A morphological template conformal filtering method in three-dimensional terrain reconstruction

Abstract

This invention proposes a morphological template-based conformal filtering method for 3D terrain reconstruction, belonging to the technical field of 3D terrain reconstruction. It solves the problem that current surface filtering methods cannot accurately identify and preserve the edges and corners of terrain features, resulting in the inability to accurately restore the true contours of features. First, a terrain representation map of the target feature is obtained. Then, multiple morphological templates are designed based on the regular shapes of the corners. These templates are used to detect the positions of the edges and corners to be processed in the terrain representation map. Next, it is determined whether the pixel code of the terrain representation map matches any one of the morphological templates. If it does, a conformal weighted filtering process is applied to the current pixel code; otherwise, a median filtering process is applied. Finally, the filtered terrain representation map is post-processed, and the 3D model of the target feature is reconstructed based on the processing results. This method achieves accurate identification and preservation of feature edges and corners, and accurate restoration of the true contours of features.

Description

Technical Field

[0001] This invention relates to the technical field of three-dimensional terrain reconstruction, and in particular to a filtering method based on morphological template conformal mapping in three-dimensional terrain reconstruction. Background Technology

[0002] When restoring the true landforms of surface buildings, the buildings are usually reconstructed in 3D using multi-view images. This is done by post-processing and optimizing the intermediate results, such as disparity maps, depth maps, or digital surface models (DSMs). The disparity maps, depth maps, and DSMs of the landforms can be directly back-calculated to the point cloud of the target landforms from the camera pose. Therefore, ensuring that the features of the landforms in the disparity maps and depth maps are clear, smooth, and complete can provide better initial point cloud values ​​for the subsequent reconstruction of higher-quality 3D models, thereby restoring the true 3D reconstructed landform contour features.

[0003] Traditional methods typically use median filtering, bilateral filtering, or DSM (Disparity Mapping Syndrome) images for post-processing. However, these methods only ensure a certain smoothing effect at the edges, failing to recover accurate contours at the corners of features, thus affecting the quality and accuracy of subsequent reconstruction models. Furthermore, edge enhancement processing is required on the corrected image; if the edge information in the corrected image is already inaccurate, the post-processing result will worsen, failing to restore the true terrain features. Existing technology discloses a surface filtering method that first extracts buildings from the digital orthophoto image, then weights the extracted buildings with geometric information from the DSM image to calculate a "building factor" within a sliding window of the DSM image. This factor is then used to dynamically adjust the filtering parameters within the sliding window based on the terrain, ultimately achieving adaptive parameter progressive morphological filtering. This method can filter out non-ground points while protecting features from loss. However, it cannot accurately identify and preserve the edges and corners of features, failing to fully utilize the image information. Summary of the Invention

[0004] To address the problem that current surface filtering methods cannot accurately identify and preserve the edges and corners of ground features, resulting in the inability to accurately restore the true contours of ground features, this invention proposes a morphological template-based conformal filtering method for 3D terrain reconstruction. This method performs weighted filtering on the positions of the edges and corners to be processed in the image, accurately identifying and preserving the edges and corners of ground features, and accurately restoring the true contours of ground features.

[0005] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows:

[0006] A conformal filtering method based on morphological templates for 3D terrain reconstruction includes the following steps:

[0007] S1. Identify the target features to be processed and obtain a geomorphological representation map of the target features;

[0008] S2. Determine the regular shape of the corner points, and construct multiple morphological templates with the same window size based on the regular shape of the corner points;

[0009] S3. Collect a specific window of the landform representation map, preprocess the specific window, and obtain the pixel code of the specific window;

[0010] S4. Determine whether the pixel code obtained in S3 matches any morphological template. If yes, the specific window corresponding to the current pixel code is a corner point. Perform corner-conserving weighted filtering on the current pixel code and execute step S5. Otherwise, the specific window corresponding to the current pixel code is an edge. Perform median filtering on the current pixel code and execute step S5.

[0011] S5. Process the filtered landform map to obtain the processed landform map;

[0012] S6. Based on the processed geomorphological map obtained in S5, reconstruct a three-dimensional model of the target features.

[0013] In this technical solution, the target feature to be processed is first identified, and a geomorphic representation map of the target feature is obtained. Then, multiple morphological templates of the same specifications are designed based on the regular shapes of the corner points. The purpose is to include the shapes of all regular corner points, which facilitates subsequent corner point matching and preservation. Then, the designed morphological templates are used to detect the positions of the edges and corner points to be processed in the geomorphic representation map. It is determined whether the pixel code of a specific window in the geomorphic representation map matches any morphological template. If the match is successful, the specific window corresponding to the current pixel code is a corner point, and the current pixel code is subjected to corner-conserving weighted filtering. Otherwise, the current pixel code is subjected to median filtering to ensure the smoothness of the image at the edges. The filtered geomorphic representation map is then processed, and the three-dimensional model of the target feature is reconstructed based on the final geomorphic representation map. This achieves accurate identification and preservation of feature edges and corner points, and accurate restoration of the true contour of the feature.

[0014] Preferably, the topographic representation map is a depth map, a disparity map, or a DSM map. In step S2, each morphological template corresponds to a regular shape of a corner point. The pixel codes of the color image and the pixel codes of the elevation image in each morphological template are templates of binary strings. The rule for binary encoding is as follows: the difference between the gray value of the center pixel of the morphological template and the gray value of the surrounding pixels is used as the encoded binary template string. If the difference between the gray value of the center pixel and the gray value of the surrounding pixels is greater than the set depth threshold, then the binary bit encoding of the current pixel is 1; otherwise, the binary bit encoding of the current pixel is 0.

[0015] Preferably, in step S3, the specific windows for acquiring the topographic representation map are all the same size.

[0016] Preferably, in step S3, the specific process of preprocessing is as follows:

[0017] S31. Use the pixels of the topographic representation map as the center pixels of the corresponding specific window;

[0018] S32. If the terrain representation map is a depth or parallax map, then perform grayscale conversion; if the terrain representation map is a DSM map, then do not perform grayscale conversion.

[0019] S33. Encode the current grayscale image or DSM image according to the rules of binary encoding, and output the pixel encoding of a specific window of the terrain representation map.

[0020] Preferably, in step S33, during the process of encoding the pixels of the current grayscale image or the pixels of the DSM image, the current grayscale image is binary encoded pixel by pixel or the elevation of the pixels of the current DSM image is binary encoded pixel by pixel.

[0021] Preferably, in step S4, if the binary string of the current pixel code is consistent with the binary string of any morphological template, then the current pixel code matches the current morphological template successfully; otherwise, the current pixel code fails to match the current morphological template.

[0022] Preferably, in step S4, the specific method for performing conformal weighted filtering is as follows: if the terrain representation map is a disparity map or a depth map, then perform conformal filtering on the disparity map or depth map based on color distance weight; if the terrain representation map is a DSM map, then perform conformal filtering on the DSM map based on elevation distance weight.

[0023] Preferably, the calculation expression for conformal filtering based on color distance weight is:

[0024]

[0025] Where Δd represents the color distance, R c G c B c R represents the RGB color value of the center pixel of a specific window in the disparity map or depth map, respectively. n G n B n Let represent the RGB color values ​​of the neighboring pixels corresponding to the center pixel of a specific window in the disparity map or depth map, respectively. The expression for calculating the color weight of the neighboring pixels corresponding to the center pixel of a specific window in the disparity map or depth map is as follows:

[0026]

[0027] Where alpha represents a hyperparameter of 0.5 and gamma represents a hyperparameter of 60.

[0028] Preferably, the calculation expression for conformal filtering based on elevation distance weight is:

[0029]

[0030] Where Δd represents the elevation distance, Depth c Depth represents the elevation value of the center pixel of a specific window in a DSM map. n The elevation weight of the neighboring pixels corresponding to the center pixel of a specific window in a DSM image represents the elevation value of the neighboring pixels of the center pixel of a specific window in a DSM image. The expression for calculating the elevation weight of the neighboring pixels corresponding to the center pixel of a specific window in a DSM image is as follows:

[0031]

[0032] Where alpha represents a hyperparameter of 0.5 and gamma represents a hyperparameter of 60.

[0033] Preferably, in step S5, the processing of the filtered terrain representation map includes: if the terrain representation map is a disparity map, then the filtered disparity map is examined to obtain the final disparity map; if the terrain representation map is a depth map or a DSM map, then the filtered depth map or DSM map is subjected to bilateral filtering to obtain the final depth map or DSM map.

[0034] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0035] This invention proposes a morphological template-based conformal filtering method for 3D terrain reconstruction. First, the target feature to be processed is identified, and its terrain representation map is obtained. Then, multiple identical morphological templates are designed based on the regular shapes of corner points to encompass the shapes of all regular corner points, facilitating subsequent corner point matching and retention. Next, the designed morphological templates are used to detect the positions of the edges and corner points to be processed in the terrain representation map. It is determined whether the pixel code of a specific window in the terrain representation map matches any morphological template. If a match is successful, the specific window corresponding to the current pixel code is identified as a corner point, and conformal weighted filtering is applied to the current pixel code. Otherwise, median filtering is applied to the current pixel code to ensure the smoothness of the image at the edges. The filtered terrain representation map is then processed, and the 3D model of the target feature is reconstructed based on the final terrain representation map. This method achieves accurate identification and retention of feature edges and corner points, and precise restoration of the true contours of the features. Attached Figure Description

[0036] Figure 1This is a flowchart illustrating a morphological template-based conformal filtering method for three-dimensional terrain reconstruction proposed in this embodiment of the invention.

[0037] Figure 2 A comparison diagram showing the disparity map results of the present invention and the disparity map results of the prior art;

[0038] Figure 3 This is a comparison chart showing the depth map results of the present invention with those of the prior art.

[0039] Figure 4 A comparison diagram showing the DSM diagram results of the present invention and the DSM diagram results of the prior art;

[0040] Figure 5 Design drawing of a morphological template representing corner points;

[0041] Figure 6 A design drawing representing a morphological template for the edge;

[0042] Figure 7 Represents a binary-coded pixel window;

[0043] Figure 8 A schematic diagram illustrating the median filtering method. Detailed Implementation

[0044] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent.

[0045] To better illustrate this embodiment, some parts of the accompanying drawings may be omitted, enlarged, or reduced, and do not represent actual dimensions. The descriptions of directions such as "up" and "down" are not intended to limit this patent.

[0046] It is understandable to those skilled in the art that some well-known details may be omitted from the accompanying drawings;

[0047] The positional relationships depicted in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent.

[0048] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0049] Example 1

[0050] like Figure 1 As shown, a conformal filtering method based on morphological templates for 3D terrain reconstruction includes the following steps:

[0051] S1. Identify the target features to be processed and obtain a geomorphological representation map of the target features;

[0052] In step S1, the topographic representation map is a depth map, a disparity map, or a DSM map.

[0053] S2. Determine the regular shape of the corner points, and construct multiple morphological templates with the same window size based on the regular shape of the corner points;

[0054] S3. Collect a specific window of the landform representation map, preprocess the specific window, and obtain the pixel code of the specific window;

[0055] In step S3, the specific windows of the acquired topographic representation map are all the same size.

[0056] S4. Determine whether the pixel code obtained in S3 matches any morphological template. If yes, the specific window corresponding to the current pixel code is a corner point. Perform corner-conserving weighted filtering on the current pixel code and execute step S5. Otherwise, the specific window corresponding to the current pixel code is an edge. Perform median filtering on the current pixel code and execute step S5.

[0057] S5. Process the filtered landform map to obtain the processed landform map;

[0058] In step S5, the processing of the filtered terrain representation map includes: if the terrain representation map is a disparity map, the filtered disparity map is checked by performing a consistency check between the left and right disparity maps to obtain the final disparity map; if the terrain representation map is a depth map or a DSM map, the filtered depth map or DSM map is subjected to bilateral filtering to obtain the final depth map or DSM map. Bilateral filtering is a non-linear filtering method that is a compromise between spatial proximity and pixel value similarity of the image. It considers both spatial information and grayscale similarity to achieve the purpose of edge preservation and noise reduction. It has the characteristics of simplicity, non-iteration, and locality. The advantage of bilateral filtering is that it can preserve edges.

[0059] S6. Based on the processed geomorphological map obtained in S5, reconstruct a three-dimensional model of the target features.

[0060] In this embodiment, the target feature to be processed is first determined, and a topographic representation map of the target feature is obtained. Then, multiple morphological templates of the same specifications are designed based on the regular shapes of the corner points. The purpose is to include the shapes of all regular corner points, which facilitates subsequent corner point matching and preservation. Then, the designed morphological templates are used to detect the positions of the edges and corner points to be processed in the topographic representation map. It is determined whether the pixel code of a specific window in the topographic representation map matches any morphological template. If the match is successful, the specific window corresponding to the current pixel code is a corner point, and the current pixel code is subjected to corner-conserving weighted filtering. Otherwise, the current pixel code is subjected to median filtering to ensure the smoothness of the image at the edges. The filtered topographic representation map is then processed. Based on the final topographic representation map, i.e., the final depth map, disparity map, or DSM map, the three-dimensional model of the target feature is reconstructed, which realizes accurate identification and preservation of feature edges and corner points, and accurate restoration of the true outline of the feature. Figure 2 This is a comparison chart showing the disparity map results obtained in this embodiment with those in the prior art. Figure 2 It is evident that the parallax map obtained in this embodiment produces more distinct building edges and corners, which better reflects the actual scene. Figure 3 This is a comparison chart showing the depth map results of this embodiment with those of existing technology. Figure 3 It is evident that in this embodiment, the depth map results show that the edges of the red house are smoother and the outline is clearer; Figure 4 This diagram shows a comparison between the DSM diagram results of this embodiment and the DSM diagram results of the prior art. Figure 4 It is evident that the DSM diagram results in this embodiment show clearer outlines of the corner points of residential buildings.

[0061] Example 2

[0062] See Figure 1 In step S2, the topographic representation map is a depth map, a disparity map, or a DSM map. Each morphological template corresponds to the geometric shape of a regular corner point, and the pixel values ​​of the color image or the pixel values ​​of the elevation image in each morphological template are encoded into a binary string template. See [link to relevant documentation]. Figure 5 and Figure 6 A total of 40 morphological templates for corner points were designed, encompassing the regular shapes of all corner points. These templates were encoded into 40 binary strings as target feature matching objects. Additionally, 9 morphological templates for edges were designed to facilitate corner and edge matching using pixel encoding of the input depth map, disparity map, or DSM map. The binary encoding rule is as follows: the difference between the grayscale value of the center pixel of the morphological template and the grayscale values ​​of the surrounding pixels is used as the encoded binary template string. If the difference between the grayscale value of the center pixel and the grayscale value of the surrounding pixels is greater than the set depth threshold, the binary bit encoding of the current pixel is 1; otherwise, the binary bit encoding of the current pixel is 0.

[0063] See Figure 7 Taking a 3x3 window as an example, the difference between the grayscale value of the center pixel of the morphological template and the grayscale values ​​of the surrounding 8 pixels is used as the encoded binary template string. A depth threshold is set, and it is determined whether the difference between the grayscale value of the current pixel and the grayscale values ​​of the surrounding pixels is greater than the set depth threshold. If so, the corresponding binary bit is encoded as 1; otherwise, the corresponding binary bit is encoded as 0. Figure 7 The grayscale value of the center pixel in the image is 10. The grayscale values ​​of the surrounding 8 pixels, from left to right and from top to bottom, are 34, 25, 12, 15, 46, 32, 11, and 25, respectively. The differences between the grayscale value of the center pixel and the grayscale values ​​of the surrounding 8 pixels are 14, 15, 2, 5, 36, 22, 1, and 15, respectively. Assuming a depth threshold of 3, the binary bits with a difference greater than the depth threshold of 3 are encoded as 1, and the binary bits with a difference less than the depth threshold of 3 are encoded as 0. The resulting binary encoded pixel value of the 3*3 window image is 11011101.

[0064] Example 3

[0065] See Figure 1 In step S3, the specific preprocessing process is as follows:

[0066] S31. Use the pixels of the topographic representation map as the center pixels of the corresponding specific window;

[0067] S32. If the terrain feature map is a depth or parallax map, then perform grayscale conversion; if the terrain feature map is a DSM map, then do not perform grayscale conversion.

[0068] In step S32, the grayscale conversion of the depth map or disparity map is as follows: first, determine the RGB image corresponding to the depth map or disparity map, and then convert the current RGB image into a grayscale image.

[0069] S33. Encode the current grayscale image or DSM image according to the rules of binary encoding, and output the pixel encoding of a specific window of the terrain representation map.

[0070] In step S33, during the encoding of the pixels of the current grayscale image or the pixels of the DSM image, the current grayscale image is binary encoded pixel by pixel, or the elevation of the pixels of the current DSM image is binary encoded pixel by pixel.

[0071] In step S4, if the binary string of the current pixel code matches the binary string of any morphological template, then the current pixel code matches the current morphological template successfully, meaning that every binary string is identical. For example, Figure 7If the binary string of the morphological template is 11011101, then the binary string of the current pixel code must be the same as 11011101 to prove that the two match successfully; otherwise, the current pixel code fails to match the current morphological template, and median filtering for edge smoothing is required. The specific method for performing conformal weighted filtering is as follows: For disparity maps or depth maps, conformal filtering based on color distance weights is required. The calculation expression for conformal filtering based on color distance weights is:

[0072]

[0073] Where Δd represents the color distance, R c G c B c R represents the RGB color value of the center pixel of a specific window in the disparity map or depth map, respectively. n G n B n These represent the RGB color values ​​of the neighboring pixels corresponding to the center pixel of a specific window in the disparity map or depth map, and the color weights of the neighboring pixels corresponding to the center pixel of a specific window in the disparity map or depth map, respectively. weight The calculation expression is:

[0074]

[0075] Where alpha represents a hyperparameter of 0.5 and gamma represents a hyperparameter of 60.

[0076] Example 4

[0077] For DSM maps, conformal filtering based on elevation distance weights is required. The calculation expression for conformal filtering based on elevation distance weights is as follows:

[0078]

[0079] Where Δd represents the elevation distance, Depth c Depth represents the elevation value of the center pixel of a specific window in a DSM map. n The elevation weight of the neighboring pixels corresponding to the center pixel of a specific window in a DSM image represents the elevation value of the neighboring pixels of the center pixel of a specific window in a DSM image. The expression for calculating the elevation weight of the neighboring pixels corresponding to the center pixel of a specific window in a DSM image is as follows:

[0080]

[0081] Where alpha represents a hyperparameter of 0.5 and gamma represents a hyperparameter of 60.

[0082] See Figure 8The method for median filtering to smooth edges is as follows: A 3x3 pixel image window is selected. The depth or elevation value of the center pixel of the left morphological template is 10. The depth or elevation values ​​of the surrounding 8 pixels, from left to right and top to bottom, are 6, 3, 3, 2, 5, 4, 4, and 6. These depth or elevation values ​​are sorted as 2, 3, 3, 4, 4, 5, 6, 6, 10. Then, the median value of 4 is taken from this sequence. The depth or elevation value of the center pixel of the left morphological template is then updated, and the median value of 4 is used to update the depth or elevation value of the center pixel. The median filtering method in this embodiment is a non-linear smoothing... Median filtering is a non-linear signal processing technique based on sorting statistics that effectively suppresses noise. The basic principle of median filtering is to replace the value of a point in a digital image or sequence with the median of the values ​​of all points in its neighborhood, making the surrounding pixel values ​​closer to the true value, thereby eliminating isolated noise points. The method involves using a two-dimensional sliding template of a certain structure to sort the pixels within the template according to their pixel values, generating a monotonically increasing or decreasing two-dimensional data sequence. The median of this monotonically increasing two-dimensional data sequence is then taken as the output, resulting in a two-dimensional median filter.

[0083] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A conformal filtering method based on morphological templates for three-dimensional terrain reconstruction, characterized in that, Includes the following steps: S1. Identify the target features to be processed and obtain a geomorphological representation map of the target features; S2. Determine the regular shape of the corner points, and construct multiple morphological templates with the same window size based on the regular shape of the corner points; S3. Collect a specific window of the landform representation map, preprocess the specific window, and obtain the pixel code of the specific window; S4. Determine whether the pixel code obtained in S3 matches any morphological template. If yes, the specific window corresponding to the current pixel code is a corner point. Perform corner-conserving weighted filtering on the current pixel code and execute step S5. Otherwise, the specific window corresponding to the current pixel code is an edge. Perform median filtering on the current pixel code and execute step S5. In step S4, the specific method for performing conformal weighted filtering is as follows: if the terrain representation map is a disparity map or a depth map, then perform conformal filtering based on color distance weight on the disparity map or depth map; if the terrain representation map is a DSM map, then perform conformal filtering based on elevation distance weight on the DSM map. The calculation expression for conformal filtering based on color distance weights is: in, Indicates color distance. R c 、G c 、B c These represent the center pixels of a specific window in the disparity map or depth map, respectively. RGB Color value, R n 、G n 、B n These represent the neighboring pixels corresponding to the center pixel of a specific window in the disparity map or depth map, respectively. RGB Color value, the color weight of the neighboring pixels corresponding to the center pixel of a specific window in a disparity map or depth map. weight The calculation expression is: in, alpha This represents a hyperparameter of 0.

5. gamma This represents a hyperparameter of 60; The calculation expression for conformal filtering based on elevation distance weight is: in, Indicates elevation distance. Depth c This represents the elevation value of the center pixel of a specific window in a DSM map. Depth n This represents the elevation value of the neighboring pixels corresponding to the center pixel of a specific window in a DSM graph, and the elevation weight of the neighboring pixels corresponding to the center pixel of a specific window in a DSM graph. weight The calculation expression is: Where alpha represents a hyperparameter of 0.5 and gamma represents a hyperparameter of 60; S5. Process the filtered landform map to obtain the processed landform map; S6. Based on the processed geomorphological map obtained in S5, reconstruct a three-dimensional model of the target features.

2. The conformal filtering method based on morphological templates in three-dimensional terrain reconstruction according to claim 1, characterized in that, The topographic representation map is a depth map, a disparity map, or a DSM map. In step S2, each morphological template corresponds to a regular shape of a corner point. The pixel codes of the color image and the pixel codes of the elevation image in each morphological template are templates of binary strings. The rule of binary encoding is as follows: the difference between the gray value of the center pixel of the morphological template and the gray value of the surrounding pixels is used as the encoded binary template string. If the difference between the gray value of the center pixel and the gray value of the surrounding pixels is greater than the set depth threshold, the binary bit encoding of the current pixel is 1; otherwise, the binary bit encoding of the current pixel is 0.

3. The conformal filtering method based on morphological templates in three-dimensional terrain reconstruction according to claim 1, characterized in that, In step S3, the specific windows of the acquired topographic representation map are all the same size.

4. The conformal filtering method based on morphological templates in three-dimensional terrain reconstruction according to claim 2, characterized in that, In step S3, the specific preprocessing process is as follows: S31. Use the pixels of the topographic representation map as the center pixels of the corresponding specific window; S32. If the terrain feature map is a depth or parallax map, then perform grayscale conversion; if the terrain feature map is a DSM map, then do not perform grayscale conversion. S33. Encode the current grayscale image or DSM image according to the rules of binary encoding, and output the pixel encoding of a specific window of the terrain representation map.

5. The conformal filtering method based on morphological templates in three-dimensional terrain reconstruction according to claim 4, characterized in that, In step S33, during the encoding of the pixels of the current grayscale image or the pixels of the DSM image, the current grayscale image is binary encoded pixel by pixel, or the elevation of the pixels of the current DSM image is binary encoded pixel by pixel.

6. The morphological template-based conformal filtering method for three-dimensional terrain reconstruction according to claim 5, characterized in that, In step S4, if the binary string of the current pixel code is consistent with the binary string of any morphological template, then the current pixel code matches the current morphological template successfully; otherwise, the current pixel code fails to match the current morphological template.

7. The morphological template-based conformal filtering method for three-dimensional terrain reconstruction according to claim 6, characterized in that, In step S5, the processing of the filtered terrain representation map includes: if the terrain representation map is a disparity map, the filtered disparity map is examined to obtain the final disparity map; if the terrain representation map is a depth map or a DSM map, the filtered depth map or DSM map is subjected to bilateral filtering to obtain the final depth map or DSM map.

Images