An optical satellite remote sensing image digital surface model extraction method and system

Abstract

The present invention discloses a method and system for extracting a digital surface model of optical satellite remote sensing images, including: obtaining the matching cost information of each image point by using the Census measure method for each image point; aggregating the matching costs based on the minimum cost path according to the matching cost information of each image point to obtain a set of candidate disparity values for each image point; selecting the candidate disparity value with the maximum support degree from the set of candidate disparity values of each image point as the final disparity value of each image point, and obtaining a disparity map according to the final disparity value of each image point; performing refinement processing on the disparity map to obtain a refined disparity map; based on the refined disparity map and the preset image RPC parameters, obtaining object point clouds through spatial resection, and obtaining a digital surface model of optical satellite remote sensing images by grid processing of the object point clouds. The present invention greatly reduces the computational redundancy and improves the processing efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing data processing, and particularly relates to a method and system for extracting a digital surface model of optical satellite remote sensing images. Background Art

[0002] Dense Matching is a core link in the extraction of the digital surface model (DSM) of optical satellite remote sensing images. The matching result and matching efficiency directly determine the quality and efficiency of the final operation result. Dense Matching generally can be divided into four parts: cost calculation, cost aggregation, disparity calculation, and disparity optimization. According to different cost aggregation strategies, it can be divided into local methods and (semi-)global methods. Compared with local methods, (semi-)global methods are more computationally complex but have higher reliability. Since the classic semi-global matching method (SGM) was proposed in 2008, it uses multi-path one-dimensional cost aggregation to approximately achieve two-dimensional cost aggregation, taking into account both matching accuracy and efficiency, and has been widely used in fields such as DSM extraction. Currently, it has also been adopted by leading remote sensing image processing software in the industry, such as Geomatica, Inpho, ERDAS IMAGING, RPC StereoProcessor (RSP), MicMac, and SURE.

[0003] In SGM, cost aggregation updates the matching cost of the current point according to the matching cost of neighboring points and the smoothing term, and realizes two-dimensional smoothing constraints through multiple one-dimensional direction cost aggregations. This strategy needs to aggregate the matching costs corresponding to all candidate disparity values of all pixels. Taking single-path cost aggregation as an example, its computational complexity is n×D, where n is the number of pixels in a single path and D is the disparity range. The cost aggregation strategy in SGM treats all disparity values within the disparity range of a single pixel equally, resulting in a large amount of computational redundancy. Especially when the disparity range is large, it seriously affects the matching efficiency. Summary of the Invention

[0004] The technical problem solved by the present invention is: overcoming the deficiencies of the prior art, providing a method and system for extracting a digital surface model of optical satellite remote sensing images, and solving the problem that a large amount of computational redundancy in semi-global dense matching leads to a long time-consuming dense matching and low efficiency of DSM extraction operations.

[0005] The object of the present invention is achieved by the following technical solutions: An optical satellite remote sensing image digital surface model extraction method, comprising: obtaining the matching cost information of each pixel by using the Census measure method for each pixel; aggregating the matching costs based on the minimum cost path according to the matching cost information of each pixel to obtain a set of candidate disparity values for each pixel; selecting the candidate disparity value with the maximum support degree from the set of candidate disparity values of each pixel as the final disparity value of each pixel, and obtaining a disparity map according to the final disparity value of each pixel; refining the disparity map to obtain a refined disparity map; based on the refined disparity map and preset image rational polynomial parameters, obtaining object space point clouds through spatial resection, and obtaining an optical satellite remote sensing image digital surface model by grid processing of the object space point clouds.

[0006] In the above optical satellite remote sensing image digital surface model extraction method, the Census measure method is: constructing a feature space with the binary information of the size relationship between the corresponding pixels in the two templates and their central pixels, thereby obtaining two Census sequences, and then using the Hamming distance between the Census sequences as the similarity measure.

[0007] In the above optical satellite remote sensing image digital surface model extraction method, if the gray value of a certain pixel is greater than or equal to the gray value of the central pixel, the value of this pixel in the feature space is 1, otherwise it is 0.

[0008] In the above optical satellite remote sensing image digital surface model extraction method, aggregating the matching costs based on the minimum cost path according to the matching cost information of each pixel to obtain a set of candidate disparity values for each pixel includes: for each pixel, by determining the minimum cost path in the 8-neighborhood direction of each pixel, obtaining the candidate disparity values in the 8-neighborhood direction of each pixel, and forming a set of candidate disparity values for each pixel.

[0009] In the above optical satellite remote sensing image digital surface model extraction method, the minimum cost path is obtained by the following formula:

[0010]

[0011] where L is the minimum cost path, i is a point on the path, n is the number of points on the path, j is the connection line between two adjacent pixels, c i is the matching cost of the selected disparity value of pixel i, c j is the smooth value when two adjacent pixels select disparity values.

[0012] In the above method for extracting the digital surface model of optical satellite remote sensing images, selecting the candidate disparity value with the maximum support degree from the candidate disparity value set of each image point as the final disparity value of each image point includes: first, counting the support degree of each disparity candidate value in the candidate disparity value set of each image point, and then adopting the WTA strategy to select the candidate disparity value with the maximum support degree as the final disparity value of each image point.

[0013] In the above method for extracting the digital surface model of optical satellite remote sensing images, the support degree of each disparity candidate value is: the reciprocal of the sum of the absolute values of the differences between each candidate disparity value and other candidate disparity values is used as the support degree of each candidate disparity value.

[0014] In the above method for extracting the digital surface model of optical satellite remote sensing images, the refinement processing includes: median filtering, left-right consistency detection, outlier detection and interpolation.

[0015] In the above method for extracting the digital surface model of optical satellite remote sensing images, the mathematical model of spatial resection is obtained through the following formula:

[0016]

[0017]

[0018] where (l n , s n ) is the normalized image plane coordinate of the image point, and (U, V, W) is the normalized object space coordinate of the object point;

[0019] Num L (U, V, W) = a1 + a2V + a3U + a4W + a5VU + a6VW + a7UW + a s V 2 + a9U 2 + a 10 W 2 + a 11 VUW + a 12 V 3 + a 13 VU 2 + a 14 VW 2 + a 15 V 2 U + a 16 U 3 + a 17 UW 2 + a 18 V 2 W + a 19 U 2 W + a 20 W 3 ;

[0020] Den L (U, V, W) = b1 + b2V + b3U + b4W + b5VU + b6VW + b7UW + b8V 2 + b9U 2 + b 10 W 2 + b 11 VUW + b 12 V 3 + b 13 VU 2 + b 14 VW 2 + b 15 V 2 U + b 16 U 3 + b 17 UW 2 + b 18 V 2 W + b 19 U 2 W + b 20 W 3 ;

[0021] Num S (U, V, W) = c1 + c2V + c3U + c4W + c5VU + c6VW + c7UW + c s V 2 + c9U 2 + c 10 W 2 + c 11 VUW + c 12 V 3 + c 13 VU 2 + c 14 VW 2 + c 15 V 2 U + c 16 U 3 + c 17 UW 2 + c 18 V 2 W + c 19 U 2 W + c 20 W 3 ;

[0022] Den S (U, V, W) = d1 + d2V + d3U + d4W + d5VU + d6VW + d7UW + d8V 2 + d9U 2 + d 10 W 2 + d11 VUW + d 12 V 3 + d 13 VU 2 + d 14 VW 2 + d 15 V 2 U + d 16 U 3 + d 17 UW 2 + d 18 V 2 W + d 19 U 2 W + d 20 W 3 ;

[0023] Among them, Num L (U, V, W), Den L (U, V, W), Num S (U, V, W) and Den s (U, V, W) are all polynomials of the normalized object - space coordinates of object - space points. a q , b q , c q and d q are all image rational polynomial parameters. Among them, q = 1, 2,..., 20, and q is a subscript.

[0024] An optical satellite remote - sensing image digital surface model extraction system, comprising: a first module for obtaining the matching - cost information of each image point by using the Census measure method for each image point; a second module for aggregating the matching costs based on the minimum - cost path for the matching - cost information of each image point to obtain a set of candidate disparity values for each image point; a third module for selecting the candidate disparity value with the maximum support degree from the set of candidate disparity values of each image point as the final disparity value of each image point, and obtaining a disparity map according to the final disparity value of each image point; a fourth module for refining the disparity map to obtain a refined disparity map; a fifth module for obtaining an object - space point cloud through spatial resection based on the refined disparity map and preset image RPC parameters, and obtaining an optical satellite remote - sensing image digital surface model by grid - processing the object - space point cloud.

[0025] The present invention has the following beneficial effects compared with the prior art:

[0026] The present invention makes the computational complexity of the disparity map no longer related to the disparity range, greatly reducing the computational redundancy and improving the processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered a limitation of the present invention. Moreover, throughout the drawings, the same reference numerals are used to denote the same components. In the drawings:

[0028] Figure 1 is a schematic flowchart of a method for extracting a digital surface model from an optical satellite remote sensing image provided by an embodiment of the present invention;

[0029] Figure 2 is a schematic diagram of the Census matching cost provided by an embodiment of the present invention;

[0030] Figure 3 is a schematic diagram of the minimum cost path provided by an embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of point cloud gridding provided by an embodiment of the present invention. Detailed Embodiments

[0032] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0033] Figure 1 is a schematic flowchart of a method for extracting a digital surface model from an optical satellite remote sensing image provided by an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0034] Step 1: Obtain the matching cost information of each image point by using the Census measure method for each image point.

[0035] Use the Census measure to calculate the matching cost corresponding to each disparity value within the disparity range for each image point, and obtain the matching costs of all image points within all disparity ranges.

[0036] Step 2: Aggregate the matching costs based on the minimum cost path according to the matching cost information of each image point to obtain the set of candidate disparity values for each image point.

[0037] For a single pixel, by separately determining the minimum cost paths in its 8-neighborhood directions, the candidate disparity values in its 8-neighborhood directions are obtained, forming the candidate disparity value set for this pixel. After processing each pixel one by one, the candidate disparity value sets for all pixels are obtained.

[0038] Step 3: Select the candidate disparity value with the maximum support degree from the candidate disparity value set of each pixel as the final disparity value of each pixel, and obtain the disparity map according to the final disparity value of each pixel.

[0039] For the candidate disparity value set obtained after cost aggregation for a single pixel, the cumulative value of the absolute values of the differences between a single disparity value and the other disparity values is used as the support degree of this disparity value. Then, the "Winner Take All" (WTA) strategy is adopted to select the candidate disparity value with the maximum support degree as the final disparity value of this pixel. After processing each pixel one by one, the disparity map is obtained.

[0040] Step 4: Refine the disparity map to obtain the refined disparity map.

[0041] Adopt strategies such as median filtering to refine the noise points in the disparity map to obtain the refined disparity map.

[0042] Step 5: Based on the refined disparity map and the preset Rational Polynomial Coefficients (RPC) of the image, the object space point cloud is obtained through spatial resection, and the digital surface model of the optical satellite remote sensing image is obtained by grid processing of the object space point cloud.

[0043] Specifically, matching cost calculation:

[0044] The matching cost is used to measure the similarity between two points in the corresponding feature space and is calculated using the Census measure. As Figure 2 shown, Census constructs the feature space with the binary information of the magnitude relationship between the corresponding points in the two templates and their respective central points' gray values, thereby obtaining two Census sequences. Then, the Hamming distance between the Census sequences is used as the similarity measure. If the gray value of a certain point is greater than or equal to the gray value of the central point, the value of this point in the feature space is 1, otherwise it is 0.

[0045] Cost aggregation based on the minimum cost path:

[0046] The minimum cost path refers to the minimum sum of the cost values of the connections between points on a single aggregation path, as shown in the following formula:

[0047]

[0048] Among them, L is the minimum cost path, i is a point on the path, n is the number of points on the path, j is the connection between two adjacent points, and c i is the matching cost of the selected disparity value for point i, and c j is the smoothness value when the selected disparity values for two adjacent points are considered.

[0049] As Figure 3 shown, similar to the SGM cost aggregation strategy, for the starting point p of the MCP single path, the disparity value adopted is the disparity value d0 corresponding to its minimum matching cost. However, different from SGM that aggregates the matching costs corresponding to all disparity candidate values of each subsequent point on the current path, MCP only selects the minimum cost path from the four disparity candidate values d0, d0 + 1, d0 - 1, and d c of the subsequent point q, where d c is the disparity value corresponding to the minimum matching cost of point q. The path cost is the sum of the matching cost of the disparity value corresponding to point q and the smoothness value of the path between it and the determined disparity value d0 of point p. Taking the disparity value d0 of point q as an example, where the smoothness term is the same as that of SGM. Repeat the processing backward in sequence until all points on the current path are processed. It can be seen that in the single path cost aggregation, different from SGM that processes all disparity candidate values of a single pixel, only four disparity candidate values of a single pixel in MCP participate in the processing, and its computational complexity is n×4, where n is the number of pixels in a single path, which will effectively reduce the computational redundancy and improve the processing efficiency, especially when the disparity range D has a large value.

[0050] Disparity determination based on maximum support:

[0051] Different from SGM that obtains a new matching cost through cost aggregation, MCP obtains n disparity candidate values, where n is the number of paths, that is, each path provides a candidate value for the final disparity value of the pixel. At this time, the WTA cannot be directly used to determine the disparity. First, count the support of each disparity candidate value, and use the reciprocal of the sum of the absolute values of the differences between a single disparity candidate value and other disparity candidate values as its support, as shown in the following formula. Then, adopt the WTA strategy to select the disparity candidate value with the maximum support as the final disparity value.

[0052] d = argmax 1 / (∑|d i - d j |) i，j∈D ;

[0053] where, D is the set of disparity candidate values, and d is the final disparity value.

[0054] Disparity refinement:

[0055] Adopt strategies such as median filtering, left - right consistency detection, outlier detection, and interpolation to refine the disparity map.

[0056] DSM Generation:

[0057] Based on the disparity map obtained from dense matching and the RPC parameters of the image, DSM is generated through spatial resection and gridding. Spatial resection refers to the process of determining the object coordinates of homologous points by intersecting homologous rays based on the interior and exterior orientation elements of the image and the coordinates of homologous points after relative orientation. Its mathematical model is a rational function model.

[0058] Num L (U, V, W) = a1 + a2V + a3U + a4W + a5VU + a6VW + a7UW + a8V 2 + a9U 2 + a 10 W 2 + a 11 VUW + a 12 V 3 + a 13 VU 2 + a 14 VW 2 + a 15 V 2 U + a 16 U 3 + a 17 UW 2 + a 18 V 2 W + a 19 U 2 W + a 20 W 3

[0059] Den L (U, V, W) = b1 + b2V + b3U + b4W + b5VU + b6VW + b7UW + b8V 2 + b9U 2 + b 10 W 2 + b 11 VUW + b 12 V 3 + b 13 VU 2 + b 14 VW 2 + b 15 V 2 U + b 16 U 3 + b 17 UW 2 + b 18 V 2 W + b 19 U 2 W + b 20 W 3 Num S(U, V, W) = c1 + c2V + c3U + c4W + c5VU + c6VW + c7UW + c8V 2 + c9U 2 + c 10 W 2 + c 11 VUW + c 12 V 3 + c 13 VU 2 + c 14 VW 2 + c 15 V 2 U + c 16 U 3 + c 17 UW 2 + c 18 V 2 W + c 19 U 2 W + c 20 W 3

[0060] Den S (U, V, W) = d1 + d2V + d3U + d4W + d5VU + d6VW + d7UW + d8V 2 + d9U 2 + d 10 W 2 + d 11 VUW + d 12 V 3 + d 13 VU 2 + d 14 VW 2 + d 15 V 2 U + d 16 U 3 + d 17 UW 2 + d 18 V 2 W + d 19 U 2 W + d 20 W 3

[0061] Among them, a q , b q , c q and d q are all image rational polynomial parameters. Among them, q = 1, 2,..., 20, q is a subscript. Generally, b1 and d1 are both taken as 1.

[0062] The object space point cloud is obtained by forward intersection. However, since the object space point cloud is not regularly arranged, and the DSM is essentially a discrete representation of the object space by a series of regular grids, with a fixed elevation value adopted within each grid, gridification is a necessary process for generating DSM from the object space point cloud. As Figure 4 shown, within the local object space range, a 3*3 grid R is used for discrete representation, and the elevation value of each grid unit is determined by the elevation values of the object space points falling into it. Common elevation value determination strategies include interpolation methods for irregular triangular meshes, median methods, etc.

[0063] This embodiment also provides an optical satellite remote sensing image digital surface model extraction system, including: a first module for obtaining the matching cost information of each image point by using the Census measure method for each image point; a second module for aggregating the matching costs based on the minimum cost path according to the matching cost information of each image point to obtain a set of candidate disparity values for each image point; a third module for selecting the candidate disparity value with the maximum support degree in the set of candidate disparity values of each image point as the final disparity value of each image point, and obtaining a disparity map according to the final disparity value of each image point; a fourth module for performing refinement processing on the disparity map to obtain a refined disparity map; a fifth module for obtaining the object space point cloud through spatial forward intersection based on the refined disparity map and preset image RPC parameters, and obtaining the optical satellite remote sensing image digital surface model by gridifying the object space point cloud.

[0064] The present invention first transforms the aggregation cost of a single pixel within its disparity range in dense matching into a set of candidate disparity values through aggregation of matching costs based on the minimum cost path; then, by statistically calculating the support degree of the candidate disparity values of a single pixel and adopting the "Winner Take All" (WTA) strategy, selects the candidate disparity value with the maximum support degree as the final disparity value of this pixel. After processing each pixel one by one, a disparity map is obtained; this makes the computational complexity of the disparity map no longer related to the disparity range, greatly reducing the computational redundancy and improving the processing efficiency.

[0065] Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention all fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for extracting a digital surface model from optical satellite remote sensing images, characterized in that Including: Obtaining the matching cost information of each image point by using the Census measure method for each image point; Based on the matching cost information of each image point, aggregating the matching cost based on the minimum cost path to obtain the candidate disparity value set of each image point; Selecting the candidate disparity value with the maximum support degree from the candidate disparity value set of each image point as the final disparity value of each image point, and obtaining the disparity map according to the final disparity value of each image point; Performing refinement processing on the disparity map to obtain the refined disparity map; Based on the refined disparity map and the preset image rational polynomial parameters, obtaining the object space point cloud through spatial resection, and obtaining the digital surface model of the optical satellite remote sensing image by grid processing the object space point cloud; Aggregating the matching cost based on the minimum cost path according to the matching cost information of each image point to obtain the candidate disparity value set of each image point, including: For each image point, by determining the minimum cost path in the 8-neighborhood direction of each image point, obtaining the candidate disparity values in the 8-neighborhood direction of each image point, and forming the candidate disparity value set of each image point; The minimum cost path is obtained by the following formula: Among them, L is the minimum cost path, i is a point on the path, n is the number of points on the path, j is the connection line between two adjacent image points, and c i is the matching cost for selecting the disparity value for image point i, and c j is the smoothness value when selecting the disparity value for two adjacent image points; Similar to the SGM cost aggregation strategy, for the starting point p of the MCP single path, the disparity value d0 corresponding to its minimum matching cost is adopted for the disparity value. However, different from SGM that aggregates the matching costs corresponding to all disparity candidate values for each subsequent point on the current path, MCP only selects the minimum cost path from the four disparity candidate values of d0, d0 + 1, d0 - 1, and d of the subsequent point q, where d c is the disparity value corresponding to the minimum matching cost of point q; the path cost is the sum of the matching cost of the disparity value corresponding to point q and the smooth value of the path between it and the determined disparity value d0 of point p; repeat the processing backward in sequence until all points on the current path are processed. c ​ 2. The method for extracting a digital surface model of an optical satellite remote sensing image according to claim 1, wherein: The Census measure method is: constructing a feature space with the binary information of the size relationship between the corresponding image points in the two templates and their central image points, thereby obtaining two Census sequences, and then using the Hamming distance between the Census sequences as the similarity measure.

3. The method for extracting a digital surface model of an optical satellite remote sensing image according to claim 2, wherein: If the gray value of a certain image point is greater than or equal to the gray value of the central image point, the value of this image point in the feature space is 1, otherwise it is 0.

4. The method for extracting the digital surface model of optical satellite remote sensing images according to claim 1, wherein: Selecting the candidate disparity value with the maximum support degree from the candidate disparity value set of each image point as the final disparity value of each image point, including: First, counting the support degree of each disparity candidate value in the candidate disparity value set of each image point, and then adopting the WTA strategy to select the candidate disparity value with the maximum support degree as the final disparity value of each image point.

5. The method for extracting a digital surface model of an optical satellite remote sensing image according to claim 1, wherein: The support degree of each disparity candidate value is: the reciprocal of the sum of the absolute values of the differences between each candidate disparity value and other candidate disparity values is used as the support degree of each candidate disparity value.

6. The method for extracting a digital surface model of optical satellite remote sensing images according to claim 1, wherein: The refinement processing includes: median filtering, left-right consistency detection, outlier detection and interpolation.

7. The method for extracting a digital surface model of optical satellite remote sensing images according to claim 1, wherein: The mathematical model of spatial resection is obtained by the following formula: where (l n , s n ) are the normalized image coordinates of the image point, and (U, V, W) are the normalized object coordinates of the object point; Num L (U, V, W) = a1 + a2V + a3U + a4W + a5VU + a6VW + a7UW + a8V 2 + a9U 2 + a 10 W 2 + a 11 VUW + a 12 V 3 + a 13 VU 2 + a 14 VW 2 + a 15 V 2 U + a 16 U 3 + a 17 UW 2 + a 18 V 2 W + a 19 U 2 W + a 20 W 3 ; Den L (U, V, W) = b1 + b2V + b3U + b4W + b5VU + b6VW + b7UW + b8V 2 + b9U 2 + b 10 W 2 + b 11 VUW + b 12 V 3 + b 13 VU 2 + b 14 VW 2 + b 15 V 2 U + b 16 U 3 + b 17 UW 2 + b 18 V 2 W + b 19 U 2 W + b 20 W 3 ; Num S (U, V, W) = c1 + c2V + c3U + c4W + c5VU + c6VW + c7UW + c8V 2 + c9U 2 + c 10 W 2 + c 11 VUW + c 12 V 3 + c 13 VU 2 + c 14 VW 2 + c 15 V 2 U + c 16 U 3 + c 17 UW 2 + c 18 V 2 W + c 19 U 2 W + c 20 W 3 ; Den S (U, V, W) = d1 + d2V + d3U + d4W + d5VU + d6VW + d7UW + d8V 2 + d9U 2 + d 10 W 2 + d 11 VUW + d 12 V 3 + d 13 VU 2 + d 14 VW 2 + d 15 V 2 U + d 16 U 3 + d 17 UW 2 + d 18 V 2 W + d 19 U 2 W + d 20 W 3 ; Among them, Num L (U, V, W), Den L (U, V, W), Num S (U, V, W) and Den S (U, V, W) are all polynomials of the normalized object - space coordinates of object - space points, a q , b q , c q and d q are all image rational polynomial parameters, where q = 1, 2, …, 20, and q is a subscript.

8. An optical satellite remote sensing image digital surface model extraction system, the system executes the method according to any one of claims 1-7, characterized in that Including: The first module is used to obtain the matching cost information of each image point by using the Census measure method for each image point; The second module is used to aggregate the matching cost based on the minimum cost path according to the matching cost information of each image point to obtain the candidate disparity value set of each image point; The third module is used to select the candidate disparity value with the maximum support degree from the candidate disparity value set of each image point as the final disparity value of each image point, and obtain the disparity map according to the final disparity value of each image point; The fourth module is used to perform refinement processing on the disparity map to obtain the refined disparity map; The fifth module is used to obtain the object space point cloud through spatial resection based on the refined disparity map and the preset image RPC parameters, and obtain the digital surface model of the optical satellite remote sensing image by grid processing the object space point cloud.

Images