A lidar-based method and apparatus for surface clutter processing using unmanned surface vessels.

Abstract

This invention proposes a method and apparatus for processing surface clutter using lidar based on unmanned surface vessels (USVs). The method includes: acquiring point cloud data of the USV's surface environment using lidar; performing coordinate system transformation and point cloud correction on the acquired point cloud data; segmenting the corrected data into lidar point cloud clusters using grids; classifying the segmented lidar point cloud clusters by extracting multiple features; and performing real-time frame-by-frame matching and target tracking of USV surface targets using the extracted feature vectors and the positioning information of the point cloud clusters, calculating the target's motion elements, and using the continuity of the target and the randomness of the waves to remove clutter points and accurately track the target. This method can filter out outliers and clutter points caused by plankton, and simultaneously filter out wake waves and spray.

Description

Technical Field

[0001] This invention pertains to unmanned surface vessel (USV) surface environment perception and three-dimensional lidar information processing, specifically relating to a lidar surface clutter processing method and device based on USV. Background Technology

[0002] For unmanned surface vessels (USVs) to safely and autonomously complete various tasks, they must possess excellent environmental awareness capabilities. LiDAR, as an active measurement sensor, boasts strong environmental adaptability and high ranging accuracy, making it widely used in unmanned vehicles with promising results. In recent years, LiDAR has also seen some applications in the complex navigation environments of USVs. However, when using LiDAR to scan water surface targets, clutter points often exist in the point cloud data, which can be misidentified as obstacles by environmental awareness algorithms, affecting the normal navigation of the USV. Therefore, algorithms are needed to filter out these clutter points.

[0003] Lasers can penetrate water, so when the water is calm and clean, there is virtually no effective echo. However, wakes from ships and whitecaps from wind waves can reflect the laser back to the radar receiver, creating surface clutter. During propagation, interference from clouds, smoke, rain, and snow can create outliers. Furthermore, when distant targets are severely obstructed by nearby targets, chaotic, highly sparse, and meaningless points can appear. When there is a large amount of plankton in the seawater, it can create concentric rings of surface clutter around the unmanned surface vessel. Therefore, filtering outliers and plankton-induced clutter, while simultaneously filtering wakes and whitecaps, is a pressing problem to be solved. Summary of the Invention

[0004] In view of this, the present invention provides a lidar-based surface clutter processing method for unmanned surface vessels. This method can be used to filter out outliers and clutter points caused by plankton, and simultaneously filter out wake waves and spray, including:

[0005] Step S1: Use lidar to acquire point cloud data of the unmanned surface vessel's water environment;

[0006] Step S2: Perform coordinate system transformation on the obtained point cloud data and perform point cloud correction;

[0007] Step S3: Perform grid segmentation on the data after point cloud correction to obtain lidar point cloud clusters;

[0008] Step S4: Perform multiple feature extractions on the segmented lidar point cloud clusters to complete the classification of the point cloud clusters;

[0009] Step S5: Use the extracted feature vectors and the localization information of the point cloud clusters to complete the frame-by-frame real-time matching and target tracking of the unmanned surface vessel (USV) on the water, calculate the motion elements of the target, and use the continuity of the target and the randomness of the waves to complete the removal of clutter points and accurate tracking of the target.

[0010] Specifically, step S2 includes:

[0011] Step S2.1: The coordinate system transformation involves converting the angle and distance information acquired by the 3D LiDAR in real-time scanning from the spherical coordinate system to a Cartesian coordinate system (x, y, z), where the X-axis represents the bow direction, the Y-axis represents the lateral direction of the hull, and the Z-axis represents the vertical upward direction; the spherical coordinates (d, θ) returned by the 3D LiDAR in real-time scanning of the target point... vertical ,θ horizon ), which are the distance to the return point, the vertical angle, and the horizontal offset angle, respectively. The conversion formula is as follows:

[0012]

[0013] Step S2.2: The point cloud correction is necessary because the data obtained by the 3D lidar deviates from the actual data due to the pitch and roll of the unmanned surface vessel (USV), affecting obstacle localization and clutter suppression. The USV's motion attitude (α, β, θ) can be acquired in real time using an attitude angle sensor, representing the USV's roll angle, pitch angle, and heading angle, respectively. The attitude transformation matrix for point cloud correction is:

[0014]

[0015] Specifically, step S3 includes:

[0016] Step S3.1: Divide the map into a series of grid cells with rows and columns according to the LiDAR range. The unmanned surface vessel (USV) is always at the center of the grid map. The grid resolution can be adjusted. Then, project the point cloud data onto the grid cells based on the coordinates of the point cloud data, and assign an attribute to each grid cell. The grid attributes include grid coordinates, number of return points, average height difference, average reflection intensity, and maximum height difference, defined as {(x,y),S,NumPoint,AvgHeight,AvgInten,HeightDiff}, where S indicates whether the grid cell contains obstacles. This is determined by setting a specific threshold for each grid cell. Grid cells with more than the threshold are considered obstacles, while those with fewer than the threshold are considered outliers or clutter. The threshold is set based on the spatial resolution of the LiDAR point cloud. The formula for calculating the horizontal spatial resolution of the LiDAR is as follows:

[0017]

[0018] Where N is the total amount of data acquired by the lidar per second, which is a fixed value; M is the number of lidar beams; f is the radar scanning frequency; d is the detection distance of the point; analysis shows that, under the condition of a fixed radar scanning frequency, the horizontal range resolution of the lidar is linearly proportional to the distance between the grid and the center point. The farther away from the center point, the smaller the threshold for judging whether there is an obstacle in the grid.

[0019] Step S3.2: Clutter removal caused by reflections from plankton or turbid water is achieved by using the corrected height value Z and the height information of the lidar installed on the unmanned surface vessel.

[0020] Step S3.3: Raster segmentation. The raster is segmented using the fast region labeling method of eight-neighborhood dilation. Connected rasters are grouped together and identified as a point cloud cluster. Since clutter points caused by wake waves and spray generally have higher reflection intensity, smaller height difference, and lower average height compared to real targets, in order to prevent over-segmentation and under-segmentation during the segmentation process, intensity and height thresholds are set according to the maximum height difference, average height, and average reflection intensity of the raster attributes obtained in S3.1, and targets such as ships and wake waves are segmented into different point cloud clusters.

[0021] Specifically, step 3.1, which involves projecting the 3D coordinate points onto the raster map and determining the raster attributes, includes:

[0022] Step 3.1.1: Input all 3D data points; iterate through each 3D data point and project it onto the corresponding grid cell of the raster map according to its x and y coordinates; for each grid cell, increment its count and determine whether the point is the highest or lowest point. If so, update the highest or lowest height of the grid cell; add the 3D coordinates and reflection intensity of the point to the data vector of the grid class; output the raster map.

[0023] Step 3.1.2: Then input all grids and the threshold for whether a grid is an obstacle in a two-dimensional array, and traverse each grid; if the number of points N in the grid is greater than the corresponding threshold, the grid is considered an obstacle, and the average intensity value and average height value of all points in the grid are calculated; otherwise, the points in the grid are determined to be outlier clutter, and the points in the grid are removed.

[0024] Specifically, step S4 includes:

[0025] Step S4.1: Feature extraction, which classifies point clouds based on their geometric features and physical characteristics of distribution. The physical features include the average height, maximum height difference, and average reflection intensity of the point cloud clusters calculated using the raster attributes from S3.1. The geometric features are calculated using the eigenvectors and eigenvalues ​​of the point cloud cluster covariance matrix, such as linear exponent, areal exponent, divergent exponent, isotropic, anisotropic, eigenentral entropy, eigenvalue sum, and curvature variation.

[0026] Step S4.2: Using the classifier with multiple decision trees in the random forest algorithm, the output category is determined by the mode of the categories output by individual trees. First, a large amount of point cloud data was collected at sea, including waves, ship wakes, fishing boats, cruise ships, and buoys. Then, the type of each target was manually labeled, and feature samples were built using the features proposed in step S4.1 for training to construct a random forest classification model. When the unmanned surface vessel is navigating, the trained classifier is used in real time to predict the type of point cloud clusters segmented in step S3, and clutter of type wakes and waves is removed.

[0027] This invention also proposes a lidar surface clutter processing device based on an unmanned surface vessel, comprising:

[0028] The point cloud data acquisition module is used to acquire point cloud data of the unmanned surface vessel's water environment using lidar;

[0029] The preprocessing module is used to perform coordinate system transformation and point cloud correction on the obtained point cloud data;

[0030] The grid segmentation module is used to perform grid segmentation on the data after point cloud correction to obtain lidar point cloud clusters;

[0031] The point cloud cluster classification module is used to perform multiple feature extractions on the segmented lidar point cloud clusters to complete the classification of the point cloud clusters;

[0032] The clutter removal and target precision tracking module is used to perform frame-by-frame real-time matching and target tracking of unmanned surface vessels (USVs) using the extracted feature vectors and the positioning information of point cloud clusters. It calculates the motion elements of the target and uses the continuity of the target and the randomness of the waves to remove clutter and perform precise target tracking.

[0033] Specifically, the preprocessing module includes:

[0034] The coordinate system transformation module converts the angle and distance information acquired by the 3D LiDAR in real-time scanning from the spherical coordinate system to a Cartesian coordinate system (x, y, z), where the X-axis is the bow direction, the Y-axis is the lateral direction of the hull, and the Z-axis is the vertical upward direction; the spherical coordinates (d, θ) returned by the 3D LiDAR in real-time scanning of the target point are also converted. vertical ,θhorizon ), which are the distance to the return point, the vertical angle, and the horizontal offset angle, respectively. The conversion formula is as follows:

[0035]

[0036] The point cloud correction module is used to correct discrepancies between the data obtained by the 3D LiDAR and the actual data due to the pitch and roll of the unmanned surface vessel (USV), which affects obstacle localization and clutter suppression. It utilizes attitude angle sensors to acquire the USV's motion attitude (α, β, θ) in real time, representing the USV's roll angle, pitch angle, and yaw angle, respectively. The attitude transformation matrix for point cloud correction is:

[0037]

[0038] Specifically, the grid segmentation module includes:

[0039] The outlier clutter removal module divides the map into a series of grid cells based on the LiDAR range, with the unmanned surface vessel (USV) always at the center of the grid map. The grid resolution is adjustable. Then, it projects the point cloud data onto the grid cells based on the coordinates of the point cloud data, assigning each grid cell an attribute. These attributes include grid coordinates, number of return points, average height difference, average reflection intensity, and maximum height difference, defined as {(x,y),S,NumPoint,AvgHeight,AvgInten,HeightDiff}, where S indicates whether the grid cell contains obstacles. This is determined by setting a specific threshold for each grid cell. Grid cells with more points than the threshold are considered obstacles, while those with fewer points are considered outlier clutter. The threshold is set based on the LiDAR's point cloud spatial resolution. The formula for calculating the LiDAR's horizontal spatial resolution is as follows:

[0040]

[0041] Where N is the total amount of data acquired by the lidar per second, which is a fixed value; M is the number of lidar beams; f is the radar scanning frequency; d is the detection distance of the point; analysis shows that, under the condition of a fixed radar scanning frequency, the horizontal range resolution of the lidar is linearly proportional to the distance between the grid and the center point. The farther away from the center point, the smaller the threshold for judging whether there is an obstacle in the grid.

[0042] The surface clutter removal module is used to remove clutter generated by surface plankton or water turbidity. It uses the corrected height value Z and the height information of the lidar installed on the unmanned surface vessel for removal.

[0043] The point cloud cluster segmentation module is used to segment the grid using the fast region labeling method of eight-neighbor expansion. Connected grids are grouped together and identified as a point cloud cluster. Since clutter points caused by wakes and sprays generally have higher reflection intensity, smaller height difference, and lower average height compared to real targets, in order to prevent over-segmentation and under-segmentation during the segmentation process, intensity and height thresholds are set according to the maximum height difference, average height, and average reflection intensity of the grid attributes obtained by the outlier clutter removal module, so that targets such as ships and wakes are segmented into different point cloud clusters.

[0044] Specifically, the determination of the projection of 3D coordinate points onto the raster map and the determination of raster attributes in the outlier clutter removal module includes:

[0045] The raster map output module is used to input all 3D data points; it iterates through each 3D data point and projects it onto the corresponding raster of the raster map according to its x and y coordinates; for each raster, it increments its count by one and determines whether the point is the highest or lowest point. If so, it updates the highest or lowest height of the raster; it adds the 3D coordinates and reflection intensity of the point to the data vector of the raster class; and outputs the raster map.

[0046] The grid traversal module is used to traverse each grid after inputting a two-dimensional array threshold for all grids and whether a grid is an obstacle. If the number of points N in a grid is greater than the corresponding threshold, the grid is considered an obstacle, and the average intensity value and average height value of all points in the grid are calculated. Otherwise, the points in the grid are determined to be outlier clutter, and the points in the grid are removed.

[0047] Specifically, the clutter removal and target precision tracking module includes:

[0048] The feature extraction module is used for point cloud classification based on the geometric features and physical features of the point cloud distribution. The physical features include the average height, maximum height difference, and average reflection intensity of the point cloud clusters calculated from the raster attributes in the outlier clutter removal module. The geometric features are calculated using the eigenvectors and eigenvalues ​​of the point cloud cluster covariance matrix, such as linear exponent, areal exponent, divergent exponent, isotropic, anisotropic, feature entropy, eigenvalue sum, and curvature variation.

[0049] The Random Forest algorithm classification module utilizes a classifier with multiple decision trees within the Random Forest algorithm, determining the output category by the mode of the categories output by individual trees. First, a large amount of point cloud data was collected at sea, including waves, ship wakes, fishing boats, cruise ships, and buoys. Then, the type of each target was manually labeled, and feature samples were built using features extracted by the feature extraction module for training, constructing a Random Forest classification model. During unmanned surface vessel (USV) navigation, the trained classifier is used in real-time to predict the type of point cloud clusters segmented by the grid segmentation module, removing clutter of wake and wave types.

[0050] Beneficial effects:

[0051] 1) This invention can solve the problem of outliers caused by interference such as clouds, smoke, rain and snow during the propagation of laser beams;

[0052] 2) Through this invention, clutter points caused by plankton can be filtered out using height threshold screening and horizontal plane fitting methods;

[0053] 3) This invention can effectively filter out wake waves and spray, adapting to high sea states and complex conditions, with high accuracy and good robustness; Attached Figure Description

[0054] Figure 1 A flowchart of a lidar surface clutter processing method based on an unmanned surface vessel provided by the present invention;

[0055] Figure 2 This is a schematic diagram of the three-dimensional lidar coordinate system used in an embodiment of the present invention;

[0056] Figures 3a-3b This is a schematic diagram of a three-dimensional lidar grid representation according to an embodiment of the present invention;

[0057] Figures 4a-4c The images show point cloud data of ship wakes and clutter data collected from turbid water surfaces, as well as data collected by a three-dimensional lidar system, in this embodiment of the invention. Detailed Implementation

[0058] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0059] This invention provides a lidar-based method for processing surface clutter on unmanned surface vessels, such as... Figure 1 This includes the following steps:

[0060] Step S1. Use lidar to acquire point cloud data of the unmanned surface vessel's water environment;

[0061] In this example, the RS-Helios-1615 lidar from RoboSense is used. The lidar is horizontally mounted on the top of the unmanned surface vessel (USV) in an unobstructed position, and rotates at a constant speed of 10Hz to scan the point cloud data of the water surface environment within a 360° field of view around the USV. First, the point cloud data is parsed according to the lidar's communication protocol to obtain the horizontal angle, distance information, and reflection intensity of each reflection point in spherical coordinates. Then, the vertical angle of the reflection point is determined based on the laser emission angle. Reflection points outside the lidar's measurement range are removed. A schematic diagram of the lidar's spherical coordinate system is shown below. Figure 2 As shown.

[0062] Step S2. Preprocess the obtained point cloud data, including coordinate system transformation and point cloud correction;

[0063] Step S2.1 Coordinate transformation: Transform the coordinates (d, θ) in the spherical coordinate system obtained in step S1. vertical ,θ horizon To convert (x, y, z) to Cartesian coordinates, the conversion formula is as follows:

[0064]

[0065] Step S2.2 Point Cloud Correction: Install attitude angle sensors and BeiDou navigation on the unmanned surface vessel (USV). Calibrate the zero point of the lidar with the zero point of the attitude angle sensors. Use the output real-time USV motion attitude (α, β, θ), representing the roll, pitch, and yaw angles of the USV, to correct the three-dimensional coordinates obtained in Step S2.1 (time alignment). The attitude transformation matrix for point cloud correction is:

[0066]

[0067] After the data is corrected, we will obtain three-dimensional point cloud data in the geodetic coordinate system, which solves the problem of inaccurate target positioning by lidar measurement caused by pitch and roll during unmanned surface vessel navigation.

[0068] Step S3. Perform raster segmentation on the point cloud data to obtain lidar point cloud clusters;

[0069] Step S3.1 The RS-Helios-1615 LiDAR has 70,000 3D data points per frame. To improve the real-time performance of data processing and subsequent unmanned surface vessel (USV) path planning, this example first establishes a grid map with a size of 400*400. The USV is always centered on the grid map. The grid resolution is 1m and can be adjusted as needed. See the schematic diagram below. Figures 3a-3bAs shown, the 3D points in the current frame are then projected onto a grid map. Each grid contains several points, which retain height and intensity information. Finally, the grid's attributes are calculated based on these points. The grid attributes include grid coordinates, number of returned points, average height difference, average reflection intensity, and maximum height difference, defined as {(x,y),S,NumPoint,AvgHeight,AvgInten,HeightDiff}, where S indicates whether the grid contains obstacles. This is determined by setting a specific threshold for each grid. Grids with more points than the threshold are considered obstacles, while those with fewer points are considered outliers or clutter. The threshold is set using the spatial resolution of the LiDAR point cloud, represented by a two-dimensional array whose size is the same as the grid size. The formula for calculating the horizontal spatial resolution of the LiDAR is as follows:

[0070]

[0071] Where N is the total amount of data acquired by the lidar per second, a fixed value of 700,000; M is the number of lidar beams, 32 lines; f is the radar scanning frequency, 10Hz; d is the detection distance of the point, calculated based on the grid coordinates (x,y); analysis shows that, with a fixed radar scanning frequency, the horizontal range resolution of the lidar is linearly proportional to the distance between the grid and the center point. The farther away from the center point, the smaller the threshold for judging whether there is an obstacle in the grid.

[0072] The specific process for projecting 3D coordinate points into a raster map and determining raster attributes is as follows:

[0073]

[0074]

[0075] Step S3.2: The raster map obtained in the previous step contains 160,000 graticules. For subsequent target feature extraction and classification, target segmentation and clustering are required. This example uses the eight-neighbor dilation fast region labeling method to segment the graticules, grouping connected graticules into a single point cloud cluster. Since clutter points caused by wake waves and spray generally have higher reflection intensity, smaller height differences, and lower average height compared to real targets, to prevent over-segmentation and under-segmentation during the segmentation process, intensity and height thresholds are set based on the maximum height difference, average height, and average reflection intensity of the graticule attributes obtained in step S3.1. This separates targets such as ships from wake waves into different point cloud clusters. The specific segmentation process is as follows:

[0076]

[0077]

[0078] Step S4. Perform multiple feature extraction on the segmented point cloud clusters to complete the classification of the point cloud clusters;

[0079] Step S4.1 Feature extraction: First, construct an n*3 matrix MatrixXf using the 3D coordinates of all points in the point cloud cluster. Then, calculate the three eigenvalues ​​of the matrix covariance using EigenSolver from the Eigen matrix library, and construct the feature vector based on the geometric features of the point cloud cluster according to the table below.

[0080]

[0081] Step S4.2 Random Forest Algorithm Classification: Random Forest is a classifier containing multiple decision trees, and its output class is determined by the mode of the classes output by individual trees. In this example, a large amount of point cloud data was first collected at sea near Sanya, including waves, ship wakes, fishing boats, cruise ships, buoys, etc. Then, the type of each target was manually labeled using the point cloud data labeling tool CloudCompare, and feature samples were established using the features proposed in step S4.1. Then, the random forest cv::ml::RTrees provided by OpenCV was directly used for model training and validation. When the unmanned surface vessel was navigating, the trained classifier was used in real time to predict the type of point cloud clusters segmented in step S3, and clutter of type wake and waves was removed. In this embodiment, the point cloud data map of ship wakes collected by 3D LiDAR and the clutter map collected from turbid water surface are shown in the figure. Figures 4a-4c As shown.

[0082] Step S5. Using the feature vectors extracted in step S4.1 and the localization information of the point cloud clusters, the unmanned surface vessel (USV) performs frame-by-frame real-time matching of water targets. Kalman filtering is used to track the target, and the motion elements of the target, including the target's heading and speed, are calculated. The continuity of the target and the randomness of the waves are used to remove clutter and accurately track the target.

[0083] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0084] It will be apparent to those skilled in the art that the embodiments of the present invention are not limited to the details of the exemplary embodiments described above, and that the embodiments of the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the embodiments of the present invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the embodiments of the present invention is defined by the appended claims rather than the foregoing description. Therefore, all variations falling within the meaning and scope of equivalents of the claims are intended to be encompassed within the embodiments of the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units, modules, or devices recited in the system, apparatus, or terminal claims may also be implemented by the same unit, module, or device through software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention and are not intended to limit them. Although the embodiments of the present invention have been described in detail with reference to the above preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the embodiments of the present invention should not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for processing surface clutter using lidar based on unmanned surface vessels, characterized in that, include: Step S1: Use lidar to acquire point cloud data of the unmanned surface vessel's water environment; Step S2: Perform coordinate system transformation on the obtained point cloud data and perform point cloud correction; Step S3: Perform grid segmentation on the data after point cloud correction to obtain lidar point cloud clusters; Step S3 specifically includes: Step S3.1: Divide the map into a series of grid cells with rows and columns according to the lidar range. The unmanned surface vessel (USV) remains at the center of the grid map. The grid resolution is adjustable. Then, project the point cloud data onto the grid cells based on the coordinates of the point cloud data, and assign each grid cell an attribute. The grid attributes include grid coordinates, number of return points, average height difference, average reflection intensity, and maximum height difference, defined as follows: Where S represents whether the grid contains obstacles, which is determined by setting a specific threshold for each grid. Grids with more points than the threshold are considered obstacles, while those with fewer points are considered outliers or clutter. The threshold is set based on the point cloud spatial resolution of the LiDAR. The formula for calculating the horizontal spatial resolution of the LiDAR is as follows: Where N is the total amount of data acquired by the lidar per second, which is a fixed value; M is the number of lidar beams; f is the radar scanning frequency; d is the detection distance of the point; analysis shows that, under the condition of a fixed radar scanning frequency, the horizontal range resolution of the lidar is linearly proportional to the distance between the grid and the center point. The farther away from the center point, the smaller the threshold for judging whether there is an obstacle in the grid. Step S3.2: Clutter removal caused by reflections from plankton or turbid water is achieved by using the corrected height value Z and the height information of the lidar installed on the unmanned surface vessel. Step S3.3: Grid segmentation. The grid is segmented using the fast region labeling method of eight-neighbor dilation. Connected grids are grouped together and identified as a point cloud cluster. Since clutter points caused by wake waves and spray generally have higher reflection intensity, smaller height difference, and lower average height compared to real targets, in order to prevent over-segmentation and under-segmentation during the segmentation process, intensity and height thresholds are set according to the maximum height difference, average height, and average reflection intensity of the grid attributes obtained in S3.1, and the ship and wake waves are segmented into different point cloud clusters. Step S4: Perform multiple feature extractions on the segmented lidar point cloud clusters to complete the classification of the point cloud clusters; Step S5: Use the extracted feature vectors and the localization information of the point cloud clusters to complete the frame-by-frame real-time matching and target tracking of the unmanned surface vessel (USV) on the water, calculate the motion elements of the target, and use the continuity of the target and the randomness of the waves to complete the removal of clutter points and accurate tracking of the target.

2. The lidar surface clutter processing method as described in claim 1, characterized in that, Step S2 specifically includes: Step S2.1: The coordinate system transformation involves converting the angle and distance information acquired by the 3D LiDAR in real-time scanning from the spherical coordinate system to a Cartesian coordinate system (x, y, z), where the x-axis represents the bow direction, the y-axis represents the lateral direction of the hull, and the z-axis represents the vertical upward direction; the spherical coordinates returned by the 3D LiDAR in real-time scanning of the target point... These are the distance to the return point, the vertical angle, and the horizontal offset angle, respectively. The conversion formula is as follows: Step S2.2: The point cloud correction is necessary because the data obtained by the 3D lidar deviates from the actual data as the unmanned surface vessel pitches and rolls, affecting obstacle localization and clutter suppression; the attitude angle sensor is used to acquire the unmanned surface vessel's motion attitude in real time. , respectively, represent the roll angle, pitch angle, and heading angle of the unmanned surface vessel. The attitude transformation matrix for point cloud correction is: 。 3. The lidar surface clutter processing method as described in claim 2, characterized in that, The determination of the projection of the 3D coordinate points into the raster map and the raster attributes in step 3.1 includes: Step 3.1.1: Input all 3D data points; iterate through each 3D data point and project it onto the corresponding grid cell of the raster map according to its x and y coordinates; for each grid cell, increment its count and determine whether the point is the highest or lowest point. If so, update the highest or lowest height of the grid cell; add the 3D coordinates and reflection intensity of the point to the data vector of the grid class; output the raster map. Step 3.1.2: Then input all grids and the threshold for whether a grid is an obstacle in a two-dimensional array, and traverse each grid; if the number of points N in the grid is greater than the corresponding threshold, the grid is considered an obstacle, and the average intensity value and average height value of all points in the grid are calculated; otherwise, the points in the grid are determined to be outlier clutter, and the points in the grid are removed.

4. The lidar surface clutter processing method as described in claim 2 or 3, characterized in that, Step S4 specifically includes: Step S4.1: Feature extraction, which classifies point clouds based on the geometric features and physical features of the point cloud distribution; the physical features include the average height, maximum height difference, and average reflection intensity of the point cloud clusters calculated using the grid attributes in S3.1; the geometric features are calculated using the eigenvectors and eigenvalues ​​of the point cloud cluster covariance matrix. Step S4.2: Using the classifier with multiple decision trees in the random forest algorithm, the output category is determined by the mode of the categories output by individual trees. First, a large amount of point cloud data was collected at sea, including waves, ship wakes, fishing boats, cruise ships, and buoys. Then, the type of each target was manually labeled, and feature samples were built using the features proposed in step S4.1 for training to construct a random forest classification model. When the unmanned surface vessel is navigating, the trained classifier is used in real time to predict the type of point cloud clusters segmented in step S3, and clutter of type wakes and waves is removed.

5. A lidar surface clutter processing device based on an unmanned surface vessel, characterized in that, include: The point cloud data acquisition module is used to acquire point cloud data of the unmanned surface vessel's water environment using lidar; The preprocessing module is used to perform coordinate system transformation and point cloud correction on the obtained point cloud data; A grid segmentation module is used to perform grid segmentation on the data after point cloud correction to obtain lidar point cloud clusters; the grid segmentation module specifically includes: The outlier clutter removal module divides the map into a series of grid cells (rows and columns) based on the lidar range, with the unmanned surface vessel (USV) always at the center of the grid map. The grid resolution is adjustable. Then, it projects the point cloud data onto the grid cells based on the coordinates of the point cloud data, assigning each grid cell an attribute. These grid attributes include grid coordinates, number of return points, average height difference, average reflection intensity, and maximum height difference, defined as follows: Where S represents whether the grid contains obstacles, which is determined by setting a specific threshold for each grid. Grids with more points than the threshold are considered obstacles, while those with fewer points are considered outliers or clutter. The threshold is set based on the point cloud spatial resolution of the LiDAR. The formula for calculating the horizontal spatial resolution of the LiDAR is as follows: Where N is the total amount of data acquired by the lidar per second, which is a fixed value; M is the number of lidar beams; f is the radar scanning frequency; d is the detection distance of the point; analysis shows that, under the condition of a fixed radar scanning frequency, the horizontal range resolution of the lidar is linearly proportional to the distance between the grid and the center point. The farther away from the center point, the smaller the threshold for judging whether there is an obstacle in the grid. The surface clutter removal module is used to remove clutter generated by surface plankton or water turbidity. It uses the corrected height value Z and the height information of the lidar installed on the unmanned surface vessel for removal. The point cloud cluster segmentation module is used to segment the grid using the fast region labeling method of eight-neighbor expansion. Connected grids are grouped together and identified as a point cloud cluster. Since clutter points caused by wake waves and spray generally have higher reflection intensity, smaller height difference, and lower average height compared to real targets, in order to prevent over-segmentation and under-segmentation during the segmentation process, intensity and height thresholds are set according to the maximum height difference, average height, and average reflection intensity of the grid attributes obtained by the outlier clutter removal module during the segmentation process, so as to segment the ship and wake waves into different point cloud clusters. The point cloud cluster classification module is used to perform multiple feature extractions on the segmented lidar point cloud clusters to complete the classification of the point cloud clusters; The clutter removal and target precision tracking module is used to perform frame-by-frame real-time matching and target tracking of unmanned surface vessels (USVs) using the extracted feature vectors and the positioning information of point cloud clusters. It calculates the motion elements of the target and uses the continuity of the target and the randomness of the waves to remove clutter and perform precise target tracking.

6. The lidar surface clutter processing device based on an unmanned surface vessel as described in claim 5, characterized in that, The preprocessing module specifically includes: The coordinate system transformation module converts the angle and distance information acquired by the 3D LiDAR in real-time scanning from the spherical coordinate system to a Cartesian coordinate system (x, y, z), where the x-axis represents the bow direction, the y-axis represents the lateral direction of the hull, and the z-axis represents the vertical upward direction; the spherical coordinates returned by the 3D LiDAR in real-time scanning of the target point... These are the distance to the return point, the vertical angle, and the horizontal offset angle, respectively. The conversion formula is as follows: The point cloud correction module is used to correct discrepancies between the data obtained by the 3D LiDAR and the actual data due to the pitch and roll of the unmanned surface vessel (USV), which affects obstacle localization and clutter suppression. It utilizes an attitude angle sensor to acquire the USV's motion attitude in real time. , respectively, represent the roll angle, pitch angle, and heading angle of the unmanned surface vessel. The attitude transformation matrix for point cloud correction is: 。 7. The lidar surface clutter processing device based on an unmanned surface vessel as described in claim 6, characterized in that, The outlier clutter removal module includes the following: the projection of 3D coordinate points onto the raster map and the determination of raster attributes. The raster map output module is used to input all 3D data points; it iterates through each 3D data point and projects it onto the corresponding raster of the raster map according to its x and y coordinates; for each raster, it increments its count by one and determines whether the point is the highest or lowest point. If so, it updates the highest or lowest height of the raster; it adds the 3D coordinates and reflection intensity of the point to the data vector of the raster class; and outputs the raster map. The grid traversal module is used to traverse each grid after inputting a two-dimensional array threshold for all grids and whether a grid is an obstacle. If the number of points N in a grid is greater than the corresponding threshold, the grid is considered an obstacle, and the average intensity value and average height value of all points in the grid are calculated. Otherwise, the points in the grid are determined to be outlier clutter, and the points in the grid are removed.

8. The lidar surface clutter processing device based on an unmanned surface vessel as described in claim 6 or 7, characterized in that, The clutter removal and precise target tracking module specifically includes: The feature extraction module is used for point cloud classification based on the geometric features of the point cloud and the physical features of the point cloud distribution. The physical features include the average height, maximum height difference, and average reflection intensity of the point cloud clusters calculated from the grid attributes in the outlier clutter removal module. The geometric features are calculated using the eigenvectors and eigenvalues ​​of the point cloud cluster covariance matrix. The Random Forest algorithm classification module utilizes a classifier with multiple decision trees within the Random Forest algorithm, determining the output category by the mode of the categories output by individual trees. First, a large amount of point cloud data was collected at sea, including waves, ship wakes, fishing boats, cruise ships, and buoys. Then, the type of each target was manually labeled, and feature samples were built using features proposed by the feature extraction module for training, constructing a Random Forest classification model. During unmanned surface vessel (USV) navigation, the trained classifier is used in real-time to predict the type of point cloud clusters segmented by the grid segmentation module, removing clutter of wake and wave types.

Images