Array interferometric synthetic aperture radar three-dimensional point cloud registration method

Abstract

This invention relates to a method for registering three-dimensional point clouds using array interferometric synthetic aperture radar (InSAR). The method utilizes an airborne platform equipped with an array InSAR antenna to acquire array SAR images, performs image registration, and calculates multi-band three-dimensional point clouds pixel by pixel. It calculates the nearest distance to each point in the point cloud, averages all distance values ​​to obtain the point cloud resolution, extracts strong target points, and constructs intensity feature descriptors to register the array InSAR three-dimensional point clouds. This method is suitable for solving the problem of difficult registration of array InSAR three-dimensional point clouds, improving the relative geometric accuracy between array InSAR point clouds obtained from different flight zones, and is beneficial for the mapping applications of array InSAR technology.

Description

Technical Field

[0001] This patent relates to a method for processing and applying array interferometric SAR (Synthetic Aperture Radar) data, and in particular to a three-dimensional point cloud registration method for array interferometric synthetic aperture radar based on strong point targets. Background Technology

[0002] Compared to two-dimensional synthetic aperture radar (SAR) imaging, three-dimensional SAR imaging can eliminate target overlay caused by side-looking radar imaging, significantly improving target resolution and identification capabilities, and is one of the important directions for the development of SAR technology. The array interferometric SAR technology proposed by the Chinese Academy of Sciences utilizes a cross-course array antenna for side-looking imaging. Based on the acquired multi-angle observation data, it achieves three-dimensional reconstruction of complex scenes, obtaining regional three-dimensional point clouds. Due to factors such as orbital errors, there may be some displacement and rotation between array InSAR point clouds obtained from different flight zones, affecting the further application of the point clouds.

[0003] To unify point clouds into a single coordinate system, registration processing is required for different point clouds. Commonly used point cloud registration methods mainly include: (1) ICP (Iterative Closest Point) algorithm, which iteratively solves the transformation relationship between point clouds by finding the nearest point; (2) Global optimal registration method based on feature points. Commonly used feature point extraction algorithms include NARF (Normal Aligned Radial Feature) algorithm, ISS (Intrinsic Shape Signatures) algorithm, etc. The extracted point cloud feature points are generally corner points, vertices, etc. in the point cloud. Commonly used point cloud feature descriptors include FPFH (FastPoint Feature Histogram) algorithm, SHOT (Signature of Histograms of Orientations) algorithm, Spin image algorithm, etc. These feature descriptors match feature points based on the geometric features of the point cloud distribution. In addition, there are point cloud registration algorithms that convert to the frequency domain. These algorithms have achieved good application results in point clouds obtained by LiDAR and depth cameras. Unlike other point cloud acquisition methods, array InSAR technology cannot directly obtain point cloud echoes from the surface of the target. Its point cloud acquisition primarily involves recalculating after registering SAR images. Furthermore, because SAR is a coherent imaging method, array InSAR point clouds contain numerous clutter points. Therefore, array InSAR point clouds are characterized by non-surface distribution and a high number of clutter points, leading to poor performance of traditional point cloud registration algorithms. Even after registration, there is still some displacement and rotation between the two point clouds. Summary of the Invention

[0004] This invention provides a three-dimensional point cloud registration method for array interferometric synthetic aperture radar (InSAR). This method, based on strong point targets, aims to address the challenge of applying existing point cloud registration techniques to array InSAR point clouds. It fully utilizes the intensity information of the array InSAR point clouds, extracting strong point targets and constructing intensity feature descriptors to achieve point cloud registration. This method effectively solves the problem of difficult array InSAR point cloud registration and significantly improves the relative geometric accuracy of the point clouds.

[0005] A method for registering three-dimensional point clouds of an array interferometric synthetic aperture radar (AAP) system involves extracting strong point targets, constructing and matching intensity feature descriptors, and calculating the transformation relationship between point clouds to achieve registration between point clouds from different flight zones. The method includes the following steps:

[0006] S1: Use an array InSAR antenna mounted on an airborne platform to acquire array SAR images, perform image registration, and calculate pixel-by-pixel three-dimensional point clouds for multi-strips.

[0007] S2: Select adjacent array interferometric SAR point clouds as the point clouds to be processed, calculate the distance to the nearest point of each point in the point cloud, and average all distance values ​​to obtain the point cloud resolution.

[0008] S3: Extracting strong point targets from point clouds based on intensity information;

[0009] S4: For each point in the point cloud, obtain the nearest points based on the coordinate information, and further calculate the intensity normal vector;

[0010] S5: Construct a feature descriptor for the intensity of a strong point target based on the intensity normal vector and the intensity information of neighboring points;

[0011] S6: After obtaining the features of strong point targets, under the set distance constraints, obtain the corresponding points in the point cloud based on the similarity criterion;

[0012] S7: Solve the transformation relationship between adjacent point clouds based on the same points in the point cloud.

[0013] Furthermore, in step S3, strong point targets are used as point cloud feature points for point cloud registration, and (X) is set. i ,Y i Z i ,I i ), (X j ,Y j Z j ,I j(x, y, z) and (i) represent the coordinates and intensity of the two acquired array interferometric InSAR point clouds P and Q, respectively, where (x, y, z) and (i) represent the coordinates and intensity of the point cloud, respectively, i = [1 … M], j = [1 … N], and M and N represent the number of points in P and Q, respectively. The extraction method of the strong point target is expressed as follows:

[0014] I>I0 (1)

[0015] Where I0 represents the intensity threshold for extracting strong point targets.

[0016] Furthermore, in step S4, for a strong point target p, using a multiple of the resolution D as the radius value, points within a certain radius range are selected as the neighboring point set P', and the intensity gradient vector is calculated using the following formula:

[0017]

[0018] Where p' represents a point in the neighborhood, p'∈P', and N represents the number of points in the neighborhood.

[0019] Furthermore, in step S5, an intensity feature descriptor based on the intensity gradient is constructed. For each point, points within a set distance range are selected as neighborhood points. The intensity normal vector calculated by equation (2) is then used... This means converting the intensity normal vector from a Cartesian coordinate array to spherical coordinates:

[0020]

[0021]

[0022]

[0023] Where az, el, and r represent the azimuth, elevation, and radius, respectively;

[0024] After converting the intensity vectors of neighboring points into spherical coordinates, the distance between the point scatterer and its neighboring points is divided into equal parts. The mean value of the gradient vector r in each interval is calculated to obtain the intensity gradient histogram f, which serves as the feature descriptor of the point scatterer.

[0025] Furthermore, in step S6, corresponding points in the point cloud are obtained using the sum of squares measure, and the points with the minimum sum of squares measure of the feature vectors are selected as corresponding points:

[0026] argmin||f p1 -f p2 || (6)

[0027] Where f represents the feature vector, and p1 and p2 represent strong point targets in different point clouds; based on the same points, the coordinate transformation relationship between different point clouds can be established through the random sampling consensus algorithm.

[0028] Furthermore, in step S7, the random sampling consensus algorithm establishes the coordinate transformation relationship between different point clouds as follows:

[0029]

[0030] Where T represents the translation parameter, s represents the scale parameter, and R represents the rotation matrix.

[0031] This invention provides a three-dimensional point cloud registration method for interferometric synthetic aperture radar (SAR) based on strong point targets, which can solve the registration problem of SAR point clouds. Considering that tomographic SAR technology is affected by the characteristics of SAR technology itself, and shares similarities with SAR, the point scatterer-based registration approach proposed in this paper is also applicable to solving the registration problem of tomographic SAR point clouds. This invention registers point clouds based on strong point targets rather than feature points, effectively avoiding the problem of not being able to extract feature points due to the non-surface distribution of SAR point clouds. This invention constructs feature descriptors for strong point targets based on intensity gradients, which can effectively express the characteristics of strong point targets and obtain corresponding points in different point clouds. This invention effectively improves the geometric accuracy between SAR point clouds, and can further promote the application of SAR technology in surveying and mapping. Attached Figure Description

[0032] The present invention will now be described in further detail with reference to the accompanying drawings, so that the above-mentioned advantages of the present invention become more apparent.

[0033] Figure 1 This is a flowchart of the three-dimensional point cloud registration method for array interferometric synthetic aperture radar described in this invention;

[0034] Figure 2 This is a comparison image of point cloud distribution and LiDAR point cloud obtained based on array interferometric SAR technology;

[0035] Figure 3 This is a diagram showing the transformation relationship between the Cartesian coordinate system and the spherical coordinate system;

[0036] Figure 4A and Figure 4B These are comparison images showing the effects of applying this invention. Figure 4A To register the previous point cloud map, Figure 4B The result is after registration using the algorithm of this invention. Detailed Implementation

[0037] The invention will now be described in further detail with reference to the accompanying drawings. The invention achieves registration between point clouds from different flight zones by extracting strong point targets, constructing intensity feature descriptors, matching them, and calculating the transformation relationship between point clouds. The entire processing flow is as follows: Figure 1 As shown:

[0038] Step S1: Use an array interferometric SAR antenna mounted on an airborne platform to acquire array SAR images, perform image registration, and calculate pixel by pixel to obtain a multi-strip 3D point cloud.

[0039] Step S2: Select adjacent array interferometric SAR point clouds as the point clouds to be processed. For each point in the point cloud, calculate the distance to the nearest point, and average the nearest distances of all points to obtain the resolution D of the point cloud.

[0040] Step S3, extract strong point targets from the point cloud based on intensity information: For each point cloud, select strong point targets as feature points. The characteristics of the non-surface distribution of array interferometric SAR point clouds are as follows... Figure 2 As shown, traditional feature point matching algorithms based on geometric features utilize feature points on the point cloud surface, which are not applicable to array interferometric SAR point clouds. This invention utilizes intensity information to extract strong point targets from the point cloud, selecting points with intensity greater than a certain threshold as strong point targets.

[0041] Assume (X) i ,Y i Z i ,I i ), (X j ,Y j Z j ,I j Let P and Q represent the two obtained array interferometric SAR point clouds, respectively, where (X,Y,Z) and (I) represent the coordinates and intensity of the point cloud, respectively, i = [1 … M], j = [1 … N], and M and N represent the number of points in P and Q, respectively. Then, the extraction method of strong point targets in step (3) can be expressed as follows:

[0042] I>I0 (1)

[0043] Where I0 represents the intensity threshold for extracting strong point targets.

[0044] Step S4: For each point in the point cloud, obtain neighboring points based on coordinate information, and further calculate the intensity normal vector: For a strong point target p, use a multiple of the resolution D as the radius value, and select points within a certain radius range as the neighboring point set P'. The intensity gradient vector can be further calculated using Equation 2:

[0045]

[0046] Where p' represents a point in the neighborhood, p'∈P', (X p ,Y p Z p (X) represents the coordinates of the strong point target. p' ,Y p' Z p' ) represents the coordinates of a point in the neighborhood, and N represents the number of points in the neighborhood.

[0047] For each point, points within a certain distance range are selected as neighborhood points. The intensity normal vector calculated by Equation 2 is used... This indicates that the intensity normal vector is converted from a Cartesian coordinate system to spherical coordinates. The conversion relationship between Cartesian and spherical coordinates is as follows: Figure 3 The diagram shows the range of the azimuth angle (az) and elevation angle (el), where the range is [-180, 180] and the range is [-90, 90]. The conversion relationship between Cartesian coordinate arrays and spherical coordinates is as follows:

[0048]

[0049]

[0050]

[0051] Where az, el, and r represent the azimuth, elevation, and radius, respectively.

[0052] Step S5: After converting the intensity vector of the neighboring points into a spherical coordinate system, the distance between the point scatterer and the neighboring points is divided equally, and the mean value of the gradient vector r in each interval is calculated to obtain the intensity gradient histogram f, which serves as the feature descriptor of the point scatterer.

[0053] Step S6: After obtaining the features of the strong point target, corresponding points are obtained based on the similarity criterion under the set distance constraint.

[0054] Select the point with the minimum value of the sum of squares of the eigenvectors as the corresponding point:

[0055] argmin||f p1 -f p2 || (6)

[0056] Where f represents the feature vector, and p1 and p2 represent strong point targets in different point clouds.

[0057] Step S7: Solve the transformation relationship between adjacent point clouds based on corresponding points in the point cloud: obtain corresponding points using the sum of squared eigenvectors as the measure. Further, based on these corresponding points, a random sampling consensus algorithm can be used to establish the coordinate transformation relationship between different point clouds.

[0058]

[0059] Where T represents the translation parameter, s represents the scale parameter, and R represents the rotation matrix.

[0060] The method described herein is applicable to solving the problem of difficult registration of three-dimensional point clouds in array interferometric SAR, improving the relative geometric accuracy between array interferometric SAR point clouds obtained from different flight zones, and is beneficial to the mapping application of array interferometric SAR technology.

[0061] The 3D point cloud registration method based on strong point targets described in this invention effectively avoids the problem of feature point extraction failure caused by the non-surface distribution of array InSAR point clouds. This invention constructs feature descriptors for strong point targets based on intensity gradients, which can effectively express the characteristics of strong point targets and obtain corresponding points in different point clouds. This invention effectively improves the geometric accuracy between array interferometric SAR point clouds, and can further promote the application of array InSAR technology in surveying and mapping. The 3D point cloud registration method based on strong point targets described in this invention is applicable to point clouds generated by array interferometric SAR and tomographic SAR techniques.

[0062] An array InSAR point cloud registration experiment was conducted using this invention, and the results are shown in Figure 4. The upper figure shows the relationship between point clouds before registration, while the lower figure shows the point cloud relationship after applying the algorithm of this invention. As can be seen from the areas marked with black circles in the figures, the relative accuracy of the point clouds was significantly improved after applying this invention.

[0063] The specific embodiments described above are merely exemplary. Under the guidance of the teachings of this invention, those skilled in the art can make various improvements and modifications based on the above embodiments, and such improvements or modifications fall within the protection scope of this invention. Those skilled in the art should understand that the above specific description is only for the purpose of explaining this invention and is not intended to limit this invention. The protection scope of this invention is defined by the claims and their equivalents.

Claims

1. A method for three-dimensional point cloud registration in array interferometric synthetic aperture radar, characterized in that: By extracting strong target points, constructing intensity feature descriptors and matching them, and then calculating the transformation relationship between point clouds, registration between point clouds of different flight strips is achieved, including the following steps: S1: Use an array InSAR antenna mounted on an airborne platform to acquire array SAR images, perform image registration, and calculate pixel-by-pixel three-dimensional point clouds for multi-strips. S2: Select adjacent array interferometric SAR point clouds as the point clouds to be processed, calculate the distance to the nearest point of each point in the point cloud, and average all distance values ​​to obtain the point cloud resolution. S3: Extracting strong point targets from point clouds based on intensity information; S4: For each point in the point cloud, obtain the nearest points based on the coordinate information, and further calculate the intensity gradient vector; strong targets With resolution Using multiples of a certain value as the radius, select points within a certain radius range as the nearest neighbor set. The intensity gradient vector is calculated using the following formula: ， in Represents points within the neighborhood. , Indicates the number of points in the neighborhood; S5: Construct a feature descriptor for the intensity of a strong point target based on the intensity gradient vector and the intensity information of neighboring points; S6: After obtaining the features of strong point targets, under the set distance constraints, obtain the corresponding points in the point cloud based on the similarity criterion; S7: Solve the transformation relationship between adjacent point clouds based on the same points in the point cloud.

2. The three-dimensional point cloud registration method for array interferometric synthetic aperture radar according to claim 1, characterized in that, In step S3, strong point targets are used as point cloud feature points for point cloud registration, and settings are made. , These represent the two acquired array interferometric InSAR point clouds. ,in , These represent the coordinates and intensity of the point cloud, respectively. , and They represent The number of midpoints, and the extraction method of the strong point targets, are expressed as follows: ， in, This represents the intensity threshold for extracting strong targets.

3. The three-dimensional point cloud registration method for array interferometric synthetic aperture radar according to claim 1, characterized in that, In step S5, an intensity feature descriptor based on the intensity gradient is constructed. For each point, points within a set distance range are selected as neighborhood points. The intensity gradient vector calculated by equation (1) is used... This means converting the intensity gradient vector from a Cartesian coordinate array to spherical coordinates: ， （4）， ， in, These represent the azimuth, elevation, and radius, respectively. Divide the distance between the point scatterer and its neighboring points into equal parts, and calculate the mean of the intensity gradient vector in each interval to obtain the intensity gradient histogram f, which serves as the feature descriptor.

4. The three-dimensional point cloud registration method for array interferometric synthetic aperture radar according to claim 1, characterized in that, In step S6, corresponding points in the point cloud are obtained using the sum-of-squares measure, and the points with the minimum sum-of-squares measure of the feature descriptors are selected as corresponding points. ， in, Represents a feature descriptor. It represents strong point targets in different point clouds; based on the same points, the coordinate transformation relationship between different point clouds can be established through the random sampling consensus algorithm.

5. The three-dimensional point cloud registration method for array interferometric synthetic aperture radar according to claim 4, characterized in that, In step S7, the random sampling consensus algorithm establishes the coordinate transformation relationship between different point clouds as follows: ， in, Indicates the translation parameter. Indicates the scale parameter. This represents the rotation matrix.

Images