Satellite stereo data image space compensation parameter model adaptive adjustment method

Abstract

The invention discloses an adaptive adjustment method for a satellite stereo data image space compensation parameter model. Relates to the technical field of satellite image processing, in particular to the technical field of adaptive adjustment of a satellite three-dimensional data image space compensation parameter model. The problem that in the prior art, a cubic image pair cannot be balanced, positioning errors need to be effectively eliminated, and overfitting needs to be prevented is effectively solved, self-adaptive adjustment of the image space compensation parameter order of each stereo image pair is achieved, a high-precision stereo result can be generated, and self-adaptive adjustment can be achieved. The method comprises the following steps: extracting homonymy points of all stereo image pairs in stereo production, adding a group of first-order image space compensation parameters to all stereo image pairs, circulating block adjustment and epipolar error until the precision is stable, and evaluating whether the precision index of each stereo image pair meets the requirement, if the stereo image pair does not meet the requirement, the image space compensation parameter is improved to be a second-order image space compensation parameter, and cyclic block adjustment is carried out again to obtain a high-precision adjustment result.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite image processing, and specifically to a technical field of adaptive adjustment of a satellite stereo data image space compensation parameter model. Background Art

[0002] Currently, commercial remote sensing satellites are launched frequently. As the number of satellites increases, a large amount of in-track and cross-track satellite stereo data has been obtained through shooting in the current industry. As the most important algorithm link in the stereo data production technical process, the quality of the adjustment result directly determines the accuracy and quality of the subsequent stereo results. In the current adjustment process, an image space compensation parameter model is often used to eliminate the positioning error of each single-scene stereo image.

[0003] Commonly used image space compensation parameter models include a first-order image space compensation parameter model and a second-order image space compensation parameter model. Among them, the first-order image space compensation parameter model is applicable to situations with less distortion or simple systematic errors. Its advantages are small calculation amount, fast speed, and high stability. It can effectively eliminate the positioning error when processing stereo image pairs with simple distortion or small distortion amplitude, and is applicable to the production of the vast majority of current stereo image pairs. The disadvantage is that it cannot effectively and completely eliminate the positioning error when processing a small number of certain large-range complex distortion and complex error stereo image pairs; the second-order image space compensation parameter model is applicable to situations with more complex distortion, high accuracy requirements, large error range, and non-linear characteristics. Due to its more parameters, it can provide higher correction accuracy when processing and correcting complex distortion and errors, but the calculation is relatively complex, requiring more computing resources and time. At the same time, there is a risk of overfitting, resulting in the inability to stably and safely provide high-quality effects in actual applications, and more control point data support is required compared with the first-order image space compensation parameter model to ensure the reliability of the results.

[0004] Currently, the industry production mainly judges and selects which order of image space compensation parameter model to use through human-computer interaction in the regional network adjustment link for the production of all stereo image pairs. The problem with this approach is that it is not flexible enough. When multiple stereo image pairs are jointly used to produce a certain area, and if the first-order image space compensation parameter model is used for all of them, then some cross-track or multi-source stereo images cannot effectively and completely eliminate the positioning error. If the second-order image space compensation parameter model is used for all of them, then more computing resources and time are required. At the same time, there is a risk of overfitting for some stereo image pairs with good shooting conditions; currently, there is an urgent need for a method that can automatically adjust the image space compensation parameter model according to the different distortion errors of stereo image pairs. Summary of the Invention

[0005] Aiming at the problem that the existing technology cannot balance the cube image pair and needs to effectively eliminate the positioning error and prevent overfitting, the present invention discloses an adaptive adjustment method for the image space compensation parameter model of satellite stereo data, which relates to the technical field of satellite image processing, realizes the adaptive adjustment of the order of the image space compensation parameters for each stereo image pair, can generate high-precision stereo results, and can be adaptively adjusted without human and time costs.

[0006] The method includes the following steps:

[0007] S1. Obtain and preprocess the dataset Img_Data to obtain the preprocessed dataset Img_Paris_Data;

[0008] S2. Perform feature matching on all stereo image pairs in the preprocessed dataset to obtain the homologous point dataset Tie_points_data;

[0009] S3. Set a set of first-order image space compensation parameter models for each stereo image in each stereo image pair in the preprocessed dataset;

[0010] S4. Set the error threshold Mse_one_yz in the first-order image point residual plane and calculate the block adjustment result Mse_one of the first-order image space compensation parameters for each stereo image pair;

[0011] S5. If Mse_one < Mse_one_yz, the adjustment accuracy of the first-order image space compensation parameter model of this stereo image pair meets the requirements, execute step S6, otherwise use the second-order image space compensation parameter model and execute step S7;

[0012] S6. Set the epipolar error threshold EPI_error_value and calculate the epipolar error MSE_epi_one of the first-order image space compensation parameters for each stereo image pair that meets the adjustment accuracy requirements of the first-order image space compensation parameter model. If MSE_epi_one < EPI_error_value, the first-order image space compensation parameter model can eliminate the positioning error of this stereo image pair, execute step S11, otherwise use the second-order image space compensation parameter model and execute step S7;

[0013] S7. Set a set of second-order image space compensation parameter models for each stereo image in the stereo image pair that uses the second-order image space compensation parameter model;

[0014] S8. Set the error threshold Mse_two_yz in the second-order image point residual plane and calculate the block adjustment result Mse_two of the second-order image space compensation parameters for each stereo image pair;

[0015] S9. If Mse_two < Mse_two_yz, the adjustment accuracy of the second-order image-space compensation parameter model for this stereo image pair meets the requirements, and step S10 is executed; otherwise, the stereo image pair is replaced and step S11 is executed.

[0016] S10. Calculate the epipolar error MSE_epi_two of the second-order image-space compensation parameter model for each stereo image pair that meets the adjustment accuracy requirements of the second-order image-space compensation parameter model. If MSE_epi_two < EPI_error_value, the second-order image-space compensation parameter model can eliminate the positioning error of this stereo image pair, and step S11 is executed; otherwise, the stereo image pair is replaced and step S11 is executed.

[0017] S11. Mark the stereo image pairs for which the positioning error can be eliminated using the first-order or second-order compensation parameter model as Mark_usful, and mark the stereo image pairs that need to be replaced as Mark_unusful.

[0018] S12. Perform stereo data replacement on the stereo image pairs marked as Mark_unusful to obtain the replaced stereo image pairs.

[0019] S13. Repeat steps S2 to S12 for the replaced stereo image pairs and the stereo image pairs marked as Mark_usful to perform adaptive adjustment of the order of the image-space compensation parameter model until there are no stereo image pairs that need to be replaced, and complete the adaptive adjustment of the order of the image-space compensation parameter model.

[0020] Furthermore, the preprocessing is specifically as follows: Available images are sequentially selected from Img_Data in the order of same-orbit, different-orbit, and multi-source to construct joint coverage stereo image pairs for the target area to be stereoprocessed. When the selected series of stereo image pairs completely cover the target area, the series of stereo image pairs is Img_Paris_Data.

[0021] Furthermore, each stereo image pair in Img_Paris_Data includes a main image img_main and a secondary image img_sec; the feature matching is specifically as follows: On img_main, the reference object-space points (X i , Y i ) of the main image are calculated using the RPC parameters of the main image and the auxiliary DEM, and then the registration points (r i ′, c i ′) of each reference object-space point are calculated using the RPC parameters of the secondary image and the auxiliary DEM. For each (X i , Y i ) and (r i ′, c iCentered on ′), image blocks of m×n pixels are taken on the main image and the secondary image respectively. Extreme feature points of the main image and the secondary image are extracted on the two image blocks, and the extreme feature points of the main image and the secondary image are subjected to SIFT feature matching to extract all homologous points of the stereo pair, obtaining Tie_points_data.

[0022] Further, the specific method for calculating the first-order image-space compensation parameter block adjustment result Mse_one of each stereo pair is as follows: Based on Tie_points_data, control information, and the first-order image-space compensation parameter model, a first-order image-space compensation parameter block adjustment model is constructed, and the Ceres library is used to perform least-squares solution with the minimum image point residual as the convergence direction to obtain the first-order image-space compensation parameter values and the corresponding homologous point residual values. If there are gross error points in the homologous point residual values, after removing the gross error points, the least-squares solution is performed again, and the loop is repeated until there are no gross error points in the homologous point residual values. The homologous point residual values without gross error points and the corresponding first-order image-space compensation parameter values form Mse_one.

[0023] Further, the specific method for calculating the first-order image-space compensation parameter epipolar error MSE_epi_one of each stereo pair that meets the adjustment accuracy requirements of the first-order image-space compensation parameter model is as follows: The epipolar resampling method based on the unified elevation plane is used for the Tie_points_epi_data dataset to obtain the up and down differences between the homologous points of the stereo pair and the epipolar image coordinates; the Tie_points_epi_data dataset is specifically: a set of homologous points evenly distributed extracted from Tie_points_data.

[0024] Further, the method for calculating the second-order image-space compensation parameter block adjustment result Mse_two of each stereo pair is the same as the method for calculating the first-order image-space compensation parameter block adjustment result Mse_one of each stereo pair.

[0025] Further, the method for calculating the second-order image-space compensation parameter epipolar error MSE_epi_two of each stereo pair that meets the adjustment accuracy requirements of the second-order image-space compensation parameter model is the same as the method for calculating the first-order image-space compensation parameter epipolar error MSE_epi_one of each stereo pair that meets the adjustment accuracy requirements of the first-order image-space compensation parameter model.

[0026] Further, the values of m and n are both odd numbers.

[0027] The beneficial effects of the present invention are:

[0028] In the present invention, while calculating the adjustment residuals of a stereo image pair, the corresponding epipolar image coordinates of homologous points are calculated to obtain the vertical parallax of the homologous points, thereby realizing the adaptive adjustment of the order of the image-space compensation parameter model for each stereo image pair, achieving the adaptive hybrid use of the first-order and second-order image-space compensation parameter models. Compared with the conventional method, it has the characteristics of small computational amount, high speed and high stability of the first-order image-space compensation parameter model, and also has the advantages of the second-order image-space compensation parameter model that can be applied to the situation with complex distortion, high precision requirements, large error range and non-linear characteristics, ensuring high precision and high efficiency and stability in the production process. Description of the Drawings

[0029] Figure 1 It is a flowchart of the method described in the embodiment of the present invention;

[0030] Figure 2 It is a diagram showing the vector coverage of the selected stereo image pairs in the main urban area of Changchun in the embodiment of the present invention;

[0031] Figure 3 It is a diagram showing the image coverage of the selected stereo image pairs in the main urban area of Changchun in the embodiment of the present invention;

[0032] Figure 4 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date1 in the embodiment of the present invention;

[0033] Figure 5 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date2 in the embodiment of the present invention;

[0034] Figure 6 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date3 in the embodiment of the present invention;

[0035] Figure 7 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date4 in the embodiment of the present invention;

[0036] Figure 8 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date5 in the embodiment of the present invention;

[0037] Figure 9 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date6 in the embodiment of the present invention;

[0038] Figure 10 It is a schematic diagram of the stereo epipolar image pair produced by the first-order image-space compensation parameter model of date7 in the embodiment of the present invention;

[0039] Figure 11Schematic diagram of the stereo epipolar image pair produced by the date8 first-order image space compensation parameter model in the embodiment of the present invention;

[0040] Figure 12 Schematic diagram of the stereo epipolar image pair produced by the date1 second-order image space compensation parameter model in the embodiment of the present invention;

[0041] Figure 13 Schematic diagram of the stereo epipolar image pair produced by the date2 second-order image space compensation parameter model in the embodiment of the present invention;

[0042] Figure 14 Schematic diagram of the stereo epipolar image pair produced by the date3 second-order image space compensation parameter model in the embodiment of the present invention;

[0043] Figure 15 Schematic diagram of the stereo epipolar image pair produced by the date4 second-order image space compensation parameter model in the embodiment of the present invention;

[0044] Figure 16 Schematic diagram of the stereo epipolar image pair produced by the date5 second-order image space compensation parameter model in the embodiment of the present invention;

[0045] Figure 17 Schematic diagram of the stereo epipolar image pair produced by the date6 second-order image space compensation parameter model in the embodiment of the present invention;

[0046] Figure 18 Schematic diagram of the stereo epipolar image pair produced by the date7 second-order image space compensation parameter model in the embodiment of the present invention;

[0047] Figure 19 Schematic diagram of the stereo epipolar image pair produced by the date8 second-order image space compensation parameter model in the embodiment of the present invention;

[0048] Figure 20 DSM result map of the main urban area of Changchun in the embodiment of the present invention;

[0049] Figure 21 Comparison diagram of the stereo imaging effects of the image space compensation parameter models selected by the human-computer interaction judgment method and the adaptive adjustment method described in the present invention in the embodiment of the present invention. Detailed implementation manners

[0050] Next, the technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] As Figure 1As shown in the figure, this embodiment provides a method for adaptively adjusting the compensation parameter model of satellite stereo data image space, and the method includes the following steps:

[0052] S1. Obtain and preprocess the dataset Img_Data to obtain the preprocessed dataset Img_Paris_Data.

[0053] Determine the stereo data according to the production requirements: Taking the main urban area of Changchun as an example, set the target area to be stereoscopically produced as S_Area, analyze the production requirements for the target area according to the production requirements, and extract information parameters such as the coverage range, image shooting time requirement (Time_win), image resolution requirement (GSD), and image data source.

[0054] According to the requirements such as the coverage range, shooting time, and image resolution, combined with the image data source information, determine the joint coverage stereo image dataset Img_Data that can be used to cover the target area S_Area.

[0055] Select available images from the dataset Img_Data to construct a joint coverage stereo image pair for the target area S_Area. The selection priority is same-orbit - different-orbit - multi-source. Finally, the target area S_Area is completely covered by a series of stereo image pairs (it can be covered by a single stereo image pair or jointly covered by multiple stereo image pairs). Let the selected series of stereo image pair datasets be Img_Pairs_Data. Taking the stereo data of the Jilin-1 satellite constellation as an example, the vector coverage distribution of the stereo image pairs is as Figure 2 As shown, in this embodiment, there are a total of 8 pairs of stereo image pairs in Img_Pairs_Data, including: date1, date2, date3, date4, date5, date6, date7, and date8, and 16 satellite images; the image coverage of the selected stereo image pairs (mostly different-orbit multi-source stereo image pairs) in the main urban area of Changchun is as Figure 3 As shown, in this embodiment, panchromatic satellite images with a resolution better than 1m are selected to construct stereo image pairs; the 8 pairs of stereo image pairs are all taken on October 19, 2022 and October 20, 2022, and each stereo image pair belongs to different-orbit stereo image pair data.

[0056] S2. Perform feature matching on all stereo image pairs of the preprocessed dataset to obtain the homologous point dataset Tie_points_data.

[0057] The two stereo images of each stereo image pair in the stereo image pair dataset Img_Pairs_Data have a large overlap, and the overlap area is the collection range area of homologous points; let one of the stereo images in a single stereo image pair be the main image img_main, and the other stereo image be the secondary image img_sec.

[0058] Select N pixel points evenly distributed on the main image img_main as the image reference points (r i , c i ). Using the RPC parameters of the main image and the publicly available auxiliary DEM, solve the object coordinates corresponding to the reference points in the main image, which are the reference object points (X i , Y i ). The projection expression is

[0059]

[0060] where p′ q represents the inverse solution form of the RPC polynomial; Z is the iterative elevation value of the object coordinates provided by the auxiliary DEM, i = [1, 2, …, N], q = [1, 2, 3, 4].

[0061] Using the RPC parameters of the secondary image and the auxiliary DEM, solve the image point coordinates corresponding to each reference object point on the search image, denoted as the registration points (r′ i , c′ i ). The projection expression is:

[0062]

[0063] In the formula, X i , Y i , Z i are the ground coordinates corresponding to the matching point pairs, and the specific form is:

[0064] p q (X, Y, Z) = a1 + a2X + a3Y + a4Z + a5XY + a6XZ + a7YZ + a8X 2 + a9Y 2 + a 10 Z 2 + a 11 YXZ + a 12 X 3 + a 13 XY 2 + a 14 XZ 2 + a 15 X 2 Y + a 16 Y 3 + a 17 YZ 2 + a 18 X 2 Z + a 19 Y 2 Z + a 20 Z 3

[0065] where pq Represents the positive solution form of the RPC polynomial, a1 - a 20 The parameter values given for the RPC parameters, X, Y, and Z are the ground point coordinates,

[0066] For each (X i , Y i ), and (r i ′, c i ′) as the center, an m×n (both m and n are odd) image block is taken on the main image and the secondary image respectively, and extreme feature points are extracted from the two image blocks. Perform SIFT feature matching on the extreme feature points extracted from the main image and the secondary image, obtain high-precision matching homologous points according to the matching similarity, save all the homologous points extracted from the stereo pair, and obtain the dataset Tie_points_data.

[0067] S3. Set a set of first-order image space compensation parameter models for each stereo image in each stereo pair in the preprocessed dataset.

[0068] Set a set of first-order image space compensation parameter models for each stereo image in each stereo pair in the preprocessed dataset to eliminate the positioning error of the stereo image. The specific form is:

[0069]

[0070] Among them, (l′, s′) is the correct row and column values corresponding to a certain ground point T(X, Y, Z) on the satellite stereo image, (l, s) is the row and column values on the satellite stereo image calculated by the satellite stereo image RPC (Rational Polynomial Coefficients) parameters of point T, a0, a l , a s , b0, b l , b s are the first-order image space compensation parameter models corresponding to each satellite stereo image.

[0071] S4. Set the error threshold Mse_one_yz in the first-order image point residual plane and calculate the first-order image space compensation parameter block adjustment result Mse_one of each stereo pair.

[0072] In this embodiment, Mse_one_yz is set to 1 pixel.

[0073] Jointly adjust Tie_points_data, control information, and the first-order image space compensation parameter model. The adjustment model is as follows:

[0074]

[0075] Among them, Xi , Y i , Z i are the ground coordinates corresponding to the matching point pairs, R s , R0, C s , C0 are the normalization parameters in the RPC parameters of the video image, and the specific form is:

[0076] p q (X, Y, Z) = a1 + a2X + a3Y + a4Z + a5XY + a6XZ + a7YZ + a8X 2 + a9Y 2 + a 10 Z 2 + a 11 YXZ + a 12 X 3 + a 13 XY 2 + a 14 XZ 2 + a 15 X 2 Y + a 16 Y 3 + a 17 YZ 2 + a 18 X 2 Z + a 19 Y 2 Z + a 20 Z 3

[0077] where a1 - a 20 are the corresponding parameter values in the RPC parameter file.

[0078] Use the Ceres library to solve the adjustment model by least squares. The solution is carried out with the minimum image point residual as the convergence direction to obtain the first-order image-side compensation parameter values and the corresponding homologous point residual values. Compare the homologous point residual values with the gross error threshold (Error_value). The gross error threshold is generally set to 3 times the value of the mean square error in the plane (Mse). Therefore, the homologous points corresponding to the residuals exceeding 3 times the mean square error in the plane are considered gross error values. If there are any, delete all the gross error points and then perform the least squares solution again. Repeat the cycle until there are no gross error points in the homologous point residual values. The homologous point residual values without gross error points and the corresponding first-order image-side compensation parameter values form Mse_one. The regional network adjustment result of the first-order image-side compensation parameter Mse_one is shown in Table 1. The plane error of date5 is greater than that of other stereo image pairs, but it also does not exceed the error threshold Mse_one_yz in the plane of the first-order image point residual; where pixel represents pixel.

[0079] Table 1

[0080]

[0081] S5. If Mse_one < Mse_one_yz, then the adjustment accuracy of the first-order image-space compensation parameter model for this stereo image pair meets the requirements, and the first-order image-space compensation parameter model can be used, and step S6 is executed; otherwise, the second-order image-space compensation parameter model is used instead, and step S7 is executed.

[0082] S6. Set the epipolar error threshold EPI_error_value, and calculate the first-order image-space compensation parameter epipolar error MSE_epi_one for each stereo image pair that meets the adjustment accuracy requirements of the first-order image-space compensation parameter model. If MSE_epi_one < EPI_error_value, then the first-order image-space compensation parameter model can eliminate the positioning error of this stereo image pair, and step S11 is executed; otherwise, the second-order image-space compensation parameter model is used instead, and step S7 is executed.

[0083] In this embodiment, the epipolar error threshold EPI_error_value is set to 0.6 pixels. If the adjustment accuracy is high enough, the Y values of the epipolar image coordinates of corresponding points should be the same.

[0084] Extract a number of (not less than 30) evenly distributed corresponding point sets from Tie_points_data to form the Tie_points_epi_data data set.

[0085] The calculation of the epipolar error MSE_epi_one of the first-order image space compensation parameter for each stereo pair that meets the adjustment accuracy requirements of the first-order image space compensation parameter model is as follows: For the Tie_points_epi_data dataset, the epipolar resampling method based on the unified elevation plane is adopted to obtain the vertical and horizontal differences between the corresponding points of the stereo pair and the epipolar image coordinates: Select the elevation normalization translation scale parameter H in the RPC parameters of the stereo images as the projection elevation plane. Take an image point A at the center position of the main image Img_main. Let D1 and D2 be two points on its projection ray near the projection elevation plane H, and their elevations are H+Hs and H-Hs (Hs is the elevation normalization scaling scale parameter in the RPC parameters). The longitude and latitude coordinates of D1 and D2 are obtained according to the inverse solution equation of the RPC model; then, the pixel coordinates of the image points D3 and D4 of D1 and D2 on the secondary image Img_fz are calculated according to the forward solution equation of the RPC model; according to the RPC parameters and the RPC inverse solution model of the secondary image Img_fz, the longitude and latitude coordinates D5 and D6 of D3 and D4 on the projection elevation plane H are calculated. At this time, the connection direction of the longitude and latitude coordinates of D5 and D6 is the approximate epipolar direction of the stereo pair, that is, the arrangement direction of the approximate epipolar of the stereo image pair on the projection elevation plane. The four boundaries of the epipolar image are determined according to the coverage range of the two images of the stereo pair on the elevation projection plane. Finally, the mapping relationship (lepi, sepi) = F(l, s) between the row and column coordinates of each point on the stereo image and the corresponding row and column coordinates on the epipolar image is determined, where l and s are the row and column coordinates of a point Tone on the stereo image, and lepi and sepi are the row and column coordinates of Tone on the corresponding epipolar image.

[0086] 8 pairs of stereo epipolar image pairs produced by the first-order image space compensation parameter model of the stereo pair, as Figures 4 - 11 shown; The epipolar errors of the first-order image space compensation parameter model of 8 pairs of stereo pairs are shown in Table 2. Corresponding to the results of the first-order image space compensation parameter block adjustment, the epipolar error of the first-order image space compensation parameter of data5 is also greater than that of other stereo pairs, and the epipolar error of the first-order image space compensation parameter of date5 exceeds the epipolar error threshold EPI_error_value. It is necessary to use the second-order image space compensation parameter model. The other stereo data all meet the first-order epipolar error requirements, and the first-order image space compensation parameter model can be used;

[0087] Table 2

[0088]

[0089] S7. Set a set of second-order image space compensation parameter models for each stereo image in the stereo pairs that will use the second-order image space compensation parameter model.

[0090] Set a set of second-order image-space compensation parameters for each stereo image in each stereo image pair to eliminate the positioning error of the stereo image. The specific form is as follows:

[0091] l′ = l + a0 + a l *l + a s *s + a ll *l 2 + a ss *s 2 + a ls *l*s

[0092] s′ = s + b0 + b l *l + b s *s + b ll *l 2 + b ss *s 2+ b ls *l*s

[0093] Among them, (l′, s′) is the correct row and column values corresponding to a certain point T(X, Y, Z) on the ground in the satellite stereo image, and (l, s) is the row and column values on the satellite stereo image obtained by solving the satellite stereo image RPC (Rational Polynomial Coefficients) parameters of point T.

[0094] a0, a l , a s , a ll , a ss , a ls , b0, b l , b s , b ll , b ss and b ls are the second-order image-space compensation parameter models corresponding to each satellite stereo image;

[0095] Joint adjustment is performed on Tie_points_data, control information, and the second-order image-space compensation parameter model. The adjustment model is as follows:

[0096]

[0097] Among them, X i , Y i , Z i are the ground coordinates corresponding to the matching point pairs, and r s , R0, C s , C0 are the normalization parameters in the video image RPC parameters.

[0098] S8. Set the error threshold Mse_two_yz in the second-order image point residual plane, and calculate the regional network adjustment result Mse_two of the second-order image space compensation parameters for each stereo image pair;

[0099] In this embodiment, Mse_two_yz is set to 1 pixel.

[0100] The method for calculating the regional network adjustment result Mse_two of the second-order image space compensation parameters for each stereo image pair is the same as the method for calculating the regional network adjustment result Mse_one of the first-order image space compensation parameters for each stereo image pair.

[0101] The regional network adjustment result Mse_two of the second-order image space compensation parameters is shown in Table 3. The regional network adjustment results of the second-order image space compensation parameters for the 8 cube data are all less than Mse_two_yz, and the errors are all very small;

[0102] Table 3

[0103]

[0104] S9. If Mse_two < Mse_two_yz, then the adjustment accuracy of the second-order image space compensation parameter model for this stereo image pair meets the requirements, and step S10 is executed; otherwise, the stereo image pair is replaced and step S11 is executed.

[0105] S10. Calculate the second-order image space compensation parameter model epipolar error MSE_epi_two for each stereo image pair that meets the adjustment accuracy requirements of the second-order image space compensation parameter model. If MSE_epi_two < EPI_error_value, then the second-order image space compensation parameter model can eliminate the positioning error of this stereo image pair, and step S11 is executed; otherwise, the stereo image pair is replaced and step S11 is executed.

[0106] The method for calculating the second-order image space compensation parameter model epipolar error MSE_epi_two for each stereo image pair that meets the adjustment accuracy requirements of the second-order image space compensation parameter model is the same as the method for calculating the first-order image space compensation parameter epipolar error MSE_epi_one for each stereo image pair that meets the adjustment accuracy requirements of the first-order image space compensation parameter model.

[0107] The stereo epipolar image pairs produced by the second-order image space compensation parameter models of 8 pairs of stereo image pairs are as Figures 12 - 19 shown; the second-order image space compensation parameter model epipolar errors of 8 pairs of stereo image pairs are shown in Table 4. It can be seen that after date5 is adaptively adjusted to the second-order image space compensation parameter model, the second-order image space compensation parameter model epipolar error is greatly reduced and has been less than the epipolar error threshold EPI_error_value. Therefore, the second-order image space compensation parameter model can already completely eliminate the positioning error of date5.

[0108] Table 4

[0109]

[0110] S11. Identify the stereo image pairs that can eliminate the positioning error using the first-order or second-order compensation parameter model as Mark_usful, and identify the stereo image pairs that need to replace the stereo image pairs as Mark_unusful;

[0111] S12. Perform stereo data replacement on the stereo image pairs marked as Mark_unusful to obtain the replaced stereo image pairs;

[0112] S13. Repeat steps S2 to S12 for the replaced stereo image pairs and the stereo image pairs marked as Mark_usful to perform adaptive adjustment of the order of the image-side compensation parameter model until there are no stereo image pairs that need to replace the stereo image pairs, and complete the adaptive adjustment of the order of the image-side compensation parameter model.

[0113] The identification results of each stereo image pair and the corresponding adaptive adjustment results of the image-side compensation parameter model are shown in Table 5. It can be seen that date5 is adaptively adjusted to the second-order image-side compensation parameter model, and other data are adaptively adjusted to the first-order image-side compensation parameter model, and each stereo image pair can be used for subsequent production; the DSM result maps of the main urban area of Changchun produced by each stereo image pair are as Figure 20 shown.

[0114] Table 5

[0115]

[0116] The current industry production mainly judges and selects which order of image-side compensation parameter model to use through human-computer interaction in the regional network adjustment link, and the selected first-order or second-order image-side compensation parameter model will be used for the production of all stereo image pairs. In the human-computer interaction judgment method for date5, it should be judged to use the first-order image-side parameter model; as Figure 21 shown, it is a comparison chart of the stereo imaging effects of date5 applying the first-order image-side parameter model and the second-order image-side parameter model. It can be clearly seen that after date5 selects to use the second-order image-side parameter model through the adaptive adjustment method of the present invention, the stereo imaging effect is significantly improved, the houses are clear and not sticky, and it is closer to the real surface features.

Claims

1. A method for adaptively adjusting a satellite stereo data image compensation parameter model, characterized in that: The method includes the following steps: S1. Obtain and preprocess the dataset Img_Data to obtain the preprocessed dataset Img_Paris_Data; S2. Perform feature matching on all stereo image pairs in the preprocessed dataset to obtain the corresponding point dataset Tie_points_data; S3. Set a set of first-order image space compensation parameter models for each stereo image in each stereo image pair in the preprocessed dataset; S4. Set the error threshold Mse_one_yz in the first-order image point residual plane and calculate the first-order image space compensation parameter block adjustment result Mse_one for each stereo image pair; S5. If Mse_one < Mse_one_yz, the adjustment accuracy of the first-order image space compensation parameter model for this stereo image pair meets the requirements, and step S6 is executed. Otherwise, switch to the second-order image space compensation parameter model and execute step S7; S6. Set the epipolar error threshold EPI_error_value and calculate the first-order image space compensation parameter epipolar error MSE_epi_one for each stereo image pair that meets the adjustment accuracy requirements of the first-order image space compensation parameter model. If MSE_epi_one < EPI_error_value, the first-order image space compensation parameter model can eliminate the positioning error of this stereo image pair, and step S11 is executed. Otherwise, switch to the second-order image space compensation parameter model and execute step S7; S7. Set a set of second-order image space compensation parameter models for each stereo image in the stereo image pair that switches to the second-order image space compensation parameter model; S8. Set the error threshold Mse_two_yz in the second-order image point residual plane and calculate the second-order image space compensation parameter block adjustment result Mse_two for each stereo image pair; S9. If Mse_two < Mse_two_yz, the adjustment accuracy of the second-order image space compensation parameter model for this stereo image pair meets the requirements, and step S10 is executed. Otherwise, replace the stereo image pair and execute step S11; S10. Calculate the second-order image space compensation parameter epipolar error MSE_epi_two for each stereo image pair that meets the adjustment accuracy requirements of the second-order image space compensation parameter model. If MSE_epi_two < EPI_error_value, the second-order image space compensation parameter model can eliminate the positioning error of this stereo image pair, and step S11 is executed. Otherwise, replace the stereo image pair and execute step S11; S11. Mark the stereo image pairs whose positioning errors can be eliminated using the first-order or second-order compensation parameter models as Mark_usful, and mark the stereo image pairs that need to be replaced as Mark_unusful; S12. Perform stereo data replacement on the stereo image pairs marked as Mark_unusful to obtain the replaced stereo image pairs; S13. Repeat steps S2 to S12 for the replaced stereo image pairs and the stereo image pairs marked as Mark_usful to perform adaptive adjustment of the order of the image space compensation parameter model until there are no stereo image pairs that need to be replaced, and complete the adaptive adjustment of the order of the image space compensation parameter model.

2. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 1, characterized in that: The preprocessing is specifically as follows: select available images from Img_Data in the order of same track, different track and multiple sources to construct stereo image pairs that jointly cover the target area to be stereo produced. When the selected series of stereo image pairs completely covers the target area, the series of stereo image pairs is Img_Paris_Data.

3. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 2, characterized in that: Each stereo image pair in Img_Paris_Data includes a main image img_main and a secondary image img_sec; the feature matching is specifically as follows: on img_main, the RPC parameters of the main image and the auxiliary DEM are used to solve the reference object point (X i ,Y i ), and then use the RPC parameters of the secondary image and the auxiliary DEM to solve the registration point (r′) of each reference object point i ,c′ i ), with each (X i ,Y i ) and (r′ i ,c′ i ) is taken as the center, and image blocks of m×n pixels are taken on the main image and the secondary image respectively. The extreme feature points of the main image and the secondary image are extracted on the two image blocks respectively, and the extreme feature points of the main image and the secondary image are matched by SIFT features to extract the same-name points of all stereo image pairs and obtain Tie_points_data.

4. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 3, characterized in that: The calculation of the first-order image compensation parameter block adjustment result Mse_one of each stereo image pair is specifically as follows: a first-order image compensation parameter block adjustment model is constructed according to Tie_points_data, control information and the first-order image compensation parameter model, and a least squares solution is performed using the Ceres library with the minimum image point residual as the convergence direction to obtain the first-order image compensation parameter value and the corresponding homonymous point residual value; if there are gross error points in the homonymous point residual value, the gross error points are removed and the least squares solution is performed again, and the cycle is repeated until there are no gross error points in the homonymous point residual value, and the homonymous point residual value without gross error points and the corresponding first-order image compensation parameter value constitute Mse_one.

5. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 1, characterized in that: The calculation of the first-order image compensation parameter epipolar error MSE_epi_one of each stereo pair that meets the adjustment accuracy requirements of the first-order image compensation parameter model is specifically as follows: an epipolar resampling method based on a unified elevation surface is used on the Tie_points_epi_data data set to obtain the upper and lower differences between the stereo pair's homonymous points and the epipolar image coordinates; the Tie_points_epi_data data set is specifically as follows: a number of evenly distributed homonymous point sets extracted from the Tie_points_data.

6. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 4, characterized in that: The method for calculating the block adjustment result Mse_two of the second-order image square compensation parameter of each stereo image pair is consistent with the method for calculating the block adjustment result Mse_one of the first-order image square compensation parameter of each stereo image pair.

7. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 5, characterized in that: The method for calculating the second-order image compensation parameter model epipolar line error MSE_epi_two of each stereo pair that meets the adjustment accuracy requirements of the second-order image compensation parameter model is consistent with the method for calculating the first-order image compensation parameter epipolar line error MSE_epi_one of each stereo pair that meets the adjustment accuracy requirements of the first-order image compensation parameter model.

8. The method for adaptively adjusting the satellite stereo data image compensation parameter model according to claim 3, characterized in that: The values ​​of m and n are both odd numbers.

Images