Three-dimensional point cloud extraction method and device

Abstract

The application discloses a three-dimensional point cloud extraction method and device, wherein the method comprises the following steps: obtaining the circumferential scanning GBSAR three-dimensional data of a three-dimensional SAR image, extracting the three-dimensional maximum value projection in the three-dimensional SAR image to obtain a three-view image; detecting the three-view image to obtain a three-view mask; projecting the three-view mask to obtain a three-dimensional mask; taking the intersection of the three-dimensional mask and the three-dimensional SAR image to obtain potential target area data; detecting the potential target area data to obtain a three-dimensional target point cloud; suppressing false alarms in the three-dimensional target point cloud to obtain clustering results; distinguishing the target and strong sidelobes in the clustering results according to preset characteristics; removing the class in which the strong sidelobes are located in the clustering results to obtain the three-dimensional point cloud after false alarm suppression. The application can accurately extract the three-dimensional point cloud under the influence of the strong sidelobes and suppress false alarms.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for extracting three-dimensional point clouds. Background Technology

[0002] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section.

[0003] Ground-based Synthetic Aperture Radar (GBSAR) is a new type of GBSAR that introduces the airborne circular track synthetic aperture radar (SAR) mode to a ground-based platform. It achieves three-dimensional imaging of the monitoring field of view by forming a two-dimensional circular synthetic aperture through the rotation of the antenna in the vertical plane. It can accurately extract three-dimensional information in scenes with undulating terrain and tall buildings, and has attracted widespread attention in steep terrain mapping, urban mapping, and target detection and identification.

[0004] However, the curved observation geometry of circular scanning GBSAR leads to the problem of strong sidelobes in the images. In 3D SAR data, strong sidelobes have a three-dimensional spatial distribution, spreading throughout the entire data. Therefore, at different altitudes, strong sidelobes may mask the main lobe of nearby weak targets, resulting in the loss of target points and the inability to accurately extract target point clouds from 3D SAR images due to the detection of strong sidelobes.

[0005] Currently, there is no effective method for extracting target point clouds from 3D SAR images. At present, in terms of target detection, 2D SAR target detection is mature, and the two-parameter constant false alarm rate (CFAR) target detector has adaptive threshold detection capability, making it an important algorithm that can adapt to changes in background clutter. However, when applied to 3D SAR target detection, this algorithm needs to detect layer by layer at different heights in the 3D image. Due to the influence of the strong sidelobe spatial distribution in 3D, the layer-by-layer detection results of the two-parameter CFAR contain a large number of false alarms, making it impossible to accurately extract the target point cloud from the 3D SAR image. Regarding false alarm suppression, density-based spatial clustering of applications with noise (DBSCAN) is a commonly used density-based clustering algorithm. This algorithm can quickly extract clustering information from noisy image data by adjusting two parameters: the neighborhood threshold Eps and the point count threshold MinPts, thus achieving false alarm suppression.

[0006] However, the traditional DBSCAN algorithm has drawbacks when processing 3D SAR data with uneven point cluster density and strong sidelobes. Firstly, the selection of the parameters Eps and MinPts is sensitive to changes in point density within a cluster. If the Eps parameter is too large, it cannot distinguish strong sidelobe clusters diffused throughout the 3D point cloud; if the Eps parameter is too small, some free-floating target point clusters will be missed, making it difficult to select a suitable threshold. Secondly, the 3D target point cloud detected by the CFAR algorithm still contains some strong sidelobes diffused throughout the data. The traditional DBSCAN clustering method cannot accurately cluster to distinguish targets from strong sidelobes, thus failing to achieve false alarm suppression and accurate extraction of the 3D point cloud.

[0007] In summary, current technologies cannot achieve accurate extraction of 3D point clouds. Regarding target detection, the curved observation geometry of circular scanning GBSAR results in strong sidelobes in the image, increasing the difficulty of 3D SAR image target point cloud extraction methods. The classic two-parameter CFAR algorithm cannot detect accurate target points under the influence of strong sidelobes. In terms of false alarm suppression, the traditional DBSCAN algorithm is sensitive to changes in intra-cluster point density, failing to effectively identify and segment target regions and strong sidelobes, thus failing to suppress false alarms. Summary of the Invention

[0008] This invention provides a method for extracting three-dimensional point clouds, which accurately extracts three-dimensional point clouds under the influence of strong sidelobes and suppresses false alarms. The method includes:

[0009] Obtain circular scan GBSAR 3D data of 3D SAR image, extract the maximum projection of the three dimensions in the 3D SAR image to obtain three views;

[0010] The three views are detected to obtain a three-view mask;

[0011] A three-dimensional mask is obtained by projecting the three-view mask, and the intersection of the three-dimensional mask and the three-dimensional SAR image is obtained to obtain the potential target area data.

[0012] The potential target region data is detected to obtain a three-dimensional target point cloud;

[0013] False alarms are suppressed in the three-dimensional target point cloud to obtain clustering results. The target and strong sidelobes in the clustering results are distinguished according to preset features. The class of strong sidelobes in the clustering results is removed to obtain the three-dimensional point cloud after false alarm suppression.

[0014] This invention also provides a three-dimensional point cloud extraction device for accurately extracting three-dimensional point clouds under the influence of strong sidelobes and suppressing false alarms. The device includes:

[0015] The projection extraction module is used to obtain the circular scan GBSAR 3D data of the 3D SAR image. It extracts the projection of the maximum value of the three dimensions in the 3D SAR image to obtain the three-view drawing.

[0016] The detection module is used to detect the three views and obtain a three-view mask;

[0017] The data acquisition module is used to obtain a three-dimensional mask by projecting a three-view mask, and to obtain potential target area data by intersecting the three-dimensional mask with the three-dimensional SAR image.

[0018] The 3D point cloud acquisition module is used to detect the potential target area data and obtain a 3D target point cloud;

[0019] The false alarm suppression module is used to suppress false alarms in the three-dimensional target point cloud, obtain clustering results, distinguish the target and strong sidelobes in the clustering results according to preset features, remove the class of strong sidelobes in the clustering results, and obtain the three-dimensional point cloud after false alarm suppression.

[0020] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described three-dimensional point cloud extraction method.

[0021] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described three-dimensional point cloud extraction method.

[0022] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described three-dimensional point cloud extraction method.

[0023] In this embodiment of the invention, the three-dimensional point cloud extraction method differs from existing technologies in several ways. Firstly, regarding target detection, the curved observation geometry of circular GBSAR results in strong sidelobes in the image, increasing the difficulty of extracting target point clouds from three-dimensional SAR images. The classic two-parameter CFAR algorithm cannot accurately detect target points under the influence of strong sidelobes. Secondly, regarding false alarm suppression, the traditional DBSCAN algorithm is sensitive to changes in point density within clusters, failing to effectively identify and segment target areas and strong sidelobes, thus failing to suppress false alarms. In contrast, this invention obtains three-dimensional data from circular GBSAR circular scan images of three-dimensional SAR images, extracts the projection of the maximum values ​​in the three dimensions of the three-dimensional SAR image to obtain three-view images, detects these three-view images to obtain a three-view mask, projects the three-view mask to obtain a three-dimensional mask, intersects the three-dimensional mask with the three-dimensional SAR image to obtain potential target area data, detects the potential target area data to obtain a three-dimensional target point cloud, suppresses false alarms in the three-dimensional target point cloud, obtains clustering results, distinguishes targets and strong sidelobes in the clustering results based on preset features, removes the cluster containing strong sidelobes, and obtains a three-dimensional point cloud with suppressed false alarms. This allows for accurate extraction of 3D point clouds under the influence of strong sidelobes, thus suppressing false alarms. Attached Figure Description

[0024] 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0025] Figure 1 This is a flowchart illustrating the three-dimensional point cloud extraction method in an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of the process of obtaining a three-dimensional target point cloud based on the global constant false alarm rate algorithm in an embodiment of the present invention;

[0027] Figure 3 This is a flowchart illustrating the improved density-based noise-based spatial clustering algorithm in an embodiment of the present invention.

[0028] Figure 4 This is a flowchart illustrating a three-dimensional point cloud extraction method in another embodiment of the present invention;

[0029] Figure 5 This is a schematic diagram of the structure of the three-dimensional point cloud extraction device in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0031] The challenge of extracting 3D point clouds from circularly scanned GBSAR based on two-step CFAR lies in the strong sidelobes present in the images due to the curved observation geometry of circularly scanned GBSAR. Furthermore, in 3D SAR data, these strong sidelobes have a three-dimensional spatial distribution, spreading throughout the entire 3D SAR data, increasing the difficulty of target point cloud extraction methods. Existing technologies have drawbacks when applied to 3D data with strong sidelobes. In target detection, the classic two-parameter CFAR algorithm cannot accurately detect target points under the influence of strong sidelobes. Regarding false alarm suppression, the traditional DBSCAN algorithm is sensitive to changes in intra-cluster point density, failing to effectively identify and segment target regions and strong sidelobes, thus failing to achieve false alarm suppression.

[0032] Current technologies cannot achieve accurate extraction of 3D point clouds. In target detection, the curved observation geometry of circular scanning GBSAR results in strong sidelobes in the image, increasing the difficulty of 3D SAR image target point cloud extraction methods. The classic two-parameter CFAR algorithm cannot detect accurate target points under the influence of strong sidelobes. Regarding false alarm suppression, the traditional DBSCAN algorithm is sensitive to changes in intra-cluster point density, failing to effectively identify and segment target regions and strong sidelobes, thus failing to suppress false alarms.

[0033] To address the aforementioned problems, this invention provides a method for extracting three-dimensional point clouds, accurately extracting three-dimensional point clouds even under the influence of strong sidelobes. This solves the problem that the curved observation geometry of circular scan GBSAR leads to strong sidelobes in three-dimensional SAR images, making accurate extraction of three-dimensional point clouds difficult. This method utilizes a two-step CFAR algorithm to detect target points in three-dimensional SAR images under the influence of strong sidelobes, and further suppresses false alarms using an improved DBSCAN algorithm, achieving accurate extraction of target point clouds from three-dimensional SAR images.

[0034] Figure 1 This is a flowchart illustrating the three-dimensional point cloud extraction method in an embodiment of the present invention. As shown in the figure, the method includes:

[0035] Step 101: Obtain the circular scan GBSAR 3D data of the 3D SAR image, extract the projection of the maximum value of the three dimensions in the 3D SAR image, and obtain the three views;

[0036] Step 102: Detect the three views to obtain a three-view mask;

[0037] Step 103: Obtain a 3D mask by projecting the three-view mask, and take the intersection of the 3D mask and the 3D SAR image to obtain the potential target area data;

[0038] Step 104: Detect the potential target region data to obtain a three-dimensional target point cloud;

[0039] Step 105: Suppress false alarms in the three-dimensional target point cloud to obtain clustering results. Distinguish between targets and strong sidelobes in the clustering results according to preset features. Remove the class containing the strong sidelobes in the clustering results to obtain the three-dimensional point cloud after false alarm suppression.

[0040] In one embodiment, the formula for the three views in step 101 can be expressed as:

[0041]

[0042] Where S represents the 3D data obtained from circular scanning GBSAR, S 左 This represents the left view of the three-view diagram, where x0 is the direction vector of the X-axis, and S... 主 This represents the front view of the three-view diagram, where y0 is the direction vector of the Y-axis, and S... 俯 This represents the top view of the three-view diagram. z0 is the direction vector of the Z-axis, and max represents the maximum value of the direction vector corresponding to the axis direction of the circular scan GBSAR 3D data.

[0043] In one embodiment, step 102, detecting the three views to obtain a three-view mask, may include:

[0044] The three-view image is detected using a two-parameter constant false alarm rate algorithm to obtain a three-view image mask.

[0045] In practice, the three-view mask is obtained according to the following formula:

[0046]

[0047] Among them, M 左 (y,z) represents the left view mask of the three-view mask, M 主 (x,z) represents the front view mask of the three-view mask, M 俯 (x,y) represents the top view mask of the three-view mask, and CFAR represents the two-parameter constant false alarm rate algorithm.

[0048] In one embodiment, step 104, detecting the potential target region data to obtain a three-dimensional target point cloud, may include:

[0049] The potential target region data is detected using the global constant false alarm rate algorithm to obtain a three-dimensional target point cloud.

[0050] In practice, Figure 2This is a schematic diagram illustrating the process of obtaining a 3D target point cloud based on the global constant false alarm rate algorithm in an embodiment of the present invention, as shown below. Figure 2 As shown, based on the global constant false alarm rate algorithm, the potential target region data is detected to obtain a three-dimensional target point cloud, which may include:

[0051] Step 201: Based on the global constant false alarm rate algorithm, perform global constant false alarm rate detection on the potential target area data, transform the potential target area data into the first row vector, and calculate the mean and variance of the data in the first row vector;

[0052] Step 202: Calculate the detection threshold based on the mean and variance and the constant false alarm probability;

[0053] Step 203: Traverse the potential target region data in the form of a sliding window to obtain the center pixel of the sliding window;

[0054] Step 204: Determine whether the center pixel of the sliding window is a target point based on the detection threshold, and obtain the target point detection result;

[0055] Step 205: Obtain the three-dimensional target point cloud based on the target point detection results.

[0056] In one embodiment, the formula for determining whether the center pixel of the sliding window is a target point based on the detection threshold in step 204 can be as follows:

[0057]

[0058] Among them, X i I represents the center pixel of the sliding window. c This indicates the detection threshold.

[0059] If the center pixel of the sliding window is greater than or equal to the detection threshold, the center pixel of the sliding window is determined to be the target point; if the center pixel of the sliding window is less than the detection threshold, the center pixel of the sliding window is determined not to be the target point.

[0060] In one embodiment, in step 103, the three-dimensional mask can be obtained according to the following formula:

[0061] I(x,y,z)=M 左 (i x ,y,z)∩M 主 (x,i y ,z)∩M 俯 (x,y,i z )

[0062]

[0063] Where I(x,y,z) represents a three-dimensional mask, i xi y i z Starting from 1, take integers, i x i y i z The subscripts x, y, z correspond to the directions of the x, y, and z axes; M 左 (i x (,y,z) indicates that the preset i has been supplemented. x Left view mask of direction vector, M 主 (x,i y ,z) indicates that the preset i has been supplemented. y The main view mask of the direction vector, M 俯 (x,y,i z ) indicates that the preset i has been supplemented. z Top-view mask of direction vectors;

[0064] And the potential target area data is obtained according to the following formula:

[0065] f(x,y,z)=S(x,y,z)∩I(x,y,z)

[0066] Where f(x,y,z) represents the potential target area data, S(x,y,z) represents the circular scan GBSAR three-dimensional data, and I(x,y,z) represents the three-dimensional mask.

[0067] In one embodiment, step 105, suppressing false alarms in the 3D target point cloud to obtain clustering results, and removing the class containing strong sidelobes from the clustering results, may include:

[0068] Based on an improved density-based noise-based spatial clustering algorithm, false alarms are suppressed in the three-dimensional target point cloud to obtain clustering results. Based on preset features, targets and strong sidelobes in the clustering results are distinguished, and the class containing strong sidelobes in the clustering results is removed. The improved density-based noise-based spatial clustering algorithm adopts improved clustering constraints to make the three-dimensional target point cloud cluster into a cylindrical shape to distinguish targets and strong sidelobes.

[0069] In practice, Figure 3 for Figure 3 This is a flowchart illustrating the improved density-based noise-based spatial clustering algorithm in an embodiment of the present invention, as shown below. Figure 3 As shown, the steps of the improved density-based noise-based spatial clustering algorithm include:

[0070] Step 301: Use Euclidean distance to measure the distance between points in the 3D target point cloud;

[0071] Step 302: Determine the improved clustering constraints as follows: the distance between any two point clouds in the x-coordinate and y-coordinate dimensions shall not exceed a preset distance value, and the distance in the height dimension shall not exceed a preset height value;

[0072] Step 303: Based on the improved constraints, perform clustering in the 3D point cloud, and obtain the clustering results;

[0073] Step 304: Based on preset features, distinguish between the target and strong sidelobes in the clustering results, and remove the class containing the strong sidelobes in the clustering results.

[0074] In one embodiment, the Euclidean distance in step 301 can be expressed as:

[0075]

[0076] Where d represents the Euclidean distance between the points in the 3D target point cloud, (x1,y1,z1) represents the coordinates of the first point, and (x2,y2,z2) represents the coordinates of the second point.

[0077] In one embodiment, the improved clustering constraint in step 302 can be expressed as:

[0078]

[0079] Where, d x d represents the distance along the horizontal axis. y d represents the distance along the vertical axis. z This represents the distance in the height dimension, where r is the preset distance value and h is the preset height value.

[0080] In one embodiment, the preset features in step 105 include one or any combination of angle, shape, size, and position.

[0081] Figure 4 This is a flowchart illustrating a three-dimensional point cloud extraction method according to another embodiment of the present invention. The following is a summary of the process. Figure 4 The following describes the process of the 3D point cloud extraction method in this embodiment of the invention:

[0082] Input circular scan GBSAR 3D data;

[0083] Extract the three-view projection of the maximum value of the three dimensions from a SAR 3D image;

[0084] Two-parameter CFAR is applied to the three-view image for detection to obtain the three-view mask;

[0085] A 3D mask is obtained by projecting a three-view mask onto the 3D data. The intersection of the 3D mask and the 3D data is then used to obtain the potential target area data.

[0086] Global CFAR detection is performed on the potential target region data extracted by intersection to obtain target point detection results; based on the target point detection results, a three-dimensional target point cloud is obtained.

[0087] Based on the characteristics of uneven density of target point clusters in 3D data and their linear structure in vertical planes, the traditional DBSCAN clustering algorithm is improved. The constraints of the clustering algorithm are improved, and clustering is performed in the 3D point cloud according to the constraints to further suppress false alarms. The target and strong sidelobes are distinguished by features such as angle, shape, size, and position. The class containing the strong sidelobes is removed, and the 3D point cloud after false alarm suppression is output.

[0088] This invention also provides a three-dimensional point cloud extraction device, as described in the following embodiments. Since the principle by which this device solves the problem is similar to that of the three-dimensional point cloud extraction method, the implementation of this device can be referred to the implementation of the three-dimensional point cloud extraction method, and repeated details will not be elaborated further.

[0089] Figure 5 This is a schematic diagram of the structure of the three-dimensional point cloud extraction device in an embodiment of the present invention, as shown below. Figure 5 As shown, the device includes:

[0090] The projection extraction module 01 is used to obtain the circular scan GBSAR three-dimensional data of the three-dimensional SAR image, and extract the projection of the maximum value of the three dimensions in the three-dimensional SAR image to obtain the three-view image;

[0091] Detection module 02 is used to detect the three views and obtain a three-view mask;

[0092] Data acquisition module 03 is used to obtain a three-dimensional mask by projecting a three-view mask, and to obtain potential target area data by intersecting the three-dimensional mask with the three-dimensional SAR image.

[0093] The 3D point cloud acquisition module 04 is used to detect the potential target area data and obtain a 3D target point cloud.

[0094] The false alarm suppression module 05 is used to suppress false alarms in the three-dimensional target point cloud, obtain clustering results, distinguish the target and strong sidelobes in the clustering results according to preset features, remove the class of strong sidelobes in the clustering results, and obtain the three-dimensional point cloud after false alarm suppression.

[0095] In one embodiment, the formula for the three views is expressed as:

[0096]

[0097] Where S represents the 3D data obtained from circular scanning GBSAR, S 左 This represents the left view of the three-view diagram, where x0 is the direction vector of the X-axis, and S... 主This represents the front view of the three-view diagram, where y0 is the direction vector of the Y-axis, and S... 俯 This represents the top view of the three-view diagram. z0 is the direction vector of the Z-axis, and max represents the maximum value of the direction vector corresponding to the axis direction of the circular scan GBSAR 3D data.

[0098] In one embodiment, the detection module 02 is specifically used for:

[0099] The three-view image is detected using a two-parameter constant false alarm rate algorithm to obtain a three-view image mask. The three-view image mask is then obtained using the following formula:

[0100]

[0101] Among them, M 左 (y,z) represents the left view mask of the three-view mask, M 主 (x,z) represents the front view mask of the three-view mask, M 俯 (x,y) represents the top view mask of the three-view mask, and CFAR represents the two-parameter constant false alarm rate algorithm.

[0102] In one embodiment, the 3D point cloud acquisition module 04 is specifically used for:

[0103] Based on the global constant false alarm rate algorithm, global constant false alarm rate detection is performed on the potential target area data, the potential target area data is transformed into the first row vector, and the mean and variance of the data in the first row vector are calculated.

[0104] The detection threshold is calculated based on the mean and variance, and the constant false alarm probability.

[0105] By using a sliding window, the data of potential target regions is traversed to obtain the center pixel of the sliding window;

[0106] Based on the detection threshold, determine whether the center pixel of the sliding window is a target point, and obtain the target point detection result;

[0107] Based on the target point detection results, a three-dimensional target point cloud is obtained.

[0108] In one embodiment, the data acquisition module 03 is specifically used for:

[0109] The 3D mask is obtained using the following formula:

[0110] I(x,y,z)=M 左 (i x ,y,z)∩M 主 (x,i y ,z)∩M 俯 (x,y,i z )

[0111]

[0112] Where I(x,y,z) represents a three-dimensional mask, i x i y i z Starting from 1, take integers, i x i y i z The subscripts x, y, z correspond to the directions of the x, y, and z axes; M 左 (i x (,y,z) indicates that the preset i has been supplemented. x Left view mask of direction vector, M 主 (x,i y ,z) indicates that the preset i has been supplemented. y The main view mask of the direction vector, M 俯 (x,y,i z ) indicates that the preset i has been supplemented. z Top-view mask of direction vectors;

[0113] The potential target area data is obtained using the following formula:

[0114] f(x,y,z)=S(x,y,z)∩I(x,y,z)

[0115] Where f(x,y,z) represents the potential target area data, S(x,y,z) represents the circular scan GBSAR three-dimensional data, and I(x,y,z) represents the three-dimensional mask.

[0116] In one embodiment, the false alarm suppression module 05 is specifically used for:

[0117] Euclidean distance is used to measure the distance between points in a 3D target point cloud;

[0118] The improved clustering constraints are determined as follows: the distance between any two point clouds in the x-coordinate and y-coordinate dimensions shall not exceed a preset distance value, and the distance in the height dimension shall not exceed a preset height value;

[0119] Based on the improved constraints, clustering is performed in the 3D point cloud, and the clustering results are as follows:

[0120] Based on preset features, the target and strong sidelobes in the clustering results are distinguished, and the class containing the strong sidelobes in the clustering results is removed.

[0121] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described three-dimensional point cloud extraction method.

[0122] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described three-dimensional point cloud extraction method.

[0123] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described three-dimensional point cloud extraction method.

[0124] In this embodiment of the invention, the three-dimensional point cloud extraction method differs from existing technologies in that, in terms of target detection, the curved observation geometry of circular scanning GBSAR results in strong sidelobes in the image, increasing the difficulty of three-dimensional SAR image target point cloud extraction. The classic two-parameter CFAR algorithm cannot detect accurate target points under the influence of strong sidelobes. In terms of false alarm suppression, the traditional DBSCAN algorithm is sensitive to changes in point density within clusters and cannot effectively identify and segment target areas and strong sidelobes, thus failing to suppress false alarms. Compared to other solutions, this approach obtains 3D GBSAR 3D data from a 3D SAR image, extracts the maximum projection of the three dimensions in the 3D SAR image to obtain three-view images, detects these three-view images to obtain a three-view mask, projects the three-view mask to obtain a 3D mask, and intersects the 3D mask with the 3D SAR image to obtain potential target area data. It then detects the potential target area data to obtain a 3D target point cloud, suppresses false alarms in the 3D target point cloud, obtains clustering results, distinguishes targets from strong sidelobes in the clustering results based on preset features, removes the cluster containing strong sidelobes, and obtains a 3D point cloud with suppressed false alarms. This allows for accurate extraction of the 3D point cloud under the influence of strong sidelobes, thus suppressing false alarms.

[0125] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific 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.

Claims

1. A method for extracting three-dimensional point clouds, characterized in that, include: Obtain circular scan GBSAR 3D data of 3D SAR image, extract the maximum projection of the three dimensions in the 3D SAR image to obtain three views; The three views are detected to obtain a three-view mask; A three-dimensional mask is obtained by projecting the three-view mask, and the intersection of the three-dimensional mask and the three-dimensional SAR image is obtained to obtain the potential target area data. The potential target region data is detected to obtain a three-dimensional target point cloud; False alarms are suppressed in the three-dimensional target point cloud to obtain clustering results. The target and strong sidelobes in the clustering results are distinguished according to preset features. The class of strong sidelobes in the clustering results is removed to obtain the three-dimensional point cloud after false alarm suppression. Detecting the three views to obtain a three-view mask includes: detecting the three views according to the following formula and the two-parameter constant false alarm rate algorithm to obtain the three-view mask. in, This represents the left view mask of the three-view mask. This represents the front view mask of the three-view mask. The top view mask represents the three-view mask, and CFAR represents the two-parameter constant false alarm rate algorithm. The process of detecting the potential target region data to obtain a three-dimensional target point cloud includes: performing global constant false alarm rate (CFAR) detection on the potential target region data according to a global constant false alarm rate algorithm; converting the potential target region data into a first row vector; calculating the mean and variance of the data in the first row vector; calculating a detection threshold based on the mean, variance, and constant false alarm probability; traversing the potential target region data in the form of a sliding window to obtain the center pixel of the sliding window; determining whether the center pixel of the sliding window is a target point based on the detection threshold to obtain the target point detection result; and obtaining the three-dimensional target point cloud based on the target point detection result.

2. The method as described in claim 1, characterized in that, The formula for the three views is expressed as follows: in, This indicates the acquisition of 3D GBSAR data via circular scanning. This represents the left view of the three-view diagram. It is the direction vector of the X-axis. The front view represents the three-view diagram. It is the direction vector of the Y-axis. The top view represents the three views. It is the direction vector of the Z-axis. This indicates the acquisition of 3D GBSAR data via circular scanning. The maximum value of the direction vector corresponding to the axis.

3. The method as described in claim 1, characterized in that, A 3D mask is obtained by projecting a three-view mask onto a 3D image. The intersection of the 3D mask and the 3D SAR image yields the potential target region data, including: The 3D mask is obtained using the following formula: in, Represents a 3D mask. , , Starting from 1, take integers. , , subscript correspond The direction of the axis; This indicates that the preset has been supplemented. Left view mask of direction vector, This indicates that the preset has been supplemented. The main view mask of the direction vector. This indicates that the preset has been supplemented. Top-view mask of direction vectors; The potential target area data is obtained using the following formula: in, This represents potential target area data. This represents circular scan GBSAR 3D data. This represents a three-dimensional mask.

4. The method as described in claim 1, characterized in that, False alarms are suppressed in the three-dimensional target point cloud to obtain clustering results. Based on preset features, targets and strong sidelobes in the clustering results are distinguished, and the class containing the strong sidelobes in the clustering results is removed, including: Based on an improved density-based noise-based spatial clustering algorithm, false alarms are suppressed in the three-dimensional target point cloud to obtain clustering results. Based on preset features, targets and strong sidelobes in the clustering results are distinguished, and the class containing strong sidelobes in the clustering results is removed. The improved density-based noise-based spatial clustering algorithm adopts improved clustering constraints to make the three-dimensional target point cloud cluster into a cylindrical shape to distinguish targets and strong sidelobes.

5. The method as described in claim 4, characterized in that, The steps of the improved density-based noise-based spatial clustering algorithm include: Euclidean distance is used to measure the distance between points in a 3D target point cloud; The improved clustering constraints are determined as follows: the distance between any two point clouds in the x-coordinate and y-coordinate dimensions shall not exceed a preset distance value, and the distance in the height dimension shall not exceed a preset height value; Based on the improved constraints, clustering is performed in the 3D point cloud, and the clustering results are as follows: Based on preset features, the target and strong sidelobes in the clustering results are distinguished, and the class containing the strong sidelobes in the clustering results is removed.

6. The method as described in claim 1, characterized in that, Preset features include one or any combination of angle, shape, size, and position.

7. A three-dimensional point cloud extraction device, characterized in that, include: The projection extraction module is used to obtain the circular scan GBSAR 3D data of the 3D SAR image. It extracts the projection of the maximum value of the three dimensions in the 3D SAR image to obtain the three-view drawing. The detection module is used to detect the three views and obtain a three-view mask; The data acquisition module is used to obtain a three-dimensional mask by projecting a three-view mask, and to obtain potential target area data by intersecting the three-dimensional mask with the three-dimensional SAR image. The 3D point cloud acquisition module is used to detect the potential target area data and obtain a 3D target point cloud; The false alarm suppression module is used to suppress false alarms in the three-dimensional target point cloud, obtain clustering results, distinguish the target and strong sidelobes in the clustering results according to preset features, remove the class of strong sidelobes in the clustering results, and obtain the three-dimensional point cloud after false alarm suppression. Detecting the three views to obtain a three-view mask includes: detecting the three views according to the following formula and the two-parameter constant false alarm rate algorithm to obtain the three-view mask. in, This represents the left view mask of the three-view mask. This represents the front view mask of the three-view mask. The top view mask represents the three-view mask, and CFAR represents the two-parameter constant false alarm rate algorithm. The process of detecting the potential target region data to obtain a three-dimensional target point cloud includes: performing global constant false alarm rate (CFAR) detection on the potential target region data according to a global constant false alarm rate algorithm; converting the potential target region data into a first row vector; calculating the mean and variance of the data in the first row vector; calculating a detection threshold based on the mean, variance, and constant false alarm probability; traversing the potential target region data in the form of a sliding window to obtain the center pixel of the sliding window; determining whether the center pixel of the sliding window is a target point based on the detection threshold to obtain the target point detection result; and obtaining the three-dimensional target point cloud based on the target point detection result.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 6.

Images