is obtained by integrating the ship speed data. ; S13. Load and integrate the ship's angular velocity data from the shipborne inertial measurement unit. The ship rotation increment was obtained. .

[0036] In a specific implementation, as a preferred embodiment of the present invention, step S2 includes: S21. Based on the principle of spherical interpolation, construct the rotation compensation matrix as follows:

[0037] in, Indicates the increment of ship rotation. The rotation matrix representing the current attitude of the ship. The rotation matrix representing the ship's attitude at the current moment. The rotation matrix representing the ship's attitude at the next moment. Indicates the current sampling timestamp. Indicates the next sampling timestamp. Indicates the sampling timestamp. Represents matrix exponentiation. Represents the matrix logarithm operation; S22. Transform the rotation compensation matrix to the shipborne lidar coordinate system for the original point cloud. Perform rotational compensation:

[0038] in, This represents the extrinsic parameter from the shipborne inertial measurement unit coordinate system to the shipborne lidar coordinate system. This represents the extrinsic parameters from the shipborne lidar coordinate system to the shipborne inertial measurement unit coordinate system. This represents the point cloud after rotation compensation.

[0039] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. Based on the current displacement matrix and displacement increment, calculate the displacement matrix for the next moment:

[0040] in, The displacement matrix represents the current attitude of the ship. It represents the change in the ship's displacement within a sampling period; S32. Calculate the bilinear interpolation coefficients, including the forward interpolation ratio and the backward interpolation ratio, using the following formula:

[0041]

[0042] in, Indicates the forward interpolation ratio, Indicates the backward interpolation ratio; S33. Based on the principle of bilinear interpolation, construct the displacement compensation matrix as follows:

[0043] in, Represents the displacement compensation matrix; S34. Transform the displacement compensation matrix to the shipborne lidar coordinate system and process the point cloud after rotation compensation. Perform displacement compensation:

[0044] in, This represents a 3D point cloud after motion compensation.

[0045] In specific implementation, as a preferred embodiment of the present invention, such as Figure 3 As shown, step S4 includes: S41. Read the number of laser beams of the shipborne lidar. and the number of emission points per harness ; Obtain 3D point clouds after complete motion compensation ; S42, Generate 3 sizes. The 2D feature matrices are the depth feature matrices. Angular feature matrix Intensity feature matrix ; S43. Traverse the 3D point cloud after complete motion compensation. Extract the depth value of each point Angle value and intensity value Then, normalization is performed, and the laser emission order of the shipborne lidar is stored in the depth feature matrix. Angular feature matrix Intensity feature matrix The corresponding position.

[0046] In specific implementation, as a preferred embodiment of the present invention, such as Figure 4 As shown, step S5 includes: S51. Generate three index arrays, namely the near-field index array. Midfield index array and far-field index array ; S52, Traversing the Depth Feature Matrix If the current point index depth value If it is a near-field point, it is stored in the near-field index array. ;like If it is a midfield point, it is determined to be a midfield point and stored in the midfield index array. ;like If it is a far-field point, it is stored in the index array far-field. .

[0047] In a specific implementation, as a preferred embodiment of the present invention, step S6 includes: S61. Construct a labeling function for regional adaptive point cloud denoising. Near-field points are subject to a single intensity constraint, mid-field points are subject to a dual constraint of intensity and spatial abrupt change, and far-field points are subject to a dual constraint of intensity and depth spatial continuity. The noise determination rule is as follows:

[0048] in, Indicates the tagging function, Indicates the intensity threshold. Indicates the mutation threshold. Indicates index Mean intensity within the neighborhood Indicates index Depth values ​​at each point within the neighborhood; S62, Traverse the near-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. If so, then mark that point as a noise point; S63, Traverse the midfield index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And the current point intensity value is related to its position in the background. The absolute value of the difference between the mean intensity values ​​within the middle neighborhood is greater than or equal to the mutation threshold. If so, then mark that point as a noise point; S64, Traverse the far-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And its depth feature matrix If the depth value of each point in the middle neighborhood is 0, then the point is marked as a noise point; S65. Filter out the point cloud data marked as noise points and output the denoised depth feature matrix, angle feature matrix and intensity feature matrix.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for feature transformation and noise reduction of shipborne lidar in a harbor navigation scenario, characterized in that, include: S1. Load the original data from the shipborne lidar and the original data from the shipborne inertial measurement unit, integrate the ship's acceleration and angular velocity, and calculate the ship's velocity increment, displacement increment, and rotation increment. S2. A ship motion compensation algorithm based on spherical interpolation is used to correct point cloud rotation distortion. S3. A ship motion compensation algorithm based on bilinear interpolation is used to correct point cloud displacement distortion. S4. Based on the laser emission sequence of the shipborne lidar, the motion-compensated 3D point cloud is converted into a 2D feature matrix, including depth features, angle features, and intensity features. S5. Based on the point cloud depth features of the port environment, the noise reduction range is divided into near field, mid field and far field. S6. A point cloud noise reduction algorithm based on spatial intensity and depth constraints is adopted to filter out low-frequency noise and isolated noise in the near field, mid field and far field.

2. The method for feature transformation and noise reduction of shipborne lidar in a port navigation scenario according to claim 1, characterized in that, Step S1 includes: S11. Initialize the ship's current attitude, read the raw data and start and end timestamps of the shipborne lidar, read the raw data and start and end timestamps of the shipborne inertial measurement unit, traverse all shipborne inertial measurement unit timestamps within the time period of the shipborne lidar start and end timestamps, and store the shipborne inertial measurement unit data at the corresponding time. S12. Load and integrate the ship acceleration data from the shipborne inertial measurement unit to obtain the ship velocity data, and integrate the ship velocity data to obtain the ship displacement data. S13. Load and integrate the ship's angular velocity data from the shipborne inertial measurement unit to obtain the ship's rotation increment.

3. The method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios according to claim 1, characterized in that, Step S2 includes: S21. Based on the principle of spherical interpolation, construct the rotation compensation matrix as follows: in, Indicates the increment of ship rotation. The rotation matrix representing the current attitude of the ship. The rotation matrix representing the ship's attitude at the current moment. The rotation matrix representing the ship's attitude at the next moment. Indicates the current sampling timestamp. Indicates the next sampling timestamp. Indicates the sampling timestamp. Represents matrix exponentiation. Represents the matrix logarithm operation; S22. Transform the rotation compensation matrix to the shipborne lidar coordinate system for the original point cloud. Perform rotational compensation: in, This represents the extrinsic parameter from the shipborne inertial measurement unit coordinate system to the shipborne lidar coordinate system. This represents the extrinsic parameters from the shipborne lidar coordinate system to the shipborne inertial measurement unit coordinate system. This represents the point cloud after rotation compensation.

4. The method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios according to claim 1, characterized in that, Step S3 includes: S31. Based on the current displacement matrix and displacement increment, calculate the displacement matrix for the next moment: in, The displacement matrix represents the current attitude of the ship. It represents the change in the ship's displacement within a sampling period; S32. Calculate the bilinear interpolation coefficients, including the forward interpolation ratio and the backward interpolation ratio, using the following formula: in, Indicates the forward interpolation ratio, Indicates the backward interpolation ratio; S33. Based on the principle of bilinear interpolation, construct the displacement compensation matrix as follows: in, Represents the displacement compensation matrix; S34. Transform the displacement compensation matrix to the shipborne lidar coordinate system, and process the point cloud after rotation compensation. Perform displacement compensation: in, This represents a 3D point cloud after motion compensation.

5. The method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios according to claim 1, characterized in that, Step S4 includes: S41. Read the number of laser beams of the shipborne lidar. and the number of emission points per harness ; Obtain 3D point clouds after complete motion compensation ; S42, Generate 3 sizes. The 2D feature matrices are the depth feature matrices. Angular feature matrix Intensity feature matrix ; S43. Traverse the 3D point cloud after complete motion compensation. Extract the depth value of each point Angle value and intensity value Then, normalization is performed, and the laser emission order of the shipborne lidar is stored in the depth feature matrix. Angular feature matrix Intensity feature matrix The corresponding position.

6. The method for feature transformation and noise reduction of shipborne lidar in a port navigation scenario according to claim 1, characterized in that, Step S5 includes: S51. Generate three index arrays, namely the near-field index array. Midfield index array and far-field index array ; S52, Traversing the Depth Feature Matrix If the current point index depth value If it is a near-field point, it is stored in the near-field index array. ;like If it is a midfield point, it is determined to be a midfield point and stored in the midfield index array. ;like If it is a far-field point, it is stored in the index array far-field. .

7. The method for feature transformation and noise reduction of shipborne lidar in port navigation scenarios according to claim 1, characterized in that, Step S6 includes: S61. Construct a labeling function for regional adaptive point cloud denoising. Near-field points are subject to a single intensity constraint, mid-field points are subject to a dual constraint of intensity and spatial abrupt change, and far-field points are subject to a dual constraint of intensity and depth spatial continuity. The noise determination rule is as follows: in, Indicates the tagging function, Indicates the intensity threshold. Indicates the mutation threshold. Indicates index Mean intensity within the neighborhood Indicates index Depth values ​​at each point within the neighborhood; S62, Traverse the near-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. If so, then mark that point as a noise point; S63, Traverse the midfield index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And the current point intensity value is related to its position in the background. The absolute value of the difference between the mean intensity values ​​within the middle neighborhood is greater than or equal to the mutation threshold. If so, then mark that point as a noise point; S64, Traverse the far-field index array If the current point is in the intensity feature matrix, then... The intensity value is less than or equal to the intensity threshold. And its depth feature matrix If the depth value of each point in the middle neighborhood is 0, then the point is marked as a noise point; S65. Filter out the point cloud data marked as noise points and output the denoised depth feature matrix, angle feature matrix and intensity feature matrix.

Images