Building height determination method and system, storage medium and computer program product

Abstract

The invention provides a building height determination method. The method comprises the following steps: determining a top prediction frame of a to-be-identified building in a remote sensing image and a plurality of bottom candidate frames of the to-be-identified building; according to the geometric features of the top prediction frame and the geometric features of the bottom candidate frames, determining a bottom prediction frame of the to-be-identified building in the plurality of bottom candidate frames; the pixel offset between the top prediction frame and the bottom prediction frame is analyzed, the height of the building to be recognized is determined, the pixel offset represents the pixel dimension distance from the center point of the top prediction frame to the center point of the bottom prediction frame, and the pixel offset is obtained from the remote sensing image. The invention further provides a building height determination system, a storage medium and a computer program product.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing, and in particular to a method for determining the height of a building, a system for determining the height of a building, a storage medium, and a computer program product. Background Art

[0002] The identification of building height is of great significance in the fields of urban planning, architectural design, real estate appraisal, etc. It can help relevant departments better carry out urban space planning, and help to assess the impact of natural disasters on buildings and ensure the compliance of new projects. In related technologies, geometric optics or lidar are mainly used to measure the height of buildings. The geometric optics method measures the length of the shadow of the building and the length of the shadow of an object of known height (such as a person or a pole) at the same time to infer the height of the building; this method relies on the length of the shadow of the building, but in a dense urban environment, buildings block each other and shadow measurement is more difficult. Lidar uses laser pulses to measure the time required from the emission to the building to the return of the sensor by the building to determine the height of the building; however, this method requires professional lidar equipment and will incur high measurement costs. Summary of the invention

[0003] The present disclosure proposes a building height determination method, a building height determination system, a storage medium and a computer program product.

[0004] According to one aspect of the present disclosure, a method for determining the height of a building is provided, comprising: determining a top prediction frame of a building to be identified in a remote sensing image and multiple bottom candidate frames of the building to be identified; determining a bottom prediction frame of the building to be identified from multiple bottom candidate frames based on geometric features of the top prediction frame and geometric features of the bottom candidate frame; and analyzing a pixel offset between the top prediction frame and the bottom prediction frame to determine the height of the building to be identified, the pixel offset representing a distance in pixel dimension from a center point of the top prediction frame to a center point of the bottom prediction frame, the pixel offset being obtained from the remote sensing image.

[0005] In some embodiments, determining a top prediction box of a building to be identified in a remote sensing image and a plurality of bottom candidate boxes of the building to be identified includes: identifying a top area of ​​the building to be identified, obtaining the top prediction box, and the top prediction box is used to identify a boundary of the top area; and identifying a bottom area of ​​the building to be identified, obtaining a plurality of bottom candidate boxes, the bottom of the building to be identified is covered with an occlusion, and the bottom candidate box is used to identify a boundary of the bottom area covered by the occlusion.

[0006] In some embodiments, according to the geometric features of the top prediction box and the geometric features of the bottom candidate box, determining the bottom prediction box of the building to be identified in multiple bottom candidate boxes includes: calculating the distance between the top prediction box and the bottom candidate box according to the center point pixel coordinates of the top prediction box and the center point pixel coordinates of the bottom candidate box, the geometric features including the center point pixel coordinates; and determining the bottom candidate box corresponding to the minimum distance as the bottom prediction box.

[0007] In some embodiments, before determining that the bottom candidate box corresponding to the minimum spacing is the bottom prediction box, it also includes: when the bottom candidate boxes with the minimum spacing corresponding to multiple top prediction boxes in the remote sensing image are the same, according to the area, shape of the bottom candidate box corresponding to the minimum spacing and the area and shape of each top prediction box, the geometric similarity between the bottom candidate box corresponding to the minimum spacing and each of the top prediction boxes is calculated, and the bottom candidate box with the minimum spacing is used as the bottom prediction box of the top prediction box corresponding to the maximum geometric similarity, and the geometric features include area and shape.

[0008] In some embodiments, after determining the bottom prediction frame of the building to be identified among multiple bottom candidate frames, the method includes: determining a line between the center point of the top prediction frame and the center point of the bottom prediction frame according to the pixel coordinates of the center point of the top prediction frame and the pixel coordinates of the midline point of the bottom prediction frame; and determining whether the angle between the line and the reference direction is within a first angle range; if the angle is within the first angle range, the verification of the bottom prediction frame passes; or, if the angle is not within the first angle range, the verification of the bottom prediction frame fails.

[0009] In some embodiments, before analyzing the pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified, the method includes: training a first height model using the height and pixel offset of known buildings in the remote sensing image to determine that the first height model corresponds to a first coefficient and a second coefficient of the remote sensing image; and using the first height model having the first coefficient and the second coefficient as a height model for analyzing the pixel offset.

[0010] In some implementations, the further step includes: calling the sample set to train the first top model and the first bottom model respectively, to obtain a top model for identifying the top region and a bottom model for identifying the bottom region.

[0011] In some implementations, the method further includes: performing data enhancement processing on the sample image to obtain a plurality of associated images of the sample image; and generating the sample set based on the sample image and the associated images.

[0012] According to another aspect of the present disclosure, a building height determination system is provided, including a processor and a memory, the memory storing a computer program, and when the computer program is executed by the processor, the processor performs the following processes: determining a top prediction box of a building to be identified in a remote sensing image and multiple bottom candidate boxes of the building to be identified; determining a bottom prediction box of the building to be identified from multiple bottom candidate boxes based on geometric features of the top prediction box and geometric features of the bottom candidate box; and analyzing a pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified, the pixel offset representing a pixel dimension distance from a center point of the top prediction box to a center point of the bottom prediction box, and the pixel offset being obtained from the remote sensing image.

[0013] According to another aspect of the present disclosure, a readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the method as described in any of the above embodiments.

[0014] According to another aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, it is used to implement at least the method as described in any one of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings illustrate exemplary embodiments of the present disclosure and together with the description serve to explain the principles of the present disclosure. These drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification.

[0016] Figure 1 It is a schematic diagram of an application scenario of the method for determining the building height according to an embodiment of the present disclosure.

[0017] Figure 2 It is a schematic diagram of the architecture of the method for determining the building height according to an embodiment of the present disclosure.

[0018] Figure 3 It is a schematic diagram of pairing the top prediction box and the bottom prediction box of a building according to an embodiment of the present disclosure.

[0019] Figure 4 is a flow chart of a method for determining building height according to an embodiment of the present disclosure.

[0020] Figure 5 It is a block diagram of a device for determining building height according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0021] The present disclosure is further described in detail below in conjunction with the accompanying drawings and examples. It is understood that the specific examples described herein are only used to explain the relevant content, rather than to limit the present disclosure. It should also be noted that, for ease of description, only the parts related to the present disclosure are shown in the accompanying drawings.

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in the present disclosure can be combined with each other. The technical solution of the present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0023] Identifying building heights can help relevant departments better plan urban space, assess the impact of natural disasters on buildings, and ensure the compliance of new construction projects. The traditional method of calculating building heights through geometric optics relies on measuring the length of building shadows, which is not suitable for urban environments with dense buildings and incomplete shadows. Measuring building heights through methods such as lidar will incur high measurement costs. Therefore, how to determine the height of a building in a relatively universal and low-cost way is an urgent problem to be solved in this field.

[0024] Figure 1 Schematic diagram of an application scenario of a building height determination method according to an embodiment of the present disclosure. Figure 1 As shown, the building height determination method is stored in the server 200 . When the server 200 runs the method, the height of the building to be identified in the remote sensing image 100 can be determined.

[0025] When it is necessary to obtain the height of a building to be identified, a remote sensing image containing the building to be identified is collected by remote sensing equipment, wherein the remote sensing acquisition equipment may be a drone or satellite equipped with sensors such as an optical camera and an imaging spectrometer, which are not listed here one by one. Then, the collected remote sensing image is transmitted to the server 200, and the top model and the bottom model in the server 200 analyze the remote sensing image to determine the top prediction box and the bottom candidate box of the building to be identified. Then, the server executes the relevant steps in the building height determination method, and selects the bottom prediction box that is adapted to the top prediction box from multiple bottom candidate boxes. Then, the pixel offset between the top prediction box and the bottom prediction box is obtained through the remote sensing image, and the pixel offset is analyzed using the height model in the server 200 to determine the height of the building to be identified. Among them, the server 200 can be a cloud server or a local server, which is not limited here.

[0026] The present disclosure only needs to use remote sensing acquisition equipment to obtain remote sensing images containing buildings to be identified, and allows the bottom area of ​​the buildings to be identified to be blocked, which is also feasible for urban environments with dense buildings. By processing the remote sensing images through the building height determination method installed in the server 200, the height of the building to be identified can be inferred, avoiding the investment of expensive radar equipment.

[0027] Figure 2 It is a schematic diagram of the architecture of the method for determining the building height according to an embodiment of the present disclosure.

[0028] The present disclosure proposes a method for determining the height of a building. First, the top prediction box and the bottom prediction box of the building to be identified in the remote sensing image are identified. Then, the height of the building to be identified is inferred based on the pixel offset between the top prediction box and the bottom prediction box in the remote sensing image and the height of the known building in the remote sensing image. Figure 2 The method for determining the building height is explained in detail.

[0029] Before determining the height of the building to be identified, the top model and the bottom model need to be trained first. The essence of the top model and the bottom model is the HRNet (High-Resolution Net) model, which is a deep neural network architecture that focuses on maintaining high-resolution feature maps at all stages of the network. It can capture fine details in remote sensing images and is suitable for identifying details such as buildings, roads, and vegetation in remote sensing images. In addition, HRNet can effectively capture global context information and local details at the same time by processing and fusing feature maps of different scales in parallel. This multi-scale feature fusion strategy greatly enhances the model's understanding and analysis capabilities of complex remote sensing scenes. In addition, HRNet adopts a progressive feature fusion method to ensure that as much information as possible can be retained at each processing stage, improving the accuracy of segmentation and recognition. The introduction of deep supervision further optimizes the network's feature identification at all scales, and the flexible and customizable architecture design allows R&D personnel to adjust the network according to the recognition requirements of the top and bottom areas of the building, making it more suitable for specific remote sensing image analysis tasks. The end-to-end training method ensures the differentiability of the entire process from input to output, allowing HRNet to be efficiently optimized through the standard back-propagation algorithm, thereby achieving high performance in remote sensing image tasks.

[0030] The top model is used to identify the top area of ​​the building in the remote sensing image, and the bottom model is used to identify the bottom area of ​​the building in the remote sensing image. When remote sensing images are taken by remote sensing acquisition equipment such as airplanes, drones, and satellites, the top area of ​​the building is usually exposed, and the top model needs to have the ability to segment the boundaries of the top area; however, the bottom area of ​​the building is relatively complex and may be blocked by other buildings, vegetation, entertainment equipment and other objects, or the bottom boundary may not be obvious due to shooting angle problems. Therefore, in addition to the ability to segment boundaries, the bottom model also needs to have the ability to resist interference from obstructions.

[0031] The two original HRNet models are respectively used as the first top model and the first bottom model, and the sample set is called to train the first top model and the first bottom model respectively. The sample set includes the top sample set and the bottom sample set. The top sample set should include multiple sample images and the actual top boundary of each sample image and the height information of the building. The sample image is used as the processing object of the HRNet model; the actual top boundary delineates the real top area of ​​the building in the sample image, providing a prediction standard for the recognition result of HRNet. Similarly, the bottom sample set also contains multiple sample images and the actual bottom boundary of each sample image and the height information of the building. The sample set can be selected from a map database, which has the location information, the actual top boundary and the height information of the building in the remote sensing image. However, the map database usually lacks the actual top boundary, so the building open source database (such as the BONAI dataset) can be used for data supplement. The BONAI dataset contains multiple aerial remote sensing images of multiple cities, covering a large number of building departures, and each building instance is annotated with detailed boundaries, which can obtain the actual bottom boundary of the building base, and realize the supplement of the actual bottom boundary in the sample set. Of course, other databases that can obtain sample sets can be used as sample data sources, which are not listed here one by one. In addition, the height information of the building can be the actual height or the number of floors, which is not limited here.

[0032] In some embodiments, data cleaning and preprocessing of the sample set are also included. Through cleaning, invalid or incomplete records in the sample set can be removed, such as buildings with no height information or unreliable height information. For buildings that only provide the number of floors, the total height of the building can be estimated by the local standard floor height (for example, 3 meters). After cleaning and preprocessing, the sample set can have enough building samples to cover different building types, heights and geographical locations, and improve the generalization ability of the top model and the bottom model. Preprocessing can be data enhancement processing of the sample image to obtain multiple associated images of the sample image, such as flipping, rotating and translating the sample image to improve the model's recognition ability of buildings under different viewing angles and lighting conditions. Then, the associated image is used as a new sample image, and these sample images and associated images are integrated into a sample set.

[0033] The first top model identifies the top area of ​​the building in the sample image and compares the identification result with the actual boundary of the top. Finally, the parameters of the first top model are adjusted according to the comparison result so that it can output an identification result that is consistent with or similar to the actual boundary of the top, thereby obtaining the top model.

[0034] The first bottom model predicts the bottom area of ​​the building according to the different categories and distribution of occluders in the remote sensing image to obtain a bottom candidate frame; and compares the recognition result with the actual bottom boundary. Finally, the parameters of the first bottom model are adjusted according to the comparison result so that it can output a recognition result that is consistent with or similar to the actual bottom boundary, thereby obtaining a bottom model.

[0035] Of course, the collected real-scene data can also be used to fine-tune the bottom model and the top model. In order to avoid catastrophic forgetting during the fine-tuning process, a smaller learning rate can be used.

[0036] After obtaining the top model and the bottom model, collect the remote sensing image containing the building to be identified, use the top model to identify the top area of ​​the building to be identified in the remote sensing image, obtain the top prediction box, and use the bottom model to predict the bottom area of ​​the building to be identified to obtain multiple bottom candidate boxes. The top prediction box represents the boundary of the top area of ​​the building to be identified, and contains geometric features, such as the pixel coordinates, shape, and area of ​​the center point of the top prediction box. The bottom candidate box is the prediction result of the possible boundary of the base of the building to be identified. Usually, there will be multiple bottom candidate boxes, and each bottom candidate box has geometric features, such as the pixel coordinates, shape, and area of ​​the center point of the bottom candidate box.

[0037] Since there are usually multiple bottom candidate boxes, it is necessary to combine the top prediction box and select the bottom prediction box that matches it from multiple bottom candidate boxes. The following is a detailed introduction to the pairing process between the top prediction box and the bottom candidate box.

[0038] First, calculate the distance between the top prediction box and each bottom candidate box. Calculate the Euclidean distance between the top prediction box and the bottom candidate box through the center point coordinates of the top prediction box and the center point pixel coordinates of the bottom candidate box. The result is the distance between the two. Euclidean distance is the most direct way to measure the straight-line distance between two points, which is suitable for evaluating the spatial proximity of the top prediction box and the bottom candidate box. Select the bottom candidate box corresponding to the minimum distance as the bottom prediction box adapted to the top prediction box.

[0039] There may be multiple buildings to be identified in the remote sensing image, and correspondingly there will be multiple top prediction boxes. Due to occlusion between buildings or recognition errors of prediction boxes, the minimum spacing between multiple top prediction boxes may be the same bottom candidate box. If the bottom candidate box is used as the bottom prediction box, multiple buildings will share the same bottom area, which is obviously unreasonable.

[0040] Therefore, after determining the bottom candidate box corresponding to the minimum spacing, it should also be determined whether the bottom candidate box is also associated with other top prediction boxes. If the bottom candidate box is not used as the bottom candidate box corresponding to the minimum spacing of the remaining top prediction boxes, then the bottom candidate box is the bottom prediction box of the corresponding top prediction box. On the contrary, according to the shape, area of ​​the bottom candidate box and the shape and area of ​​each associated top prediction box, the geometric similarity between the bottom candidate box and each top prediction box is calculated, and it is used as the bottom prediction box of the top prediction box corresponding to the maximum geometric similarity. The method for determining the geometric similarity can be selected from Jaccard coefficient or Z-score (standard score), etc., which is not limited here. For example, if the bottom candidate box corresponds to top prediction box 1 and top prediction box 2 at the same time, then according to the area and shape, the similarity between the bottom candidate box and top prediction box 1 and the similarity between the bottom candidate box and top prediction box 2 are determined. If the similarity between the bottom candidate box and top prediction box 1 is greater than the similarity between it and top prediction box 2, then the bottom candidate box is considered to be the bottom prediction box of top prediction box 1.

[0041] In some embodiments, after determining the bottom prediction frame of a top prediction frame, it is also necessary to verify the bottom prediction frame to eliminate the problem of inaccurate pairing due to occlusion, top prediction frame or bottom prediction frame recognition errors, etc. Specifically, based on the pixel coordinates of the center point of the top prediction frame and the pixel coordinates of the midline point of the bottom prediction frame, determine the line between the center point of the top prediction frame and the center point of the bottom prediction frame; and determine whether the angle between the line and the reference direction is within the first angle range; if the angle is within the first angle range, the verification of the bottom prediction frame passes; or, if the angle is not within the first angle range, the verification of the bottom prediction frame fails.

[0042] Specifically, the remote sensing image is an image collected by remote sensing acquisition equipment containing multiple reference objects and at least one building to be identified, and the reference objects include but are not limited to known buildings, known street lamps and other objects with known heights and bottom areas. Each object in the same remote sensing image has the same shooting angle; and, at the same shooting angle, the angle between the top center point of these objects perpendicular to the ground and the center point of the surface coincident with the ground should be the same relative to the same reference direction. Therefore, the present disclosure uses the angle between the known object and the reference direction in the remote sensing image as the first angle, the first angle as the standard value of the angle between the corresponding line of the building to be identified and the reference direction, and on the basis of the first angle, sets the fault tolerance range to form a first angle range. That is to say, when the aforementioned angle is in the first angle range, the inclination of the bottom candidate frame corresponding to the angle relative to the reference direction in the remote sensing image conforms to the shooting angle of the remote sensing image; the bottom candidate frame whose corresponding angle is not in the first angle range, whose inclination relative to the reference direction in the remote sensing image does not conform to the shooting angle of the remote sensing image, is eliminated.

[0043] Figure 3 : is a schematic diagram of the pairing of the top prediction box and the bottom prediction box of a building according to an embodiment of the present disclosure. Figure 3 The above pairing method is described.

[0044] Assume that buildings A and B are the buildings to be identified, building C is a known building, t is the top prediction box of building B, the reference direction is d, b1, b2 and b3 are the bottom candidate boxes respectively, α is the angle between building B and the reference direction d, and β is the angle between building C and the reference direction d.

[0045] If you want to match the corresponding bottom prediction box for the top prediction box t of building B, first calculate the distances between the pixel coordinates of the center point of the top prediction box t and the bottom candidate boxes b1, b2 and b3 respectively. Assuming that the top prediction box t has the minimum distance with the bottom candidate box b1, then determine whether the bottom candidate box b1 is also the bottom candidate box with the minimum distance of the other top prediction boxes. If not, the bottom candidate box b1 is the bottom prediction box corresponding to the top prediction box t. Assuming that the bottom candidate box b1 is also the bottom candidate box with the minimum distance of building A, then according to the shape and area of ​​the bottom candidate box b1, the shape and area of ​​the top prediction box of building A, and the shape and area of ​​the top prediction box t, determine the geometric similarity between the bottom candidate box b1 and the top prediction box of building A, and the geometric similarity between it and the top prediction box t. If the geometric similarity between the bottom candidate box b1 and the top prediction box t is greater than the geometric similarity between it and the top prediction box of building A, then the bottom candidate box b1 is the bottom prediction box corresponding to the top prediction box t.

[0046] After determining the bottom prediction frame of building B, it is determined whether the angle between the line between the center point of b1 and the center point of t and the reference direction d is within the first angle range. Since building C is a building with known height and bottom area, the pixel coordinates of the top center point and the bottom center point of building C are used to determine the angle β between the line between its top area and bottom area and d, and β is used as the first angle. And according to the preset fault tolerance range, such as ±5°, the first angle range is determined on the basis of β. Of course, the fault tolerance range can also use the maximum and minimum values ​​of the angles between all known objects in the remote sensing image and the reference direction d as the upper and lower limits, and the first angle can also be the mean of the angles between all known objects in the remote sensing image and the reference direction d. The way to determine the first angle and the first angle range is not exhausted here. Further, the angle between the midline point of b1 and the line between the center point of t and the reference direction d is calculated, and it is determined whether it is within the first angle range. If it is within the first angle range, it proves that the inclination of b1 relative to the reference direction d meets the shooting angle of the remote sensing image, and the verification is passed; otherwise, the verification fails.

[0047] After determining the bottom prediction box of the building to be identified, the height model is used to analyze the pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified. The pixel offset can be read from the remote sensing image, which represents the distance from the center point of the top prediction box to the center point of the bottom prediction box in the pixel dimension.

[0048] Specifically, the first height model is trained using the height and pixel offset of known buildings in the remote sensing image, and the first coefficient and second coefficient of the first height model corresponding to the remote sensing image are determined; and the first height model with the first coefficient and the second coefficient is used as the height model. The height model can characterize the relationship between the pixel offset between the top area and the bottom area of ​​each object in the remote sensing image and the height of the object. The relationship can be represented by Y=aX+b, where a represents the first coefficient, b represents the second coefficient, X represents the pixel offset, and Y represents the height corresponding to the pixel offset. Each remote sensing image contains known buildings, so the values ​​of a and b in the remote sensing image can be determined based on the pixel offsets and heights of multiple known buildings. The height model can be a linear regression or other appropriate machine learning algorithm, which is used to establish the relationship between the pixel offsets of each remote sensing image and the height of the object.

[0049] Figure 4 FIG. 1 is a flow chart of a method for determining building height according to an embodiment of the present disclosure. Figure 4 As shown, the present disclosure proposes a building height determination method M400 for implementing the aforementioned technical solution, including: step S410, determining a top prediction frame of a building to be identified and multiple bottom candidate frames of the building to be identified in a remote sensing image; step S420, determining a bottom prediction frame of the building to be identified from multiple bottom candidate frames based on geometric features of the top prediction frame and geometric features of the bottom candidate frame; and step S430, analyzing the pixel offset between the top prediction frame and the bottom prediction frame to determine the height of the building to be identified, the pixel offset representing the pixel dimension distance from the center point of the top prediction frame to the center point of the bottom prediction frame, and the pixel offset is obtained from the remote sensing image.

[0050] This method overcomes the dependence on the length of the building's shadow and the high cost of equipment investment through the recognition and matching mechanism of the top prediction box and the bottom prediction box and the introduction of the height model; and by verifying the top prediction box, the accuracy of the top and bottom matching of the building to be identified is improved.

[0051] The present disclosure also proposes a building height determination system, including a processor and a memory, the memory storing a computer program, when the computer program is executed by the processor, the processor performs the following processes: determining a top prediction box of a building to be identified and multiple bottom candidate boxes of the building to be identified in a remote sensing image; determining a bottom prediction box of the building to be identified from multiple bottom candidate boxes based on geometric features of the top prediction box and geometric features of the bottom candidate box; and analyzing a pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified, the pixel offset representing a pixel dimension distance from a center point of the top prediction box to a center point of the bottom prediction box, the pixel offset being obtained from the remote sensing image.

[0052] In some embodiments, determining a top prediction box of a building to be identified in a remote sensing image and multiple bottom candidate boxes of the building to be identified includes: identifying a top area of ​​the building to be identified, obtaining a top prediction box, and the top prediction box is used to identify a boundary of the top area; and identifying a bottom area of ​​the building to be identified, obtaining multiple bottom candidate boxes, the bottom of the building to be identified is covered with an occlusion, and the bottom candidate box is used to identify the boundary of the bottom area covered by the occlusion.

[0053] In some embodiments, based on the geometric features of the top prediction box and the geometric features of the bottom candidate box, a bottom prediction box of the building to be identified is determined from multiple bottom candidate boxes, including: calculating the spacing between the top prediction box and the bottom candidate box based on the center point pixel coordinates of the top prediction box and the center point pixel coordinates of the bottom candidate box, the geometric features including the center point pixel coordinates; and determining the bottom candidate box corresponding to the minimum spacing as the bottom prediction box.

[0054] In some embodiments, before determining that the bottom candidate box corresponding to the minimum spacing is the bottom prediction box, it also includes: when the bottom candidate boxes with the minimum spacing corresponding to multiple top prediction boxes in the remote sensing image are the same, the geometric similarity between the bottom candidate box corresponding to the minimum spacing and each top prediction box is calculated based on the area and shape of the bottom candidate box corresponding to the minimum spacing and the area and shape of each top prediction box, and the bottom candidate box with the minimum spacing is used as the bottom prediction box of the top prediction box corresponding to the maximum geometric similarity, and the geometric features include area and shape.

[0055] In some embodiments, after determining the bottom prediction frame of the building to be identified from multiple bottom candidate frames, the method includes: determining a line between the center point of the top prediction frame and the center point of the bottom prediction frame according to the pixel coordinates of the center point of the top prediction frame and the pixel coordinates of the midline point of the bottom prediction frame; and determining whether the angle between the line and the reference direction is within a first angle range; if the angle is within the first angle range, the bottom prediction frame is verified to pass; or, if the angle is not within the first angle range, the bottom prediction frame is verified to fail.

[0056] In some embodiments, before analyzing the pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified, the method includes: training a first height model using the height and pixel offset of known buildings in the remote sensing image to determine the first coefficient and the second coefficient of the first height model corresponding to the remote sensing image; and using the first height model having the first coefficient and the second coefficient as a height model for analyzing the pixel offset.

[0057] In some implementations, the further step includes: calling the sample set to train the first top model and the first bottom model respectively, to obtain a top model for identifying the top region and a bottom model for identifying the bottom region.

[0058] In some implementations, the method further includes: performing data enhancement processing on the sample image to obtain a plurality of associated images of the sample image; and generating a sample set based on the sample image and the associated images.

[0059] This system overcomes the dependence on the length of the building's shadow and the high cost of equipment investment through the recognition and matching mechanism of the top prediction box and the bottom prediction box and the introduction of the height model; and by verifying the top prediction box, the accuracy of the top and bottom matching of the building to be identified is improved.

[0060] Figure 5 is a block diagram of a device for determining building height according to an embodiment of the present disclosure. Figure 5 As shown, the present disclosure proposes a device 1000 for determining the height of a building, including: a recognition module 1002, used to determine a top prediction frame of a building to be identified and multiple bottom candidate frames of the building to be identified in a remote sensing image; a pairing module 1004, used to determine a bottom prediction frame of the building to be identified from multiple bottom candidate frames according to geometric features of the top prediction frame and geometric features of the bottom candidate frame; and a calculation module 1006, used to analyze the pixel offset between the top prediction frame and the bottom prediction frame to determine the height of the building to be identified, the pixel offset representing the pixel dimension distance from the center point of the top prediction frame to the center point of the bottom prediction frame, and the pixel offset is obtained from the remote sensing image.

[0061] The various modules in the building height determination device 1000 are provided to implement the various steps of the building height determination method, and the implementation principles and steps thereof can be referred to above and will not be described in detail. The building height determination device 1000 also includes a processor 1200, a memory 1300 and other circuits 1400 necessary to support the operation of the aforementioned modules.

[0062] The present disclosure also provides a readable storage medium, in which execution instructions are stored, and when the execution instructions are executed by a processor, they are used to implement the above method. "Readable storage medium" can be any device that can contain, store, communicate, propagate or transmit programs for use in an instruction execution system, device or equipment or in combination with these instruction execution systems, devices or equipment. More specific examples of readable storage media include the following: an electrical connection portion (electronic device) with one or more wirings, a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable read-only memory (CDROM), etc.

[0063] The present disclosure also provides a computer program product, and the method of the present disclosure can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed, the process or function of the present disclosure is executed in whole or in part. A computer-readable storage medium can be used as an implementation of a computer program product, that is, the present disclosure also provides a computer-readable storage medium, which includes a computer program, and when the computer program is executed by a processor, any of the above methods is implemented.

[0064] The computer program or instructions may be stored in a readable storage medium or transmitted from one readable storage medium to another readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The readable storage medium may be any available medium that can be accessed or a data storage device such as a server, data center, etc. that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid state drive. The computer readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0065] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0067] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0068] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0069] In the description of this specification, the description with reference to the terms "one embodiment / method", "some embodiments / methods", "example", "specific example", or "some examples" etc. means that the specific features, structures, or characteristics described in conjunction with the embodiment / method or example are included in at least one embodiment / method or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment / method or example. Moreover, the specific features, structures, or characteristics described may be combined in any one or more embodiments / methods or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments / methods or examples described in this specification and features of different embodiments / methods or examples, unless they are contradictory.

[0070] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, "plurality" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0071] Those skilled in the art should understand that the above embodiments are only for the purpose of clearly illustrating the present disclosure, and are not intended to limit the scope of the present disclosure. For those skilled in the art, other changes or modifications may be made based on the above disclosure, and these changes or modifications are still within the scope of the present disclosure.

Claims

1. A method for determining building height, characterized in that: include: Determine a top prediction frame of a building to be identified in a remote sensing image and multiple bottom candidate frames of the building to be identified; Determining a bottom prediction frame of the building to be identified from a plurality of bottom candidate frames according to the geometric features of the top prediction frame and the geometric features of the bottom candidate frame; as well as The pixel offset between the top prediction box and the bottom prediction box is analyzed to determine the height of the building to be identified, the pixel offset represents the pixel dimension distance from the center point of the top prediction box to the center point of the bottom prediction box, and the pixel offset is obtained from the remote sensing image.

2. The method for determining building height according to claim 1, characterized in that: Determining a top prediction box of a building to be identified in a remote sensing image and a plurality of bottom candidate boxes of the building to be identified, comprising: Identify the top area of ​​the building to be identified, and obtain the top prediction box, where the top prediction box is used to identify the boundary of the top area; and The bottom area of ​​the building to be identified is identified to obtain a plurality of bottom candidate frames, wherein the bottom of the building to be identified is covered by an occlusion, and the bottom candidate frames are used to identify the boundary of the bottom area covered by the occlusion.

3. The method for determining building height according to claim 1, characterized in that: Determining a bottom prediction frame of the to-be-identified building from a plurality of bottom candidate frames according to the geometric features of the top prediction frame and the geometric features of the bottom candidate frame, comprising: Calculating a distance between the top prediction box and the bottom candidate box according to the center point pixel coordinates of the top prediction box and the center point pixel coordinates of the bottom candidate box, wherein the geometric features include the center point pixel coordinates; and Determine the bottom candidate box corresponding to the minimum distance as the bottom prediction box.

4. The method for determining building height according to claim 3, characterized in that: Before determining the bottom candidate box corresponding to the minimum spacing as the bottom prediction box, the method further includes: When the bottom candidate frames with the minimum spacing corresponding to multiple top prediction frames in the remote sensing image are the same, the geometric similarity between the bottom candidate frame corresponding to the minimum spacing and each of the top prediction frames is calculated according to the area and shape of the bottom candidate frame corresponding to the minimum spacing and the area and shape of each top prediction frame, and the bottom candidate frame with the minimum spacing is used as the bottom prediction frame of the top prediction frame corresponding to the maximum geometric similarity, and the geometric features include area and shape.

5. The method for determining building height according to claim 1, characterized in that: After determining the bottom prediction frame of the building to be identified from the multiple bottom candidate frames, the method includes: Determining a line between the center point of the top prediction box and the center point of the bottom prediction box according to the pixel coordinates of the center point of the top prediction box and the pixel coordinates of the midline point of the bottom prediction box; and Determine whether the angle between the connecting line and the reference direction is within a first angle range; if the angle is within the first angle range, the check on the bottom prediction box passes; or, if the angle is not within the first angle range, the check on the bottom prediction box fails.

6. The method for determining building height according to claim 1, characterized in that: Before analyzing the pixel offset between the top prediction box and the bottom prediction box to determine the height of the building to be identified, the method includes: Training a first height model using heights and pixel offsets of known buildings in the remote sensing image to determine first coefficients and second coefficients of the first height model corresponding to the remote sensing image; and A first height model having a first coefficient and a second coefficient is used as a height model for analyzing the pixel offset.

7. The method for determining building height according to claim 2, characterized in that: Also includes: The sample set is called to train the first top model and the first bottom model respectively, so as to obtain a top model for identifying the top area and a bottom model for identifying the bottom area.

8. The method for determining building height according to claim 7, characterized in that: Also includes: Performing data enhancement processing on the sample image to obtain multiple associated images of the sample image; as well as The sample set is generated according to the sample image and the associated image.

9. A building height determination system, characterized in that: The invention comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following process: Determine a top prediction frame of a building to be identified in a remote sensing image and multiple bottom candidate frames of the building to be identified; Determining a bottom prediction frame of the building to be identified from a plurality of bottom candidate frames according to the geometric features of the top prediction frame and the geometric features of the bottom candidate frame; as well as The pixel offset between the top prediction box and the bottom prediction box is analyzed to determine the height of the building to be identified, the pixel offset represents the pixel dimension distance from the center point of the top prediction box to the center point of the bottom prediction box, and the pixel offset is obtained from the remote sensing image.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program is used to implement the method according to any one of claims 1 to 8.

11. A computer program product, characterized in that The computer program product comprises a computer program, which is used to implement at least the method according to any one of claims 1 to 8 when executed by a processor.

Images