Three-dimensional reconstruction method and device for large-scale scene and electronic equipment

Abstract

The invention relates to a three-dimensional reconstruction method and device for a large-scale scene and electronic equipment, relates to the technical field of three-dimensional reconstruction, and solves the problem that single equipment cannot efficiently complete a three-dimensional reconstruction task of a large-scale or even super-large-scale scene. The three-dimensional reconstruction method for the large-scale scene comprises the following steps: acquiring a plurality of aerial images of the large-scale scene in an array mode; processing the plurality of aerial images by using an ortho-image generation algorithm to obtain an ortho-image of the large-scale scene; semantically segmenting the orthoimage by using a scene segmentation algorithm; blocking the orthoimage subjected to semantic segmentation by adopting a quadtree blocking algorithm to obtain a plurality of target sub-blocks; and matching one or more aerial images corresponding to each target sub-block, and performing three-dimensional reconstruction according to one or more aerial images in each target sub-block.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional reconstruction, and particularly to a method, device and electronic device for three-dimensional reconstruction of large-scale scenes. Background Art

[0002] Three-dimensional reconstruction based on drones has great application significance. It mainly captures images of a given area by aerial photography and then executes three-dimensional modeling tasks by running three-dimensional reconstruction algorithms on a high-performance server offline. Existing three-dimensional modeling systems input all images into the three-dimensional modeling algorithm at once, which works in the case of small scenes. Although the complexity of the three-dimensional modeling algorithm is usually high, the algorithm execution time is acceptable in small scenes. However, if faced with the three-dimensional reconstruction task of large-scale or even ultra-large-scale scenes, such as a scene of 5 kilometers by 5 kilometers or even the modeling of a city, it is unrealistic to rely solely on a single computing node for three-dimensional reconstruction. The reasons are as follows:

[0003] 1. The amount of collected images is huge, so the execution time of algorithms in the three-dimensional modeling stage such as feature extraction and matching on a single device is too long to be acceptable;

[0004] 2. The operation of three-dimensional reconstruction algorithms often requires a large amount of memory and hard disk storage space, which is often difficult to provide by a single device;

[0005] 3. As the scene becomes larger and the amount of images increases, the increase in algorithm complexity is not linear, but shows a quadratic growth in complexity;

[0006] 4. Unexpected errors may occur during the three-dimensional modeling process, so executing all reconstruction tasks on a single device increases the risk. Summary of the Invention

[0007] In view of the above analysis, embodiments of the present invention aim to provide a method, device and electronic device for three-dimensional reconstruction of large-scale scenes, so as to solve the problem that a single device cannot efficiently complete the three-dimensional reconstruction task of large-scale or even ultra-large-scale scenes.

[0008] In a first aspect, embodiments of the present invention provide a method for three-dimensional reconstruction of large-scale scenes, including the following steps:

[0009] Collect multiple aerial images of a large-scale scene in an arrayed manner;

[0010] Process the multiple aerial images using an orthophoto generation algorithm to obtain an orthophoto of the large-scale scene;

[0011] Semantically segment the orthophoto using a scene segmentation algorithm;

[0012] The orthophoto image after semantic segmentation is segmented using a quadtree segmentation algorithm to obtain a plurality of target sub - blocks; and

[0013] Match one or more of the aerial images corresponding to each of the target sub - blocks and perform 3D reconstruction based on one or more of the aerial images within each of the target sub - blocks.

[0014] Based on a further improvement of the above - mentioned method, the quadtree segmentation algorithm includes the following constraints:

[0015] Minimum target sub - block constraint or

[0016] Based on a further improvement of the above - mentioned method, the quadtree segmentation algorithm further includes the following constraints: minimum target sub - block constraint and maximum recursion depth constraint

[0017] An intra - block inconsistency index constraint for measuring the balance of different classes within the target sub - block, where when the intra - block inconsistency index constraint conflicts with the maximum recursion depth constraint or the minimum target sub - block constraint, the intra - block inconsistency index constraint is cancelled.

[0018] Based on a further improvement of the above - mentioned method, the intra - block inconsistency index constraint is determined in the following way:

[0019] Judge whether the total area of the target classes within the target sub - block is greater than a preset threshold. If so, the intra - block inconsistency index constraint is satisfied; otherwise, the intra - block inconsistency index constraint is not satisfied.

[0020] Based on a further improvement of the above - mentioned method, before using the orthophoto image generation algorithm to process the plurality of aerial images, the method further includes:

[0021] Adjust the sizes of the plurality of aerial images.

[0022] Based on a further improvement of the above - mentioned method, before using the scene segmentation algorithm to semantically segment the orthophoto image, the method further includes:

[0023] Adjust the size of the orthophoto image according to the accuracy requirements of the 3D modeling task.

[0024] Based on a further improvement of the above - mentioned method, performing 3D reconstruction based on one or more of the aerial images within each of the target sub - blocks includes:

[0025] Perform 3D reconstruction for each of the target sub - blocks using separate 3D modeling parameters respectively; and

[0026] Stitch the 3D reconstruction results corresponding to each of the target sub - blocks to obtain the 3D reconstruction result of the large - scale scene.

[0027] Based on the further improvement of the above method, the scene segmentation algorithm is the Unet algorithm.

[0028] In a second aspect, an embodiment of the present invention provides an apparatus for three-dimensional reconstruction of a large-scale scene, including:

[0029] An acquisition module, configured to acquire multiple aerial images of the large-scale scene in an arrayed manner;

[0030] An orthorectification module, configured to process the multiple aerial images using an orthophoto generation algorithm to obtain an orthophoto of the large-scale scene;

[0031] A semantic segmentation module, configured to semantically segment the orthophoto using a scene segmentation algorithm;

[0032] A chunking module, configured to chunk the orthophoto after semantic segmentation using a quadtree chunking algorithm to obtain multiple target sub-chunks; and

[0033] A modeling module, configured to match one or more of the aerial images corresponding to each target sub-chunk and perform three-dimensional reconstruction based on one or more of the aerial images within each target sub-chunk.

[0034] Based on the further improvement of the above apparatus, the quadtree chunking algorithm includes the following constraints:

[0035] A maximum recursion depth constraint and a minimum target sub-chunk constraint.

[0036] Based on the further improvement of the above apparatus, the quadtree chunking algorithm further includes the following constraints:

[0037] An intra-block inconsistency index constraint for measuring the balance of different categories within the target sub-chunk, where when the intra-block inconsistency index constraint conflicts with the maximum recursion depth constraint or the minimum target sub-chunk constraint, the intra-block inconsistency index constraint is cancelled.

[0038] Based on the further improvement of the above apparatus, the intra-block inconsistency index constraint is determined in the following manner:

[0039] Judge whether the total area of the target category within the target sub-chunk is greater than a preset threshold. If so, the intra-block inconsistency index constraint is satisfied; otherwise, the intra-block inconsistency index constraint is not satisfied.

[0040] In a third aspect, an embodiment of the present invention provides an electronic device, including:

[0041] At least one processor; and

[0042] A memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the method for three-dimensional reconstruction of large-scale scenes according to any one of the first aspects of the present invention.

[0043] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0044] 1. The solution of the present invention accelerates the three-dimensional reconstruction process of large-scale scenes, shortens the three-dimensional reconstruction time, and optimizes the execution efficiency of the three-dimensional modeling algorithm for large-scale scenes in the deployment stage by intelligently segmenting large-scale scenes and performing separate three-dimensional reconstruction on each block.

[0045] 2. The solution of the present invention optimizes the three-dimensional modeling process and reduces the instability of the three-dimensional reconstruction process by providing high-precision three-dimensional modeling for regions of interest and low-precision three-dimensional modeling for regions of no interest.

[0046] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent specification, and some advantages can be made obvious from the specification or understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings are only for the purpose of illustrating specific embodiments and are not considered as limiting the present invention. Throughout the drawings, the same reference signs denote the same components;

[0048] Figure 1 Shows a schematic flowchart of a method for three-dimensional reconstruction of large-scale scenes according to an embodiment of the present invention.

[0049] Figure 2 Shows an exemplary block diagram of an apparatus for three-dimensional reconstruction of large-scale scenes according to an embodiment of the present invention.

[0050] Figure 3 Shows a schematic block diagram of an example electronic device 300 that can be used to implement the embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The following will specifically describe the preferred embodiments of the present invention with reference to the drawings, wherein the drawings form a part of the present application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0052] Figure 1The flowchart shows a method for 3D reconstruction of large-scale scenes according to an embodiment of the present invention. As Figure 1 shown, the method for 3D reconstruction of large-scale scenes includes the following steps:

[0053] Step S100: Collect multiple aerial images of the large-scale scene in an arrayed manner.

[0054] In this embodiment, a drone or other aircraft can be used to systematically photograph the target area (e.g., in an arrayed manner) along a predetermined flight path. The target area is a large-scale scene. By using an arrayed manner (such as Z-shaped aerial photography), it can ensure that the images cover the entire large-scale scene, while ensuring that there is sufficient overlap between adjacent images for subsequent feature matching and 3D reconstruction.

[0055] Step S200: Process the multiple aerial images using an orthophoto generation algorithm to obtain an orthophoto of the large-scale scene.

[0056] In this embodiment, an orthophoto generation algorithm can be used to process the multiple collected aerial images to generate an orthophoto of the large-scale scene. The orthophoto algorithm here can be Map2Dfusion. The orthophoto algorithm aggregates the input multiple aerial images into a large orthophoto. In addition, the orthophoto generation algorithm can also correct the tilted aerial images into vertical images, thus eliminating the influence of terrain undulation and shooting angle. Using the orthophoto generation algorithm can improve the geometric accuracy of the images and provide a reliable basis for subsequent scene segmentation and 3D reconstruction.

[0057] In some embodiments, before generating the orthophoto, the resolution of the aerial images can be adjusted first to reduce the computational complexity.

[0058] Step S300: Semantically segment the orthophoto using a scene segmentation algorithm.

[0059] In this embodiment, the Unet algorithm can be applied to semantically segment the orthophoto. Scene segmentation is to divide the orthophoto into multiple regions with the same or similar semantics, such as buildings, roads, vegetation, etc. For example, the orthophoto is an image with a resolution of 5000×5000. After semantic segmentation, the resolution of the image is still 5000×5000, but the regions in the image have been divided according to semantics. This step provides important semantic information for subsequent chunking and 3D reconstruction, ensuring that the content within each subsequent target sub-chunk has high homogeneity, thereby improving the accuracy and efficiency of reconstruction.

[0060] Step S400: Chunk the orthophoto after semantic segmentation using a quadtree chunking algorithm to obtain multiple target sub-chunks.

[0061] The quadtree partitioning algorithm is a recursive image partitioning method that divides an image into four quadrants and further recursively partitions as needed until specific constraint conditions are met. In this embodiment, through the quadtree partitioning algorithm, a large-scale scene can be segmented into multiple target sub-blocks, and the image sets within each target sub-block have high homogeneity, thus facilitating subsequent 3D reconstruction.

[0062] In this embodiment, the quadtree partitioning algorithm includes the following constraints:

[0063] 1. Maximum recursive depth constraint: Restricts the depth of recursion to avoid increased computational complexity caused by excessive partitioning.

[0064] 2. Minimum target sub-block constraint: Ensures that each sub-block contains a sufficient number of images to support effective 3D reconstruction.

[0065] 3. Intra-block inconsistency index constraint: Measures the balance of different classes within a sub-block and ensures that the proportion of the main class within each sub-block is greater than a preset threshold.

[0066] It should be noted that among these three constraints, the minimum target sub-block constraint has the highest priority, and the maximum recursive depth constraint can be executed without violating the minimum target sub-block constraint.

[0067] In some embodiments, when the intra-block inconsistency index constraint conflicts with the maximum recursive depth constraint or the minimum target sub-block constraint, the intra-block inconsistency index constraint can be cancelled to ensure the feasibility and efficiency of partitioning.

[0068] In some embodiments, the intra-block inconsistency index constraint is determined as follows: Determine whether the total area of the target class within the target sub-block is greater than the preset threshold. If so, the intra-block inconsistency index constraint is satisfied; otherwise, it is not satisfied. For example, if the preset threshold is set to 0.8 (i.e., 80%), then it can be judged the area size of the occupied class region within the target sub-block in the target sub-block. If the area occupied by the occupied class region is greater than 80%, it meets the intra-block inconsistency index constraint; otherwise, it does not.

[0069] Step S500: Match one or more of the aerial images corresponding to each of the target sub-blocks and perform 3D reconstruction based on one or more of the aerial images within each of the target sub-blocks.

[0070] In this embodiment, for each target sub-block, an appropriate aerial image can be selected for 3D reconstruction. In this embodiment, the GPS information of the aerial image can be used to match the target sub-blocks. For example, if the GPS range of the target sub-block is X1, then the aerial images with GPS information falling within X1 can be found from the aerial images. By matching the images within the sub-blocks, the geometric structure and texture information of the scene can be restored more precisely. In this embodiment, the sets of sub-block images can be assigned to independent computing nodes, and a 3D modeling software (such as WebODM) can be used for block-by-block reconstruction. The 3D reconstruction process includes steps such as feature extraction, feature matching, sparse reconstruction, and dense reconstruction, and finally a high-precision 3D model is generated.

[0071] After obtaining the 3D reconstruction results of all sub-blocks, these 3D reconstruction results can be stitched together to generate a complete 3D reconstruction model. Through the stitching process, the consistency of the 3D models of each sub-block in terms of geometry and texture can be ensured, the differences in the possible overlapping areas can be eliminated, and finally a seamless and high-precision large-scale scene 3D model is generated.

[0072] In some embodiments, different 3D modeling parameters can be used for each target sub-block. For example, for areas of no interest such as mountains and grasslands, the accuracy of the 3D model can be lowered to save computing power. For areas of interest such as buildings and streets, the accuracy of the 3D model can be increased to meet the task requirements. In this embodiment, the 3D modeling process is optimized by block-by-block targeted optimization of the modeling algorithms for each scene, thereby reducing the instability of the 3D reconstruction process.

[0073] The following takes the 3D modeling of the 5km * 5km scale surface as an example to illustrate the method of the embodiment of the present invention.

[0074] The following takes the 5km × 5km surface modeling as an example to illustrate the implementation process of the present invention:

[0075] 1. Data acquisition stage

[0076] The aerial images are taken by an unmanned aerial vehicle (UAV) array, and the resolution of the aerial images is adjusted to 1920 × 1080 to reduce the data volume and avoid waste of computing power.

[0077] 2. Preprocessing stage

[0078] — Use the orthophoto generation algorithm Map2Dfusion to process all aerial data to obtain an almost real-time orthophoto.

[0079] — Adjust the size of the orthophoto to 5000 * 5000 to reduce the data volume and avoid waste of computing power.

[0080] — Use the improved Unet algorithm to perform semantic segmentation on the orthophoto and identify areas such as buildings and vegetation.

[0081] — Use the quadtree partitioning algorithm to process the scene segmentation map to obtain appropriate sub - blocks. The maximum recursive depth is required to be 4, each block should contain at least 100 images, and the within - block inconsistency index value of each block is greater than 0.8.

[0082] — Match the aerial image sets corresponding to each sub - block (through GPS matching) and generate 3D modeling parameters. The modeling parameters include scene categories, and borrow the modeling algorithms for each scene in 3D modeling software to optimize each block specifically.

[0083] 3. 3D Reconstruction Stage

[0084] Use the WebODM 3D modeling software to separately calculate the 3D reconstruction results for the aerial image sets corresponding to all blocks using the modeling algorithm. In this stage, the 3D reconstruction parameters adopt the 3D reconstruction parameters of their respective blocks. Here, a 3D modeling calculation node pool can be set up to adaptively execute the reconstruction tasks. Since the 3D modeling algorithm can run on multiple hosts, there are multiple calculation nodes available for adaptive use.

[0085] 4. Post - processing Stage

[0086] After obtaining all the 3D reconstruction results, the individual sub - 3D reconstruction results can be stitched into the final 3D reconstruction result, or directly generate tiles and display them on the front - end.

[0087] Figure 2 shows an exemplary block diagram of an apparatus for 3D reconstruction of large - scale scenes according to an embodiment of the present invention. As Figure 2 shown, the apparatus 200 for 3D reconstruction of large - scale scenes includes: an acquisition module 201 configured to acquire multiple aerial images of a large - scale scene in an arrayed manner; an ortho - module 202 configured to process the multiple aerial images using an orthophoto generation algorithm to obtain an orthophoto of the large - scale scene; a semantic segmentation module 203 configured to semantically segment the orthophoto using a scene segmentation algorithm; a partitioning module 204 configured to partition the semantically segmented orthophoto using a quadtree partitioning algorithm to obtain multiple target sub - blocks; and a modeling module 205 configured to match one or more of the aerial images corresponding to each of the target sub - blocks and perform 3D reconstruction based on one or more of the aerial images within each of the target sub - blocks.

[0088] It should be understood that Figure 2 each module of the apparatus 200 shown in Figure 1correspond to the respective steps in the described method. Thus, the operations, features, and advantages described above for the method also apply to the apparatus 200 and the modules included therein. For the sake of brevity, some operations, features, and advantages are not described herein again.

[0089] Figure 3 FIG. shows a schematic block diagram of an exemplary electronic device 300 that can be used to implement embodiments of the present disclosure. Referring to Figure 3 , a structural block diagram of the electronic device 300 that can be used as a server or a client of the present disclosure will now be described. It is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein. As Figure 3 shown, the electronic device 300 includes a computing unit 301 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the device 300 can also be stored. The computing unit 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304. A plurality of components in the device 300 are connected to the I / O interface 305, including: an input unit 306, such as a keyboard, a mouse, etc.; an output unit 307, such as various types of displays, speakers, etc.; a storage unit 308, such as a magnetic disk, an optical disk, etc.; and a communication unit 309, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 309 allows the device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0090] The computing unit 301 may be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 executes the various methods and processes described above, such as the method for three-dimensional reconstruction of large-scale scenes. For example, in some embodiments, the method for three-dimensional reconstruction of large-scale scenes may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed onto the device 300 via the ROM 302 and / or the communication unit 309. When the computer program is loaded into the RAM 303 and executed by the computing unit 301, one or more steps of the gift package recommendation method described above may be executed. Alternatively, in other embodiments, the computing unit 301 may be configured to execute the method for three-dimensional reconstruction of large-scale scenes in any other suitable manner (e.g., by means of firmware).

[0091] Embodiments of the present invention provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method for three-dimensional reconstruction of large-scale scenes described in any one of the above. Herein, one or more computer-readable non-transitory storage media may include one or more semiconductor-based or other integrated circuits (ICs) (e.g., a field-programmable gate array (FPGA) or an application-specific IC (ASIC)), a hard disk drive (HDD), a hybrid hard disk drive (HHD), an optical disc, an optical disc drive (ODD), a magneto-optical disc, a magneto-optical disc drive, a floppy disk, a floppy disk drive (FDD), a magnetic tape, a solid-state drive (SSD), a RAM drive, or any other suitable computer-readable non-transitory storage medium. The computer-readable non-transitory storage medium may be volatile, non-volatile, or a combination of volatile and non-volatile.

[0092] Compared with the prior art, at least one of the following beneficial effects can be achieved by the embodiments of the present invention:

[0093] 1. The solution of the present invention accelerates the three-dimensional reconstruction process of large-scale scenes, shortens the three-dimensional reconstruction time, and optimizes the execution efficiency of the three-dimensional modeling algorithm for large-scale scenes in the deployment stage.

[0094] 2. The solution of the present invention optimizes the three-dimensional modeling process by providing high-precision three-dimensional modeling for regions of interest and low-precision three-dimensional modeling for regions of no interest, thereby reducing the instability of the three-dimensional reconstruction process.

[0095] As described above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for three-dimensional reconstruction of a large-scale scene, characterized in that: The steps include: Use array-based methods to collect multiple aerial images of large-scale scenes; Processing the plurality of aerial images using an orthophoto generation algorithm to obtain an orthophoto of a large-scale scene; Semantically segmenting the orthophoto using a scene segmentation algorithm; The orthophoto image after semantic segmentation is divided into blocks using a quadtree block algorithm to obtain a plurality of target sub-blocks; as well as One or more aerial images corresponding to each target sub-block are matched and three-dimensional reconstruction of the large-scale scene is performed according to the one or more aerial images in each target sub-block.

2. The method according to claim 1, characterized in that The quadtree partitioning algorithm includes the following limitations: Minimum target subblock constraint or Minimum target sub-block constraint and maximum recursion depth constraint.

3. The method according to claim 2, characterized in that The quadtree partitioning algorithm also includes the following restrictions: An intra-block inconsistency indicator constraint is used to measure the balance of different categories in the target sub-block, wherein when the intra-block inconsistency indicator constraint conflicts with the maximum recursion depth constraint or the minimum target sub-block constraint, the intra-block inconsistency indicator constraint is cancelled.

4. The method according to claim 3, characterized in that The intra-block inconsistency indicator constraint is determined in the following way: It is determined whether the total area of ​​the target category in the target sub-block is greater than a preset threshold. If so, the intra-block inconsistency indicator constraint is satisfied; otherwise, the intra-block inconsistency indicator constraint is not satisfied.

5. The method according to claim 1, characterized in that Before using the orthophoto generation algorithm to process the plurality of aerial images, the method further includes: The plurality of aerial images are resized.

6. The method according to claim 1, characterized in that Before semantically segmenting the orthophoto using a scene segmentation algorithm, the method further includes: The size of the orthophoto is adjusted according to the accuracy requirement of the three-dimensional modeling task.

7. The method according to claim 1, characterized in that Performing three-dimensional reconstruction according to one or more aerial images in each target sub-block includes: Performing three-dimensional reconstruction for each target sub-block using separate three-dimensional modeling parameters; and The three-dimensional reconstruction results corresponding to each target sub-block are stitched to obtain a three-dimensional reconstruction result of a large-scale scene.

8. The method according to claim 1, characterized in that The scene segmentation algorithm is the Unet algorithm.

9. A device for three-dimensional reconstruction of large-scale scenes, characterized in that: include: A collection module is configured to collect multiple aerial images of a large-scale scene in an array manner; An orthophoto module is configured to process the plurality of aerial images using an orthophoto generation algorithm to obtain an orthophoto of a large-scale scene; A semantic segmentation module, configured to semantically segment the orthophoto using a scene segmentation algorithm; A block division module is configured to use a quadtree block division algorithm to divide the orthophoto after semantic segmentation into blocks to obtain a plurality of target sub-blocks; as well as The modeling module is configured to match one or more aerial images corresponding to each target sub-block and perform three-dimensional reconstruction according to one or more aerial images in each target sub-block.

10. The device according to claim 9, characterized in that The quadtree partitioning algorithm includes the following limitations: Maximum recursion depth constraint and minimum target sub-block constraint.

11. The device according to claim 10, characterized in that The quadtree partitioning algorithm also includes the following restrictions: An intra-block inconsistency indicator constraint is used to measure the balance of different categories in the target sub-block, wherein when the intra-block inconsistency indicator constraint conflicts with the maximum recursion depth constraint or the minimum target sub-block constraint, the intra-block inconsistency indicator constraint is cancelled.

12. The device according to claim 11, characterized in that The intra-block inconsistency indicator constraint is determined in the following way: It is determined whether the total area of ​​the target category in the target sub-block is greater than a preset threshold. If so, the intra-block inconsistency indicator constraint is satisfied; otherwise, the intra-block inconsistency indicator constraint is not satisfied.

13. An electronic device, comprising: at least one processor; A memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor so that the at least one processor can execute the method for three-dimensional reconstruction of a large-scale scene as described in any one of claims 1-8.

Images