Radar image matching method, device, equipment and storage medium

Abstract

The present invention discloses a radar image matching method, apparatus, device, and storage medium. The present invention calculates the phase consistency of a radar image pair to be matched and generates a corresponding first maximum moment map. Based on wavelet transforms at different scales, each first maximum moment map is converted into multiple downsampled second maximum moment maps to form an image pyramid corresponding to each radar image, and control points corresponding to each layer of the image pyramid are extracted. The control points corresponding to different radar images in the radar image pair are matched, and the generated set of matching control points is input into a preset polynomial matching model to obtain transformation parameters for the radar image pair, thereby matching the radar image pair and outputting a matching result. Compared with the prior art, the technical solution of the present invention extracts control points from the generated image pyramid based on wavelet transforms at different scales, which can increase the number of control points subsequently obtained and improve matching accuracy based on the preset polynomial matching model.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a radar image matching method, apparatus, device and storage medium. Background Art

[0002] Image matching is a crucial task in image processing. Currently, a variety of algorithms are used for SAR image registration, including SAR-SIFT, PSO-SIFT, HOPC, CFOG, and RIFT. While RIFT, considered the most effective heterogeneous image matching algorithm, suffers from several drawbacks when matching radar images: First, it lacks adaptability to feature extraction at varying scales; second, it involves numerous parameters, such as patchsize, k, and noiseMethod, but lacks a robust automatic value guide for these parameters; and third, RIFT extracts fewer feature points in images with low correlation, low signal-to-noise ratio, and weak texture, and these feature points are unevenly distributed. Therefore, even with excellent matching results, these drawbacks can still lead to reduced image matching accuracy. Summary of the Invention

[0003] The present invention proposes a radar image matching method, apparatus, device and storage medium. The method can extract control points from a generated image pyramid based on wavelet transforms of different scales, thereby increasing the number of control points subsequently obtained and improving matching accuracy based on a preset polynomial matching model.

[0004] A first aspect of an embodiment of the present invention provides a radar image matching method, the method comprising:

[0005] Acquire a pair of radar images to be matched, calculate the phase consistency of the radar image pairs according to a first preset formula, and generate a first maximum moment map corresponding to each radar image;

[0006] Based on wavelet transforms at different scales, each first maximum moment map is converted into multiple downsampled second maximum moment maps to form an image pyramid corresponding to each radar image. Control points corresponding to each layer of the image pyramid are extracted to generate a control point set.

[0007] Matching control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, inputting the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, matching the radar image pair based on the transformation parameters, and outputting a matching result.

[0008] In a possible implementation manner of the first aspect, extracting control points corresponding to images in each layer of the image pyramid specifically includes:

[0009] Obtaining a top image of the image pyramid, and extracting control points corresponding to the top image based on a FAST operator;

[0010] Acquire an intermediate layer image and a bottom layer image of the image pyramid, and extract first control points corresponding to the intermediate layer image and the bottom layer image based on a multi-operator joint algorithm;

[0011] Simultaneously, obtaining scale ratios of wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and mapping control points corresponding to the top layer image to the intermediate layer image and the bottom layer image, respectively, based on the scale ratios, to obtain mapping points corresponding to the intermediate layer image and the bottom layer image;

[0012] The first control point is compared with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

[0013] In a possible implementation manner of the first aspect, extracting the control points corresponding to the top image based on the FAST operator is specifically:

[0014] Obtain a pixel point set in the top image, and select any pixel point in the pixel point set as a center pixel point;

[0015] Based on a first preset rule, determine whether the central pixel point is a candidate point. If so, perform first candidate point detection. If not, discard the feature point corresponding to the central pixel point, update the pixel point set in the top image, and return to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixels are selected.

[0016] When performing the first candidate point detection, the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points is calculated, and based on the number of the absolute values greater than a preset threshold, whether the central pixel point is a feature point is determined; if so, the pixel point corresponding to the central pixel point is used as the feature point; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the process returns to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected;

[0017] Obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image.

[0018] In a possible implementation manner of the first aspect, before matching control point sets corresponding to different radar images in the radar image pair, the method further includes:

[0019] All pixels in each radar image are sequentially traversed to obtain a multidimensional sequence corresponding to each pixel, and maximum processing is performed on the multidimensional sequence to obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value. The channel index is set to the pixel point corresponding to the pixel point until all pixels are traversed to obtain a maximum index map corresponding to each radar image;

[0020] The control points are obtained, and with each control point as the center, the maximum index map is divided into image blocks of a second preset size, and the image blocks are divided into multiple sub-grids, multiple distribution histograms are constructed for each sub-grid, the multiple distribution histograms are connected to obtain the feature vectors of the control points, and all the obtained feature vectors are normalized.

[0021] A second aspect of an embodiment of the present invention provides a radar image matching device, the device comprising: a maximum moment map generation module, a control point extraction module, and a control point matching module;

[0022] The maximum moment map generation module is configured to obtain a pair of radar images to be matched, calculate the phase consistency of the pair of radar images according to a first preset formula, and generate a first maximum moment map corresponding to each radar image;

[0023] The control point extraction module is configured to convert each first maximum moment map into multiple downsampled second maximum moment maps based on wavelet transforms of different scales, thereby forming an image pyramid corresponding to each radar image, extracting control points corresponding to each layer of the image pyramid, and generating a control point set;

[0024] The control point matching module is configured to match control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, input the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, match the radar image pair based on the transformation parameters, and output a matching result.

[0025] In a possible implementation of the second aspect, the control point extraction module is configured to extract control points corresponding to each layer of the image in the image pyramid, specifically:

[0026] The control point extraction module is used to obtain the top image of the image pyramid and extract the control points corresponding to the top image based on the FAST operator;

[0027] The control point extraction module is used to obtain the middle layer image and the bottom layer image of the image pyramid, and extract the first control points corresponding to the middle layer image and the bottom layer image based on a multi-operator joint algorithm;

[0028] The control point extraction module is used to simultaneously obtain the scale ratios of the wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and map the control points corresponding to the top layer image to the intermediate layer image and the bottom layer image respectively according to the scale ratios to obtain the mapping points corresponding to the intermediate layer image and the bottom layer image;

[0029] The control point extraction module is configured to compare the first control point with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

[0030] In a possible implementation of the second aspect, the control point extraction module is configured to extract control points corresponding to the top image based on a FAST operator, specifically:

[0031] The control point extraction module is used to obtain a pixel point set in the top image and select any pixel point in the pixel point set as a center pixel point;

[0032] The control point extraction module is configured to determine whether the central pixel point is a candidate point based on a first preset rule; if so, perform first candidate point detection; if not, discard the feature point corresponding to the central pixel point, update the pixel point set in the top image, and return to the step of "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected;

[0033] The control point extraction module is configured to calculate the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points when performing the first candidate point detection, and determine whether the central pixel point is a feature point based on the number of the absolute values that are greater than a preset threshold. If so, the pixel point corresponding to the central pixel point is used as the feature point; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the step of "selecting any pixel point in the pixel point set as the central pixel point" is returned to until all the pixel points are selected;

[0034] The control point extraction module is used to obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image.

[0035] In a possible implementation of the second aspect, the method further includes: a feature vector acquisition module;

[0036] The feature vector acquisition module is used to sequentially traverse all pixels in each radar image, obtain a multidimensional sequence corresponding to each pixel, perform maximum processing on the multidimensional sequence, obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value, set the channel index to the pixel point corresponding to the pixel point, and obtain a maximum index map corresponding to each radar image after traversing all pixels;

[0037] The feature vector acquisition module is used to obtain the control points, divide the maximum index map into image blocks of a second preset size with each control point as the center, and divide the image blocks into multiple sub-grids, construct multiple distribution histograms for each sub-grid, connect the multiple distribution histograms, obtain the feature vectors of the control points, and normalize all the obtained feature vectors.

[0038] A third aspect of an embodiment of the present invention provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the radar image matching method as described in any one of the above items is implemented.

[0039] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes: a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the radar image matching methods described above.

[0040] Compared to the prior art, the radar image matching method, device, system, and storage medium provided by embodiments of the present invention have the following beneficial effects: the present invention obtains a radar image pair to be matched and generates a multi-scale downsampled image pyramid based on wavelet transforms at different scales to extract image control points, thereby combining coarse-scale feature information and fine-scale feature information, thereby increasing the number of control points subsequently acquired; based on a preset polynomial matching model, the obtained matching control point set is input into the preset polynomial matching model to achieve matching of the radar image pair to be matched, thereby improving matching accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 1 is a flow chart of a radar image matching method provided by one embodiment of the present invention;

[0042] Figure 2 The figure is a schematic structural diagram of a radar image matching device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0044] Reference Figure 1 , which shows a flow chart of a radar image matching method provided by an embodiment of the present invention.

[0045] As an example, the radar image matching method may include:

[0046] Before matching the radar image pair to be matched, a parameter adaptive calculation criterion is first set, that is, a polynomial matching model is set.

[0047] In one embodiment, the polynomial matching model requires at least 12 pairs of matching points, evenly distributed across the image, to achieve more accurate fitting results. Because six unknown parameters must be fitted and solved in each of the x and y directions, the number of blocks in both directions is greater than 6, for a total of greater than 36. To avoid issues such as too small image block sizes, which result in insufficient information in the local neighborhood of a feature point, making it insufficient for reliable feature point extraction or poorly distinguishable descriptors, the polynomial matching model in this embodiment divides the image into 36 blocks and adaptively calculates the block size based on the input radar image size. This configuration enables adaptive calculation and selection of phase-consistent block sizes and the number of scales in the scale space based on the radar image size and statistical characteristics during subsequent radar image matching, resulting in a more even distribution of feature points. This also reduces human intervention in parameter selection during the image matching process, thereby improving matching accuracy.

[0048] This setup differs from the existing technique of using the RIFT algorithm as the transformation model, which is only applicable to matching transformations with smaller images. Furthermore, solving the six unknown transformation parameters of the affine transformation only requires three pairs of identical feature points, which can lead to low fitting accuracy. In this embodiment, the polynomial matching model is selected to account for potentially complex deformations between the matching images, is unaffected by the size of the radar images being matched, and has a wider range of applicability.

[0049] Step 101: Acquire a pair of radar images to be matched, calculate the phase consistency of the pair of radar images according to a first preset formula, and generate a first maximum moment map corresponding to each radar image.

[0050] In one embodiment, the phase congruency feature of the radar image pair is obtained by using a filter bank of Log-Gabor filters of different orientations and scales.

[0051] In one embodiment, a first preset formula for calculating the phase consistency of the radar image pair is as follows:

[0052]

[0053] Where PC is the phase consistency value, (x, y) represents the coordinates of the point in the image, and W o (x, y) is the weighting factor for a given frequency spread, A no (x, y) is the amplitude at (x, y) for wavelet scale n and direction o, T is the noise threshold, and ε is a small constant to avoid division by zero. Indicates that the closed quantity is equal to itself when its value is positive, otherwise it is zero, ΔΦ no (x, y) is the phase difference function.

[0054] In one embodiment, it is known from classical moment analysis that the uniqueness of an image feature is reflected by the axis corresponding to the maximum moment. Therefore, the maximum moment of the image must be calculated. Before calculating the maximum moment, three intermediate quantities a, b, and c must be calculated. The calculation formulas for the intermediate quantities a, b, and c are as follows:

[0055] a=∑ o (PC(θ o )cos(θ o )) 2

[0056] b=2∑ o (PC(θ o )cos(θ o ))(PC(θ o )sin(θ o ));

[0057] c=∑ o (PC(θ o )sin(θ o )) 2

[0058] Among them, θ o is the angle of direction o, PC(θ o ) is the phase consistency at this angle.

[0059] After calculating the three intermediate quantities, the maximum moment M is calculated using the following formula:

[0060]

[0061] In one embodiment, a first maximum moment map corresponding to each radar image is calculated based on the calculated maximum moment, thereby obtaining a first maximum moment map corresponding to a pair of radar images, which is used to perform edge detection and extraction on the radar images. Specifically, by calculating the maximum moment for each pixel in each radar image, a maximum moment map having the same size as the radar image can be generated.

[0062] Step 102: Based on wavelet transforms at different scales, each first maximum moment map is converted into multiple downsampled second maximum moment maps to form an image pyramid corresponding to each radar image. Control points corresponding to each layer of the image pyramid are extracted to generate a control point set.

[0063] Based on the differences between the radar image pairs to be matched, when the differences are large, control points extracted based only on small-scale edges may not necessarily maintain stability.

[0064] In one embodiment, to address the issue of unstable control points in small-scale edge extraction, a scale space is introduced. Based on wavelet transforms at different scales, each first maximum moment map is converted into multiple downsampled second maximum moment maps at different scales. Based on these downsampled second maximum moment maps at different scales, a layered downsampled image pyramid is formed for each radar image. Control points corresponding to each layer of the image pyramid are extracted, enabling robust extraction of control points from coarse to fine scales. Furthermore, the image pyramid based on wavelet transforms at different scales offers scalable resolution and fidelity, low memory usage, fast encoding and decoding, and strong error resilience.

[0065] In one embodiment, different scales of wavelet transform are adaptively selected. In this embodiment, the number of adaptively selected different scales does not exceed five at most.

[0066] In one embodiment, the adaptive selection method mainly takes the sub-image in the upper left corner as the input image for the next wavelet transform after each wavelet transform, so that each input image is 1 / 4 of the previous scale image. When the sub-image is smaller than 32*32 or has undergone a five-scale wavelet transform, that is, the wavelet transform has been performed five times, no wavelet transform is performed again.

[0067] In one embodiment, after generating an image pyramid, control points corresponding to each layer of the image pyramid are extracted to generate a control point set. In this embodiment, the first layer of the image pyramid is set as the top layer image, the last layer of the image pyramid is set as the bottom layer image, and all images before the top layer image and the bottom layer image are set as intermediate layer images.

[0068] In one embodiment, control points corresponding to the top image of the image pyramid are first extracted. The step of extracting the control points corresponding to the top image of the image pyramid is specifically as follows:

[0069] Step 1021: Acquire the top image of the image pyramid, and extract control points corresponding to the top image based on the FAST operator.

[0070] In one embodiment, the control points corresponding to the top image are extracted based on the FAST operator mainly by:

[0071] A pixel point set in the top image is obtained, and any pixel point in the pixel point set is selected as a central pixel point; specifically, pixel point p is selected as the center, a circle with a radius of 3 is drawn based on the center, and 16 pixel points p1-p16 on the circle are selected.

[0072] Based on the first preset rule, determine whether the central pixel point is a candidate point. If so, perform the first candidate point detection. If not, discard the feature point corresponding to the central pixel point, update the pixel point set in the top image, and return to the step of "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected. Specifically, the first preset rule is: define a preset threshold, calculate the pixel difference between p1, p9 and the central pixel point p respectively, and determine whether their absolute values are all less than the threshold. If so, the central pixel point p is not a candidate point and is discarded; if not, the central pixel point p is a candidate point and the first candidate point detection is performed.

[0073] When performing the first candidate point detection, the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points is calculated, and according to the number of the absolute values greater than the preset threshold, whether the central pixel point is a feature point is determined. If so, the pixel point corresponding to the central pixel point is used as the feature point. If not, the feature point corresponding to the central pixel point is discarded, and the pixel point set in the top image is updated. Return to the step of "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected. Specifically: the first candidate point detection is divided into two parts: the first part detection The method is as follows: if the central pixel point p is a candidate point, the pixel differences between the pixel points p1, p9, p5, and p13 and the central pixel point p are calculated. If at least 3 of their absolute values exceed the threshold, the central pixel point p is regarded as a candidate point and the second part of the detection is performed; otherwise, the central pixel point p is directly discarded; the second part of the detection is as follows: if the central pixel point pp is a candidate point, the pixel differences between the 16 pixel points p1 to p16 and the central pixel point p are calculated. If at least 9 of their absolute values exceed the threshold, the central pixel point p is confirmed as a feature point; otherwise, the central pixel point p is directly discarded.

[0074] Obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image. Specifically: calculate the FAST score values at all feature points, through a neighborhood centered on the feature point P, the neighborhood size can be 3x3 or 5x5, if there are multiple feature points in the neighborhood, then determine the FAST score value of each feature point, if the feature point P is the one with the largest score value among all feature points in the neighborhood, then retain it; otherwise, suppress it. If there is only one feature point in the neighborhood, then retain it. Among them, the FAST score value of each feature point is determined mainly by calculating the sum of the absolute values of the pixel differences between the 16 pixels selected and the center pixel point p when the feature point P is the center pixel point p. The FAST score calculation formula is as follows:

[0075]

[0076] Where V represents the FAST score and t represents the preset threshold.

[0077] Step 1022: Acquire the middle layer image and the bottom layer image of the image pyramid, and extract first control points corresponding to the middle layer image and the bottom layer image based on a multi-operator joint algorithm.

[0078] In one embodiment, a multi-operator joint control point extraction is performed on the edges of the middle and bottom layers of the image, that is, four classic Harris operators, Forstner operators, Moravec operators and SUSAN operators are used for calculation simultaneously. When there is a one-pixel position deviation among at least three operators, the pixel point is determined to be the first control point.

[0079] Step 1023: Simultaneously obtain the scale ratios of the wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and map the control points corresponding to the top layer image to the intermediate layer image and the bottom layer image respectively according to the scale ratios to obtain the mapping points corresponding to the intermediate layer image and the bottom layer image.

[0080] In one embodiment, when the image pyramid has 1, 2, 3, 4, or 5 layers from the top to the bottom, the first layer image is the top layer image, the fifth layer image is the bottom layer image, and the second, third, and fourth layers images are the middle layer images. When mapping from the first layer to the second layer, the horizontal and vertical coordinates need only be multiplied by 2 to map the control points corresponding to the first layer image onto the second layer image, obtaining the mapping points corresponding to the second layer image. When mapping from the first layer to the third layer, the horizontal and vertical coordinates need to be multiplied by 4 to map the control points corresponding to the first layer image onto the third layer image, obtaining the mapping points corresponding to the third layer image. When mapping from the first layer to the fourth layer, the horizontal and vertical coordinates need to be multiplied by 8 to map the control points corresponding to the first layer image onto the fourth layer image, obtaining the mapping points corresponding to the fourth layer image. Therefore, the scale ratio of each layer is different. The top layer image is the original image, the next layer is the one-scale wavelet transform result, the next layer is the two-scale wavelet transform result, the next layer is the three-scale wavelet transform result, and so on, until the mapping of the image pyramid is completed and the mapping points corresponding to the middle layer image and the bottom layer image are obtained.

[0081] Step 1024: Compare the first control point with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

[0082] Specifically, the first control points and mapping points corresponding to the extracted intermediate layer image and the underlying image are compared, the repetition of the first control points and the mapping points in the same layer is calculated, the points with high repetition are retained, and the points with low repetition are eliminated to obtain the corresponding control points of the intermediate layer image and the underlying image.

[0083] In this embodiment, by introducing scale space, constructing wavelet pyramids of different scales and multi-operator joint control point extraction, the algorithm can combine large-scale features with small-scale features, which can greatly increase the number of matching point pairs obtained subsequently.

[0084] Step 103: Match control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, input the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, match the radar image pair based on the transformation parameters, and output a matching result.

[0085] In one embodiment, before matching the control point sets corresponding to different radar images in the radar image pair, all pixels in each radar image are traversed in sequence to obtain a multidimensional sequence corresponding to each pixel. Maximum processing is performed on the multidimensional sequence to obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value. The channel index is set to the pixel point corresponding to the pixel point until all pixels are traversed to obtain a maximum index map corresponding to each radar image.

[0086] Specifically, the multidimensional sequence corresponding to each pixel point (x, y) is obtained by obtaining the amplitude A of each pixel point for the wavelet scale n and direction o. no (x, y). Based on a direction o, the amplitudes of the pixels at all scales are added together to obtain a log-Gabor layer A o (x, y), the formula is as follows:

[0087] A o (x, y) = ∑ n A no (x, y).

[0088] Based on different directions o, obtain multiple log-Gabor layers A o (x, y), multiple log-Gabor layers are arranged in order to obtain a log-Gabor convolution sequence, which is a multi-channel convolution map Among them, each log-Gabor layer is a channel of the convolutional graph, N o is the number of directions o, the superscript ω represents the different channels of the log-Gabor convolution sequence, and the value range is ω = 1, 2, ..., N o .

[0089] Perform maximum processing on the multidimensional sequence corresponding to each pixel point. The maximum value acquisition formula is as follows:

[0090]

[0091] Get the maximum value A in a multidimensional sequence max (x, y) and its corresponding channel index ω max . max Set to the pixel value of the pixel point (x, y). Perform the above operation on each pixel to obtain the maximum index map corresponding to the radar image pair.

[0092] In one embodiment, after obtaining the maximum index map, the maximum index map is further divided into image blocks of a second preset size based on the extracted control point set and with each control point as the center, and the image blocks are divided into multiple subgrids, multiple distribution histograms are constructed for each subgrid, the multiple distribution histograms are connected to obtain the feature vector of the control point, and all the obtained feature vectors are normalized. Specifically, an image block of size KxK centered on the control point is selected, and a weight is assigned to each pixel in the image block using a Gaussian function, wherein the standard deviation of the Gaussian function is σ=K / 2; and the image block is divided into 6x6 subgrids, and N distribution histograms are constructed for each subgrid. o The distribution histogram of the boxes, where N o It is the upper limit of the value range of each pixel on the maximum index map.

[0093] In one embodiment, after obtaining the feature vectors of all control points, the control point sets corresponding to different radar images in the radar image pair are matched to generate a matched control point set. Specifically, a control point in the first radar image of the radar image pair is matched with all control points in the second radar image. Based on the feature vectors of the control points, the Euclidean distance between the control point in the first radar image and all control points in the second radar image is calculated. The feature point pair with the minimum Euclidean distance is used as the matched control point. This process is repeated for all control points in the first radar image to generate a matched control point set. False matches are then removed using the NBCS algorithm to obtain a matched control point set with high matching accuracy.

[0094] In one embodiment, the matching control point set is input into a preset polynomial matching model to obtain transformation parameters of the radar image pair, and the radar image pair is matched based on the transformation parameters to output a matching result.

[0095] Specifically, the transformation parameters of the radar image pair can be obtained by fitting the matching offsets corresponding to the feature points to obtain the matching offsets at other pixels. The transformation relationship can be described by a second-order polynomial, where the second-order polynomial is as follows:

[0096]

[0097] Where a0–a5 and b0–b5 are fitting parameters, obtained using the least squares method based on the offset of the successfully matched control points combined with the aforementioned second-order polynomial. (m, n) are the pixel coordinates in the radar image, and (u, v) are the offsets between the corresponding matching positions in the two radar images.

[0098] Based on the least squares method, the distance between the offset between the pixel coordinates in the radar image and the corresponding matching position in the radar image and the fitting parameters is minimized, the fitting degree is higher, and high-precision matching of the radar image pairs to be matched is achieved.

[0099] In this embodiment, an embodiment of the present invention provides a radar image matching method, which has the following beneficial effects: the present invention obtains a radar image pair to be matched and generates a multi-scale downsampled image pyramid based on wavelet transforms at different scales to extract image control points, thereby combining coarse-scale feature information and fine-scale feature information, thereby increasing the number of control points subsequently obtained; based on a preset polynomial matching model, the obtained matching control point set is input into the preset polynomial matching model to achieve matching of the radar image pair to be matched, thereby improving matching accuracy.

[0100] The embodiment of the present invention also provides a radar image matching device, see Figure 2 , which shows a schematic structural diagram of a radar image matching device provided by an embodiment of the present invention.

[0101] As an example, the radar image matching device may include: a maximum moment map generation module 201 , a control point extraction module 202 and a control point matching module 203 .

[0102] A maximum moment map generation module 201 is configured to obtain a pair of radar images to be matched, calculate the phase congruence of the radar image pairs according to a first preset formula, and generate a first maximum moment map corresponding to each radar image;

[0103] The control point extraction module 202 is configured to convert each first maximum moment map into multiple downsampled second maximum moment maps based on wavelet transforms at different scales, thereby forming an image pyramid corresponding to each radar image, extracting control points corresponding to each layer of the image pyramid, and generating a control point set.

[0104] The control point matching module 203 is configured to match the control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, input the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, match the radar image pair based on the transformation parameters, and output a matching result.

[0105] Optionally, the control point extraction module is used to extract control points corresponding to each layer of the image in the image pyramid, specifically:

[0106] The control point extraction module is used to obtain the top image of the image pyramid and extract the control points corresponding to the top image based on the FAST operator;

[0107] The control point extraction module is used to obtain the middle layer image and the bottom layer image of the image pyramid, and extract the first control points corresponding to the middle layer image and the bottom layer image based on a multi-operator joint algorithm;

[0108] The control point extraction module is used to simultaneously obtain the scale ratios of the wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and map the control points corresponding to the top layer image to the intermediate layer image and the bottom layer image respectively according to the scale ratios to obtain the mapping points corresponding to the intermediate layer image and the bottom layer image;

[0109] The control point extraction module is configured to compare the first control point with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

[0110] Optionally, the control point extraction module is used to extract the control points corresponding to the top image based on the FAST operator, specifically:

[0111] The control point extraction module is used to obtain a pixel point set in the top image and select any pixel point in the pixel point set as a center pixel point.

[0112] The control point extraction module is used to determine whether the central pixel point is a candidate point based on a first preset rule. If so, the first candidate point detection is performed; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the step of "selecting any pixel point in the pixel point set as the central pixel point" is returned to until all the pixel points are selected.

[0113] The control point extraction module is used to calculate the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points when performing the first candidate point detection, and determine whether the central pixel point is a feature point based on the number of the absolute values greater than a preset threshold. If so, the pixel point corresponding to the central pixel point is used as the feature point; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the step of "selecting any pixel point in the pixel point set as the central pixel point" is returned to until all the pixel points are selected.

[0114] The control point extraction module is used to obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image.

[0115] Optionally, the radar image matching device further includes: a feature vector acquisition module.

[0116] The feature vector acquisition module is used to sequentially traverse all pixel points in each radar image, obtain a multidimensional sequence corresponding to each pixel point, perform maximum value processing on the multidimensional sequence, obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value, set the channel index to the pixel point of the corresponding pixel point, until all pixel points are traversed to obtain a maximum index map corresponding to each radar image.

[0117] The feature vector acquisition module is used to obtain the control points, divide the maximum index map into image blocks of a second preset size with each control point as the center, and divide the image blocks into multiple sub-grids, construct multiple distribution histograms for each sub-grid, connect the multiple distribution histograms, obtain the feature vectors of the control points, and normalize all the obtained feature vectors.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0119] Furthermore, an embodiment of the present application also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the radar image matching method described in the above embodiment is implemented.

[0120] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the radar image matching method as described in the above embodiment.

[0121] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A radar image matching method, characterized in that: include: Acquire a pair of radar images to be matched, calculate the phase consistency of the radar image pairs according to a first preset formula, and generate a first maximum moment map corresponding to each radar image; Based on wavelet transforms at different scales, each first maximum moment map is converted into multiple downsampled second maximum moment maps to form an image pyramid corresponding to each radar image. Control points corresponding to each layer of the image pyramid are extracted to generate a control point set. matching control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, inputting the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, matching the radar image pair based on the transformation parameters, and outputting a matching result; The extracting of control points corresponding to each layer of the image in the image pyramid is specifically as follows: Obtaining a top image of the image pyramid, and extracting control points corresponding to the top image based on a FAST operator; Acquire an intermediate layer image and a bottom layer image of the image pyramid, and extract first control points corresponding to the intermediate layer image and the bottom layer image based on a multi-operator joint algorithm; Simultaneously, obtaining scale ratios of wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and mapping control points corresponding to the top layer image to the intermediate layer image and the bottom layer image, respectively, based on the scale ratios, to obtain mapping points corresponding to the intermediate layer image and the bottom layer image; The first control point is compared with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

2. A radar image matching method according to claim 1, characterized in that: The control points corresponding to the top image are extracted based on the FAST operator, specifically: Obtain a pixel point set in the top image, and select any pixel point in the pixel point set as a center pixel point; Based on a first preset rule, determine whether the central pixel is a candidate point. If so, perform first candidate point detection. If not, discard the feature point corresponding to the central pixel, update the pixel point set in the top image, and return to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixels are selected. When performing the first candidate point detection, the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points is calculated, and based on the number of the absolute values greater than a preset threshold, whether the central pixel point is a feature point is determined; if so, the pixel point corresponding to the central pixel point is used as the feature point; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the process returns to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected; Obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image.

3. The radar image matching method according to claim 1, wherein: Before matching the control point sets corresponding to different radar images in the radar image pair, the method further includes: All pixels in each radar image are sequentially traversed to obtain a multidimensional sequence corresponding to each pixel, and maximum processing is performed on the multidimensional sequence to obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value. The channel index is set to the pixel point corresponding to the pixel point until all pixels are traversed to obtain a maximum index map corresponding to each radar image; The control points are obtained, and with each control point as the center, the maximum index map is divided into image blocks of a second preset size, and the image blocks are divided into multiple sub-grids, multiple distribution histograms are constructed for each sub-grid, the multiple distribution histograms are connected to obtain the feature vectors of the control points, and all the obtained feature vectors are normalized.

4. A radar image matching device, characterized in that: include: Maximum moment map generation module, control point extraction module and control point matching module; The maximum moment map generation module is configured to obtain a pair of radar images to be matched, calculate the phase consistency of the pair of radar images according to a first preset formula, and generate a first maximum moment map corresponding to each radar image; The control point extraction module is configured to convert each first maximum moment map into multiple downsampled second maximum moment maps based on wavelet transforms of different scales, thereby forming an image pyramid corresponding to each radar image, extracting control points corresponding to each layer of the image pyramid, and generating a control point set; The control point matching module is configured to match control point sets corresponding to different radar images in the radar image pair to generate a matching control point set, input the matching control point set into a preset polynomial matching model to obtain transformation parameters of the radar image pair, match the radar image pair based on the transformation parameters, and output a matching result; The control point extraction module is used to extract the control points corresponding to each layer of the image in the image pyramid, specifically: The control point extraction module is used to obtain the top image of the image pyramid and extract the control points corresponding to the top image based on the FAST operator; The control point extraction module is used to obtain the middle layer image and the bottom layer image of the image pyramid, and extract the first control points corresponding to the middle layer image and the bottom layer image based on a multi-operator joint algorithm; The control point extraction module is used to simultaneously obtain the scale ratios of the wavelet transforms corresponding to the intermediate layer image and the bottom layer image, and map the control points corresponding to the top layer image to the intermediate layer image and the bottom layer image respectively according to the scale ratios to obtain the mapping points corresponding to the intermediate layer image and the bottom layer image; The control point extraction module is configured to compare the first control point with the mapping point to obtain control points corresponding to the intermediate layer image and the bottom layer image.

5. The radar image matching device according to claim 4, characterized in that: The control point extraction module is used to extract the control points corresponding to the top image based on the FAST operator, specifically: The control point extraction module is used to obtain a pixel point set in the top image and select any pixel point in the pixel point set as a center pixel point; The control point extraction module is configured to determine, based on a first preset rule, whether the central pixel point is a candidate point; if so, perform first candidate point detection; if not, discard the feature point corresponding to the central pixel point, update the pixel point set in the top image, and return to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected; The control point extraction module is configured to calculate the absolute value of the pixel difference between the central pixel point and a plurality of preset pixel points when performing the first candidate point detection, and determine whether the central pixel point is a feature point based on the number of the absolute values that are greater than a preset threshold. If so, the pixel point corresponding to the central pixel point is used as the feature point; if not, the feature point corresponding to the central pixel point is discarded, the pixel point set in the top image is updated, and the process returns to step "selecting any pixel point in the pixel point set as the central pixel point" until all the pixel points are selected; The control point extraction module is used to obtain all feature points in the top image, calculate the FAST score values corresponding to all feature points, and set a neighborhood of a first preset size with each feature point as the center, and use the feature point corresponding to the largest FAST score value in the neighborhood as the control point of the top image.

6. The radar image matching device according to claim 4, characterized in that: Also includes: Feature vector acquisition module; The feature vector acquisition module is used to sequentially traverse all pixels in each radar image, obtain a multidimensional sequence corresponding to each pixel, perform maximum processing on the multidimensional sequence, obtain the maximum value in the multidimensional sequence and the channel index corresponding to the maximum value, set the channel index to the pixel point corresponding to the pixel point, and obtain a maximum index map corresponding to each radar image after traversing all pixels; The feature vector acquisition module is used to obtain the control points, divide the maximum index map into image blocks of a second preset size with each control point as the center, and divide the image blocks into multiple sub-grids, construct multiple distribution histograms for each sub-grid, connect the multiple distribution histograms, obtain the feature vectors of the control points, and normalize all the obtained feature vectors.

7. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the radar image matching method according to any one of claims 1 to 3 when executing the computer program.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the radar image matching method according to any one of claims 1 to 3.

Images