Building change detection device, method, and program

Abstract

To provide a building change detection device, method, and program that can predict current or future vacant houses easily and at low cost.SOLUTION: In a construction age determination system 1, a building change detection device 10 includes: an image division unit 11 configured to divide two-dimensional building images of multiple buildings taken from the sky into preset building units to generate multiple building unit images; a high-resolution processing unit 12 configured to perform high-resolution processing on the building unit images to generate high-resolution images; a change detection unit 13 configured to compare high-resolution images including the same building generated from building images taken at different timings in time to detect information on changes in the building; and a construction age determination unit 14 configured to determine the construction age of the building based on the detected information on changes in the building.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a building change detection device, method, and program for detecting information about changes in a building based on two-dimensional building images taken from above the building. [Background technology]

[0002] In recent years, it has become important to know the age of buildings. If you know the age of a building, it becomes easier to determine which buildings are at risk of earthquakes, and it becomes easier to estimate damage such as building collapse or fire in the event of a disaster. In addition, the older a building is, the more likely it is to become an insufficiently vacant house or a specific vacant house. Identifying old buildings with outdated earthquake resistance can be used to deal with vacant houses, allowing for appropriate management and efficient publicity.

[0003] If vacant house management of buildings can be done, it can be used for formulating construction plans for surrounding infrastructure, etc. Vacant house management is also very important in the sense that it can prevent the occurrence of poorly managed vacant houses and specific vacant houses. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2021-157634 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, at present, there is no database that accumulates information on the construction ages of the above-mentioned buildings, making it extremely difficult to determine the construction ages of the buildings. Moreover, creating a database like the one mentioned above would be extremely costly. In addition, building information in fixed asset registers managed by local governments and other organizations is basically personal information, so it is difficult to obtain. Furthermore, there are unregistered buildings, and information on such buildings cannot be obtained.

[0006] In addition, to get a comprehensive and wide-area understanding of the age of buildings, including their location, it is possible to read them from either old maps or aerial photographs.

[0007] However, due to the scale and handwriting of many maps, there is some misalignment or distortion with respect to actual buildings, and although it is possible to grasp the general location, it lacks accuracy. Aerial photographs have relatively little misalignment or distortion, and can be considered useful information that has been compiled nationwide. However, they tend to be low resolution.

[0008] Patent Document 1 proposes a technology to determine the type of vacant house based on energy consumption measurement history information and generate a vacancy rate map showing the vacancy rate for each specified area on a map, but does not propose any technology to determine the construction age of the building.

[0009] In view of the above problems, the present invention aims to provide a building change detection device, method, and program that can determine the construction age of a building simply and at low cost. [Means for solving the problem]

[0010] The building change detection device of the present invention comprises an image division unit that divides two-dimensional building images of multiple buildings photographed from above into predetermined building units to generate multiple building unit images, a high-resolution processing unit that applies high-resolution processing to the building unit images, and a change detection unit that compares building unit images of the same location photographed at different times and that have been subjected to high-resolution processing to detect information about changes in the buildings. Effect of the Invention

[0011] According to the building change detection device of the present invention, two-dimensional building images of multiple buildings photographed from above are divided into predetermined building units to generate multiple building unit images, high-resolution processing is applied to the building unit images, and information on changes in the buildings is detected by comparing building unit images of the same location that were photographed at different times and that have been subjected to high-resolution processing.Therefore, the information on changes in the buildings can be used to easily and low-costly determine the construction age of the buildings.

[0012] In addition, the building change detection device of the present invention performs high-resolution processing, so even if the building image is an old image with low resolution, it is possible to detect information about changes in the building with high accuracy. In other words, it is possible to determine the construction age of a building even from an old building image. [Brief description of the drawings]

[0013] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a building age determination system using an embodiment of a building change detection device of the present invention. [Diagram 2] A diagram for explaining how to determine the size of a building unit image [Diagram 3] A flowchart to explain how to generate a trained model used to detect changes in buildings. [Figure 4] A flowchart illustrating the process flow of the construction age determination system shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] A construction age determination system using an embodiment of a building change detection device of the present invention will be described in detail below with reference to the drawings. Fig. 1 is a schematic configuration diagram of a construction age determination system 1 of this embodiment.

[0015] 1, the construction age determination system 1 of this embodiment includes a building change detection device 10, a display device 20, and an input device 30. The building change detection device 10 and the display device 20, and the building change detection device 10 and the input device 30 are connected so as to be able to communicate with each other.

[0016] The building change detection device 10 includes an image division unit 11, a high resolution processing unit 12, a change detection unit 13, and a building age determination unit 14.

[0017] The image division unit 11 acquires two-dimensional building images of multiple buildings photographed from above, and divides the acquired building images into preset building units to generate multiple building unit images. The image division unit 11 of this embodiment acquires multiple building images photographed at different times in time as building images. The image division unit 11 acquires building images of different eras, such as building images photographed in the 1980s and building images photographed in 2022.

[0018] Then, the image division unit 11 of this embodiment recognizes individual buildings based on the building image, and determines the size of the building unit image based on the size of the recognized building.

[0019] Specifically, the image segmentation unit 11 acquires polygon data of a plurality of buildings generated from an aerial image of the same location, for example, and recognizes each building from the polygon data. The polygon data may be any known data.

[0020] Then, the image division unit 11 determines the sizes W3 and W4 of each building, which are 1.5 times the sizes W1 and W2, as shown in Fig. 2, and sets these as the sizes of the building unit images. That is, the size of each building unit image is determined to be larger than the size of each building. Note that when adjacent buildings are relatively close to each other, the building unit images of the adjacent buildings will partially overlap each other.

[0021] The high-resolution processing unit 12 performs high-resolution processing on each divided building unit image. For example, aerial images and satellite images are used as building images, but images of buildings taken in the past, such as in the 1980s, are low-resolution building images because they were taken with different equipment than those used today, and there is a risk that the detection accuracy of information on changes in buildings, which will be described later, will decrease. Therefore, in this embodiment, high-resolution processing is performed on the building unit images to improve the detection accuracy of information on changes in buildings.

[0022] For the high-resolution processing, it is preferable to use a super-resolution technology, such as SwinIR or Swin2SR. For example, a building unit image of 32 x 32 pixels is made into a 64 x 64 pixel image, which is four times the resolution. However, the high-resolution processing is not limited to this, and other processing may be used.

[0023] The change detection unit 13 detects information about changes in buildings by comparing building unit images of the same location that have been captured at different times and have been subjected to high resolution processing.

[0024] The change detection unit 13 of this embodiment first performs a standardization process on each building unit image. The standardization process in this embodiment is a process of converting a dark building unit image or an overly bright building unit image into a building unit image with the same brightness as normal. Specifically, the change detection unit 13 of this embodiment performs a standardization process on each building unit image such that the average pixel value is zero and the variance is one.

[0025] Next, the change detection unit 13 inputs two building-unit images of the same location that were taken at different times and that have been subjected to high-resolution processing into a trained model obtained in advance by machine learning a plurality of images including buildings, and detects information on changes in buildings by comparing the feature amounts of each building-unit image. Specifically, the change detection unit 13 calculates the probability of change in the feature amounts of each building-unit image as information on changes in buildings.

[0026] As the machine learning technique, for example, ResNet is preferably used. Fig. 3 is a flowchart for explaining a method for generating a trained model according to the present embodiment.

[0027] As shown in Fig. 3, when generating a trained model, a building image is divided to generate building unit images (S10) in the same manner as in the case of calculating the change probability described above. Then, after performing a high-resolution process on the building unit images (S12), a standardization process is performed (S14). Next, the number of building unit images used for machine learning is increased by performing data expansion processing on the building unit images that have been subjected to the high-resolution processing and standardization processing (S16). Specifically, for example, the building unit images are rotated (for example, by 90°) or a part of the building unit images is cut out. The part of the building unit images is cut out, for example, by cutting out the periphery of the building unit images. It is desirable to determine multiple widths for the periphery to be cut out by randomly changing the number of pixels, for example, and generate multiple building unit images with the peripheries of various widths cut out.

[0028] Then, machine learning is performed using multiple building unit images that have been taken at different points in time and have undergone data augmentation processing, and a trained model that outputs the above-mentioned change probability is generated (S18).

[0029] In addition, when generating a trained model, it is preferable to train the model on images without buildings, including fields and rice paddies, in addition to building-unit images that contain buildings, so that it can detect the construction of a building where there was nothing before.

[0030] The trained model may be generated in the change detection unit 13, or a trained model generated in advance may be stored. The trained model may be stored in an external device different from the building change detection device 10 and used.

[0031] The construction age determination unit 14 determines the construction age of a building based on information on changes in the building detected by the change detection unit 13. In this embodiment, the construction age determination unit 14 identifies two building unit images (building images) for which the change probability detected by the change detection unit 13 is equal to or greater than a preset threshold, and identifies the year in which the two building unit images (building images) were photographed. Each building image is associated with a photography year, which is acquired and stored by the image division unit 11 together with each building image.

[0032] Then, the construction age determination unit 14 determines the construction ages of the buildings included in the two building unit images based on the shooting years of the two building unit images. Specifically, for example, if the shooting year of one building unit image is 2023 and the shooting year of the other building unit image is 2020, the construction age determination unit 14 determines that the construction age is between 2020 and 2023.

[0033] For example, vacant house prediction may be performed using the construction age determined by the construction age determination unit 14. Specifically, if the construction age is older than a preset threshold from the current age, the building included in the building unit image may be predicted to be vacant. Specifically, when the current age is 2023, if the construction age determined to be 1993, which is more than 30 years ago, the construction age determination unit 14 may predict that the building included in the building unit image is currently vacant.

[0034] Also, if the building is not older than a preset threshold from the current age, i.e., if the building is relatively new, the building included in the building-unit image may be predicted to become vacant in an age obtained by adding a predetermined number of years, such as 40 years or more, to the building's age. Specifically, if the determined building age is 2000, the building included in the building-unit image may be predicted to become vacant in 2040.

[0035] In this embodiment, the construction age determination unit 14 determines the construction age of a building based on the years when two building unit images (building images) with a large probability of change in the feature amount were captured. However, in another embodiment, for example, the change detection unit 13 may detect information on changes in a building due to changes in the state of vegetation overgrowth within the building site. In this case, a trained model is generated using multiple building unit images with different states of vegetation overgrowth, and two building unit images of the same location are input to this trained model, so that it is possible to output whether or not the state of the building has changed due to overgrowth of vegetation (for example, whether or not vegetation on the building site is overgrowth and covering the building). In this way, the construction age of a building due to overgrowth of vegetation (poorly managed vacant lot) can be detected. By detecting change information about the occupancy rate of a building (house), it is possible to predict whether the building is vacant or not, as described above, and to manage vacant houses.

[0036] The building change detection device 10 is composed of a computer and has a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), semiconductor memory such as ROM (Read Only Memory) and RAM (Random Access Memory), storage such as a hard disk, and a communication I / F (Interface).

[0037] An embodiment of a building change detection program of the present invention is installed in the storage of the building change detection device 10. When this building change detection program is started by a CPU or GPU included in the building change detection device 10, the functions of the above-mentioned parts of the building change detection device 10 are executed.

[0038] In addition, in this embodiment, the functions of each of the above-mentioned parts are executed by executing the building change detection program using a CPU or GPU, but some or all of the functions executed by the building change detection program may be configured using hardware such as an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or other electrical circuits.

[0039] In addition, in this embodiment, the change detection unit 13 detects information about changes in a building using a trained model, but the high resolution processing (super-resolution processing) in the high resolution processing unit 12 and the building change detection processing in the change detection unit 13 may be performed using a single trained model.

[0040] The display device 20 has a liquid crystal display or the like, and displays information on the construction age of each building determined by the building change detection device 10 as described above, and predicted vacant houses.

[0041] The input device 30 includes a keyboard, a mouse, and the like, and the user uses the input device 30 to give instructions for inputting building images, and the like.

[0042] Next, a description will be given of the flow of processing in the construction age determination system 1 of this embodiment. FIG.

[0043] First, two-dimensional building images of multiple buildings taken from above at different points in time are acquired (S30), and the acquired building images are divided to generate multiple building unit images (S32).

[0044] Next, each divided building unit image is subjected to a high-resolution process (S34) and then a standardization process (S36). Then, each of the building unit images including buildings in the same location generated from building images taken at different times in time is input into the trained model, and the probability of change in the features of each building unit image output from the trained model is calculated (S38).

[0045] Next, two building unit images with the calculated change probability equal to or greater than a preset threshold are identified, the years in which the two building unit images were photographed are identified, and the construction ages of the buildings included in the two building unit images are determined based on the years in which they were photographed (S40).

[0046] Then, information on the construction age determined for each building is output to and displayed on the display device 20 (S42).

[0047] According to the above embodiment of the construction age determination system 1, a two-dimensional building image is divided into predetermined building units to generate multiple building unit images, a high-resolution process is applied to the building unit images, and information on changes in the building is detected by comparing building unit images containing the same building generated from building images taken at different times in time.Therefore, the construction age of the building can be determined easily and at low cost using the information on changes in the building.

[0048] Furthermore, because a high-resolution process is applied to the building unit images, even if the building images are old and have low resolution, information on changes in the buildings can be detected with high accuracy.

[0049] In the construction age determination system 1 of the above embodiment, after dividing the building image into building unit images, the high-resolution processing is performed on each building unit image, but the order of the division processing and the high-resolution processing may be reversed. In other words, after the high-resolution processing is performed on the building image, the high-resolution-processed building image may be divided to generate the building unit images.

[0050] Furthermore, in the above embodiment of the building age determination system 1, information on changes to a building is detected using a trained model obtained by prior machine learning of multiple images including a building, so that information on changes to a building can be detected with high accuracy using a simple method.

[0051] Furthermore, in the construction age determination system 1 of the above embodiment, the size of the building unit image is determined based on the size of the building included in the building image, so that information on changes in the building can be detected with high accuracy.

[0052] Specifically, in a building image taken from the sky, the central part is photographed almost directly from above, so the shapes of the building and road are almost accurately represented, but for example, in the four corners of the building image, the building and road are photographed from an oblique angle, so the size of the building and the width of the road are not accurate. If such a building image is simply divided into appropriate sizes, it may not be possible to generate building unit images that include the entire building, and the accuracy of detecting information about changes in the building may decrease. Also, the accuracy of detecting information about changes in the building may be improved by including not only the building itself but also its surroundings.

[0053] Therefore, in this embodiment, the size of the building unit image is determined based on the size of the building contained in the building image. Therefore, for example, by setting a larger size, a building unit image can be generated that reliably includes the entire building and its surroundings, thereby improving the accuracy of detecting information about changes in the building.

[0054] The present invention is not limited to the above-described embodiment, and the components can be modified and embodied in the implementation stage without departing from the gist of the invention. In addition, various inventions can be formed by appropriately combining the multiple components disclosed in the above-described embodiment. For example, all the components shown in the embodiment can be appropriately combined. Of course, various modifications and applications are possible without departing from the spirit of the invention.

[0055] The present invention further discloses the following supplementary notes.

[0056] (Appendix 1) The building change detection device of the present invention comprises an image division unit that divides two-dimensional building images of multiple buildings photographed from above into preset building units to generate multiple building unit images, a high-resolution processing unit that applies high-resolution processing to the building unit images, and a change detection unit that compares building unit images of the same location photographed at different times and that have been subjected to high-resolution processing to detect information about changes in the buildings.

[0057] (Appendix 2) The building change detection device of the present invention comprises a high-resolution processing unit that applies high-resolution processing to two-dimensional building images of multiple buildings photographed from above, an image division unit that divides the high-resolution processed building images into predetermined building units to generate multiple building unit images, and a change detection unit that compares building unit images of the same location photographed at different times and subjected to high-resolution processing to detect information about changes in the buildings.

[0058] (Appendix 3) In the building change detection device described in Appendix 1, the change detection unit can detect information about changes in buildings by inputting multiple building-unit images to a trained model obtained by previously performing machine learning on multiple images including buildings, and comparing the features of each building-unit image.

[0059] (Appendix 4) In the building detection device according to Supplementary Note 1 or 2, the image dividing section can determine, as the size of the building unit image, a size larger than the size of the buildings included in the building image.

[0060] (Appendix 5) The building detection device according to any one of Supplementary Note 1 to Supplementary Note 3 may include a construction age determination unit that determines the construction age of the building being photographed based on the result of the change detection unit.

[0061] (Appendix 6) The building change detection method of the present invention divides two-dimensional building images of multiple buildings photographed from above into predetermined building units to generate multiple building unit images, applies high-resolution processing to the building unit images, and detects information about changes in buildings by comparing building unit images of the same location that were photographed at different times and that have been subjected to high-resolution processing.

[0062] (Appendix 7) The building change detection method of the present invention applies high-resolution processing to two-dimensional building images of multiple buildings photographed from above, divides the high-resolution processed building images into pre-set building units to generate multiple building unit images, and detects information about changes in buildings by comparing the building unit images of the same location photographed at different times and subjected to high-resolution processing.

[0063] (Appendix 8) The building change detection program of the present invention causes a computer to execute the steps of dividing two-dimensional building images of multiple buildings photographed from above into predetermined building units to generate multiple building unit images, applying high-resolution processing to the building unit images, and comparing the building unit images photographed at different times and subjected to high-resolution processing to detect information about changes in the buildings.

[0064] (Appendix 9) The building change detection program of the present invention causes a computer to execute the steps of: applying high-resolution processing to two-dimensional building images of multiple buildings photographed from above; dividing the high-resolution processed building images into pre-set building units to generate multiple building unit images; and comparing building unit images of the same location photographed at different times and subjected to high-resolution processing to detect information about changes in the buildings. [Explanation of symbols]

[0065] 1. Building Age Determination System 10. Building change detection device 11 Image division section 12 High-resolution processing section 13 Change detection section 14 Building age determination department 20 Display device 30 Input Devices

Claims

1. an image division unit that divides a two-dimensional building image obtained by photographing a plurality of buildings from above into a plurality of building unit images set in advance; a high-resolution processing unit that performs a high-resolution processing on the building unit images; A building change detection device comprising: a change detection unit that compares building unit images of the same location that were taken at different times in time and that have been subjected to the high-resolution processing, and detects information about changes in the building.

2. A high-resolution processing unit that performs high-resolution processing on two-dimensional building images obtained by photographing a plurality of buildings from above; an image division unit that divides the building image subjected to the high resolution processing into a plurality of building unit images set in advance; A building change detection device comprising: a change detection unit that compares building unit images of the same location that were taken at different times in time and that have been subjected to the high-resolution processing, and detects information about changes in the building.

3. The building change detection device according to claim 1 or 2, wherein the change detection unit inputs a plurality of building unit images to a trained model obtained by prior machine learning of a plurality of images including buildings, and detects information on changes in the buildings by comparing the features of each of the building unit images.

4. 3. The building change detection device according to claim 1, wherein the image dividing section determines, as the size of the building unit image, a size larger than the size of the buildings included in the building image.

5. 3. A building change detection device according to claim 1, further comprising a construction age determination section for determining the construction age of the building being photographed based on the result of said change detection section.

6. A two-dimensional building image of a plurality of buildings photographed from above is divided into a plurality of building unit images set in advance, A high-resolution process is performed on the building unit images; A building change detection method comprising: comparing the building unit images of the same location that were taken at different times in time and that have been subjected to the high resolution processing, and detecting information about changes in the building.

7. Two-dimensional building images of multiple buildings taken from the sky are processed to increase the resolution, Dividing the building image subjected to the high resolution processing into a predetermined building unit to generate a plurality of building unit images; A building change detection method comprising: comparing the building unit images of the same location that were taken at different times in time and that have been subjected to the high resolution processing, and detecting information about changes in the building.

8. A step of dividing a two-dimensional building image obtained by photographing a plurality of buildings from above into a predetermined building unit to generate a plurality of building unit images; a step of performing a resolution enhancement process on the building unit images; A building change detection program that causes a computer to execute a step of comparing the building unit images of the same location that were taken at different times in time and that have been subjected to the high-resolution processing, and detecting information about changes in the building.

9. A step of performing a high-resolution processing on two-dimensional building images obtained by photographing a plurality of buildings from above; A step of dividing the building image subjected to the high resolution processing into a plurality of building unit images set in advance; A building change detection program that causes a computer to execute a step of comparing the building unit images of the same location that were taken at different times in time and that have been subjected to the high-resolution processing, and detecting information about changes in the building.

Images