Prediction device, prediction method, and prediction program

Abstract

To provide a prediction device, prediction method, and prediction program that can improve the accuracy of human flow predictions.SOLUTION: A prediction device includes: image acquisition means for acquiring satellite images taken by an artificial satellite, and drone images that are images taken by a drone and capture an area that overlaps at least a portion of the area captured by the satellite images; and human flow prediction means for predicting human flow using the satellite images and the drone images.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present disclosure relates to a prediction device, a prediction method, and a prediction program. [Background technology]

[0002] In modern society, rapid urban development and population growth are causing various problems, such as traffic congestion, overcrowding of public spaces, and increased environmental impact. To address these problems, it is necessary to accurately understand and predict people's behavioral patterns and trends in human flow. For example, there is a vehicle density observation device that observes the vehicle density on roads using input images based on images taken from the air and map information (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-128732 Summary of the Invention [Problem to be solved by the invention]

[0004] However, there are many moving parts when predicting people flow, and it is not easy to make highly accurate predictions.

[0005] The present invention aims to provide a prediction device, a prediction method, and a prediction program that can improve the accuracy of people flow predictions. [Means for solving the problem]

[0006] A prediction device based on the present disclosure includes an image acquisition means for acquiring satellite images taken by an artificial satellite and drone images, which are images taken by a drone of an area that overlaps with at least a portion of the area captured by the satellite image, and a people flow prediction means for predicting people flow using the satellite image and the drone image.

[0007] In the prediction method according to the present disclosure, a computer acquires satellite images taken by a satellite and drone images taken by a drone of an area that overlaps with at least a portion of the area captured by the satellite images, and predicts pedestrian flow using the satellite images and drone images.

[0008] A prediction program based on the present disclosure causes a computer to perform an image acquisition process that acquires satellite images taken by a satellite and drone images that are images taken by a drone of an area that overlaps with at least a portion of the area captured by the satellite images, and a people flow prediction process that predicts people flow using the satellite images and drone images. [Effects of the Invention]

[0009] According to the present invention, it is possible to improve the accuracy of predicting people flow. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram showing an overview of control of a disaster human damage prediction device according to the present disclosure. [Figure 2] FIG. 1 is a block diagram showing an example of the configuration of a disaster human casualty prediction device. [Figure 3] 1 is a flowchart showing the operation of the disaster human casualty prediction device. [Figure 4] FIG. 1 is an explanatory diagram illustrating an example of the hardware configuration of a disaster human damage prediction device according to the present disclosure. [Figure 5] FIG. 1 is an explanatory diagram showing an overview of the control of a crime occurrence area prediction device according to the present disclosure. [Figure 6] FIG. 1 is a block diagram showing an example of the configuration of a crime occurrence area prediction device. [Figure 7] 10 is a flowchart showing the operation of the crime occurrence area prediction device. [Figure 8] 1 is an explanatory diagram illustrating an example of the hardware configuration of a crime occurrence area prediction device according to the present disclosure. [Figure 9] FIG. 1 is a block diagram showing an overview of a prediction device. DETAILED DESCRIPTION OF THE INVENTION

[0011] Embodiment 1. In this embodiment, a prediction device predicts the flow of people and predicts human casualties when a disaster occurs. Figure 1 is an explanatory diagram showing an overview of the control of the disaster human casualty prediction device in the first embodiment. As shown in Figure 1, first, a specific location is periodically photographed by artificial satellites and drones flying in the sky, and the disaster human casualty prediction device acquires the photographed image data.

[0012] Next, the disaster human casualty prediction device performs analysis using AI (artificial intelligence). Specifically, the disaster human casualty prediction device inputs image data captured by satellites and drones, and uses the input data to train a people flow prediction model that predicts people flow and traffic volume by time of day and day of the week. The disaster human casualty prediction device also inputs past weather, topography, and disaster information data, and trains a disaster damage prediction model that predicts damage in the event of a disaster. In Figure 1, AI represents a machine learning model (prediction model).

[0013] Finally, the disaster human casualty prediction device performs a simulation. Specifically, by inputting disaster information into the learned prediction models (people flow prediction model, disaster damage prediction model), it predicts human casualties in the event of a disaster.

[0014] FIG. 2 is a block diagram showing an example of the configuration of a disaster human damage prediction device according to the first embodiment.

[0015] The disaster human casualty prediction device 100 of this embodiment predicts human casualties when a disaster occurs.

[0016] As shown in Figure 2, the disaster human damage prediction device 100 includes a people flow prediction unit 110 that receives data from an artificial satellite 200 and a drone 300, a disaster damage prediction unit 120, and a disaster human damage prediction unit 130.

[0017] The artificial satellite 200 includes an imaging unit 210 that images the Earth. Hereinafter, an image captured by the imaging unit 210 may be referred to as an artificial satellite image, and the data of the image may be referred to as artificial satellite image data.

[0018] The drone 300 includes an imaging unit 310 that captures images of the ground. Hereinafter, an image captured by the imaging unit 310 may be referred to as a drone-captured image, and data of the image may be referred to as drone-captured image data.

[0019] The people flow prediction unit 110 predicts the flow of people (people flow). As shown in the figure, the people flow prediction unit 110 includes a data acquisition unit 111, a first learning unit 112, and a first prediction unit 113.

[0020] The data acquisition unit 111 periodically acquires satellite-captured image data from the satellite 200. The data acquisition unit 111 periodically acquires drone-captured image data from the drone 300. Each piece of acquired photographed data is stored in a memory area (not shown) in the people flow prediction unit 110. Each piece of photographed data corresponds to learning data (training data) for training a people flow prediction model.

[0021] The first learning unit 112 learns a people flow prediction model that predicts people flow based on the satellite-captured image data and drone-captured image data acquired by the data acquisition unit 111 and event information. Specifically, the first learning unit 112 extracts people and vehicles from the satellite-captured image data and drone-captured image data, and counts the number of people in the image capture area by counting the number of extracted people and vehicles. Furthermore, the first learning unit 112 detects the movement status (e.g., temporal positional changes) of the detected people, i.e., people flow. The satellite-captured image data and drone-captured image data are associated with information on the date and time of each image capture and information on the day of the week. Event information is stored in an event database (not shown), in which information on holidays and upcoming events is associated with the date and time. The first learning unit 112 compares the date, time, and day of the week of each captured image data with the event information to determine whether an event is occurring, and then reflects this determination result in learning the people flow prediction model. As an example, a trained people flow prediction model can predict the number of people and the flow of people in a specified area when a specific day of the week or a specific event is input.

[0022] The first prediction unit 113 predicts the flow of people using a people flow prediction model. For example, when information on the location and date and time of a disaster occurrence is input, the first prediction unit 113 predicts the flow of people using the input information and the trained people flow prediction model.

[0023] Satellite images are images of a wide area on the ground. Therefore, by analyzing the satellite images, the first learning unit 112 can grasp large-scale population density and its fluctuations. In contrast, drone-captured images are images of a localized area on the ground (an area that is smaller than the area captured by the satellite images and that overlaps with at least a portion of the area captured by the satellite images). Therefore, by analyzing the drone images, the first learning unit 112 can locally complement the analysis results of the satellite images.

[0024] The disaster damage prediction unit 120 includes a past disaster history information storage unit 121, a meteorological and topographical information storage unit 122, a second learning unit 123, and a second prediction unit .

[0025] The past disaster history information storage unit 121 stores disaster information (disaster history information data) relating to disasters that have occurred in the past.

[0026] The weather and terrain information storage unit 122 stores past weather information and terrain information (including building layouts; weather and terrain information data). The weather and terrain information data corresponds to learning data (training data) for training a disaster damage prediction model.

[0027] The second learning unit 123 learns a disaster damage prediction model that predicts damage when a disaster occurs, based on the disaster information in the past disaster history information storage unit 121 and the weather information and topographical information in the weather and topographical information storage unit 122.

[0028] The second prediction unit 124 predicts disaster damage using a disaster damage prediction model. For example, when prediction conditions such as the location of a disaster, the type of disaster (earthquake, tsunami, typhoon, flood, etc.), or the scale of the disaster (seismic intensity, etc.) are input, the second prediction unit 124 predicts disaster damage using the input prediction conditions and disaster damage prediction model. For example, the disaster damage prediction unit 120 displays on a map the flooding situation for each area due to a tsunami after an earthquake as a disaster damage prediction result.

[0029] The disaster human damage prediction unit 130 predicts human damage in the event of a disaster based on the human flow prediction result from the human flow prediction unit 110 and the disaster damage prediction result from the disaster damage prediction unit 120. For example, the disaster damage prediction unit 120 calculates human damage (number of deaths and injuries) in the event of a tsunami using the number of people staying in a specific area as the human flow prediction result and the depth of flooding in the specific area as the disaster damage prediction result.

[0030] FIG. 3 is a flowchart showing the operation of the disaster human casualty prediction device 100.

[0031] In the disaster human casualty prediction device 100, the data acquisition unit 111 in the people flow prediction unit 110 acquires satellite-captured image data and drone-captured image data from the satellite 200 and the drone 300, respectively (step S112).

[0032] In the disaster human damage prediction device 100, the first learning unit 112 in the people flow prediction unit 110 learns a people flow prediction model that predicts the flow (trends) of people based on the acquired satellite-photographed image data and drone-photographed image data (step S114).

[0033] Specifically, the first learning unit 112 extracts people and vehicles from satellite-captured image data and drone-captured image data, measures the number of people and vehicles, and uses the measurement results, the date and time, the day of the week, and information on whether or not there is an event in the capture area to learn a people flow prediction model.

[0034] The second learning unit 123 in the disaster damage prediction unit 120 learns a disaster damage prediction model based on the past disaster history information data and the weather and topography information data (step S116).

[0035] The disaster human casualty prediction device 100 receives input of prediction conditions such as the location of the disaster, the type of disaster (earthquake, tsunami, typhoon, flood, etc.), or the scale of the disaster (seismic intensity, etc.) (step S118).

[0036] The first prediction unit 113 in the people flow prediction unit 110 predicts people flow using the input prediction conditions and the trained people flow prediction model (step S120).

[0037] The second prediction unit 124 in the disaster damage prediction unit 120 predicts disaster damage using the input prediction conditions and the trained disaster damage prediction model (step S122).

[0038] The disaster human damage prediction unit 130 of the disaster human damage prediction device 100 predicts human damage in the event of a disaster based on the human flow predicted in step S120 and the disaster damage predicted in step S122 (step S124).

[0039] The following describes a specific example of the hardware configuration of the disaster human casualty prediction device 100. Fig. 4 is an explanatory diagram showing an example of the hardware configuration of the disaster human casualty prediction device 100 according to the present disclosure.

[0040] The disaster human damage prediction device 100 shown in FIG. 4 includes a CPU (Central Processing Unit) 1000, a main memory unit 1001, and an auxiliary memory unit 1002.

[0041] The disaster human damage prediction device 100 is realized by software when the CPU 1000 shown in FIG. 4 executes a program that provides the functions of each component.

[0042] That is, the CPU 1000 loads the program stored in the auxiliary storage unit 1002 into the main storage unit 1001, executes it, and controls the operation of the disaster human damage prediction device 100, thereby realizing each function by software.

[0043] The main memory unit 1001 is used as a data working area and a data temporary saving area, and is, for example, a RAM (Random Access Memory).

[0044] The auxiliary storage unit 1002 is a non-transitory tangible storage medium, such as a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a semiconductor memory.

[0045] In the disaster human damage prediction device 100 of this embodiment, the auxiliary memory unit 1002 stores programs for realizing the people flow prediction unit 110 (data acquisition unit 111, first learning unit 112, first prediction unit 113), the disaster damage prediction unit 120 (second learning unit 123, second prediction unit 124), and the disaster human damage prediction unit 130.

[0046] The signal processing system may be implemented with a circuit including hardware components such as an LSI (Large Scale Integration) that realizes the functions shown in FIG.

[0047] As described above, in this embodiment, the disaster human damage prediction device 100 predicts human flow using satellite images taken by an artificial satellite and drone images taken by a drone, thereby improving the accuracy of human flow prediction.

[0048] Furthermore, the disaster human damage prediction device 100 predicts disaster damage information related to disaster damage in the event of a disaster based on past disaster damage history information and weather and topographical information, and predicts human damage in the event of a disaster based on the human flow prediction results and disaster damage prediction results. This improves the accuracy of predicting human damage in the event of a disaster. For example, human damage in the event of a disaster can be predicted quickly and accurately, enabling more effective disaster response.

[0049] Embodiment 2. In this embodiment, the prediction device predicts human flow and also predicts areas with a high risk of crime. FIG. 5 is an explanatory diagram showing an overview of the control of the crime occurrence area prediction device in the second embodiment. As shown in FIG. 5, first, specific locations are periodically photographed by artificial satellites and drones flying in the sky, and the crime occurrence area prediction device acquires the photographed image data. Note that in FIG. 5, AI represents a machine learning model (prediction model).

[0050] Next, the crime area prediction device uses AI to perform analysis. Specifically, the crime area prediction device inputs image data taken by satellites and drones, as well as past crime information, into the AI, and analyzes the flow of people and traffic volume by time of day and day of the week, street lighting intensity, etc., to extract features related to crime areas.

[0051] Finally, the features extracted by the disaster human casualty prediction device are used to predict areas with a high risk of crime.

[0052] FIG. 6 is a block diagram showing an example of the configuration of a crime occurrence area prediction device according to the second embodiment.

[0053] The crime occurrence area prediction device 400 of this embodiment predicts areas where there is a high risk of crime occurring.

[0054] As shown in FIG. 6, the crime occurrence area prediction device 400 includes a crime occurrence area prediction unit 410 to which data is input from an artificial satellite 500 and a drone 600 .

[0055] The artificial satellite 500 includes an imaging unit 510 that images the earth. Hereinafter, an image captured by the imaging unit 510 may be referred to as an artificial satellite image, and the data of the image may be referred to as artificial satellite image data.

[0056] The drone 600 includes a photographing unit 610 that photographs the ground. Hereinafter, an image photographed by the photographing unit 610 may be referred to as a drone-photographed image, and data of the image may be referred to as drone-photographed image data.

[0057] The crime occurrence area prediction unit 410 predicts areas where there is a high risk of crime occurring. As shown in the figure, the crime occurrence area prediction unit 410 includes a data acquisition unit 411, a past crime history information storage unit 412, a third learning unit 413, and a third prediction unit 414.

[0058] The data acquisition unit 411 periodically acquires satellite-captured image data from the satellite 500. The data acquisition unit 411 periodically acquires drone-captured image data from the drone 600. Each piece of acquired photographed data is stored in a memory area (not shown) in the crime occurrence area prediction unit 410. Each piece of photographed data corresponds to learning data (training data) for learning a crime occurrence area prediction model.

[0059] The past crime history information storage unit 412 stores crime information about crimes that have occurred in the past (information indicating the date, time, location, and type of crime that has occurred in the past; crime history information data). The crime history information data corresponds to learning data (training data) for training a crime occurrence area prediction model.

[0060] The third learning unit 413 learns a crime area prediction model that predicts crime areas based on the satellite-captured image data and drone-captured image data acquired by the data acquisition unit 411 and the crime history information data stored in the past crime history information storage unit 412. Specifically, the third learning unit 413 extracts elements related to crime occurrence, such as people, vehicles, street lights, building shadows, road widths, and intersection corners, from the satellite-captured image and drone-captured image, and counts the number of people in the captured area by counting the number of extracted people and vehicles. Furthermore, the third learning unit 413 detects the movement status (people flow) of the detected people. In other words, the third learning unit 413 includes the functions of the first learning unit 112 in the first embodiment.

[0061] Furthermore, the satellite-captured image data and drone-captured image data are associated with information on the date and time and day of the week when each image was captured. Furthermore, the crime history information data stored in the past crime history information storage unit 412 includes information indicating the date, time, location, and type of crime that occurred in the past. Because the measured number of people staying in the area and the crime history information data are associated by date, time, and location, the third learning unit 413 can extract conditions (features) that increase the risk of crime occurring and can train a crime occurrence area prediction model based on these conditions.

[0062] The third prediction unit 414 predicts crime occurrence areas using this crime occurrence area prediction model. For example, when the third prediction unit 414 receives input of prediction conditions related to date, time, and location, it predicts the crime occurrence risk for each district using the prediction conditions and the crime occurrence area prediction model, and creates a crime occurrence risk map by reflecting the risk on a map. The third learning unit 413 has the function of the first learning unit 112, which predicts the number of people staying in a specified area and the flow of people based on satellite and drone images, and therefore can predict the number of people staying and the flow of people. Therefore, the third prediction unit 414 can make predictions that take these into account.

[0063] Satellite images are images of a wide area on the ground. Therefore, by analyzing the satellite images, the third learning unit 413 can grasp large-scale population density and its fluctuations. In contrast, drone-captured images are images of a localized area on the ground (an area that is smaller than the area captured by the satellite images and that overlaps with at least a portion of the area captured by the satellite images). Therefore, by analyzing the drone images, the third learning unit 413 can locally complement the analysis results of the satellite images.

[0064] FIG. 7 is a flowchart showing the operation of the crime occurrence area prediction device 400.

[0065] In the crime occurrence area prediction device 400, the data acquisition unit 411 in the crime occurrence area prediction unit 410 acquires satellite-captured image data and drone-captured image data from the satellite 500 and drone 600, respectively (step S212).

[0066] The third learning unit 413 in the crime occurrence area prediction unit 410 learns a crime occurrence area prediction model that predicts crime occurrence areas based on the acquired satellite-photographed image data and drone-photographed image data and the crime history information data stored in the past crime history information storage unit 412 (step S214).

[0067] When the third prediction unit 414 in the crime occurrence area prediction unit 410 receives input of prediction conditions related to date, time, and location (step S216), it predicts the crime occurrence area using the input prediction conditions and a crime occurrence area prediction model (step S218).

[0068] The following describes a specific example of the hardware configuration of the crime occurrence area prediction device 400. Fig. 8 is an explanatory diagram showing an example of the hardware configuration of the crime occurrence area prediction device 400 according to the present disclosure.

[0069] The crime occurrence area prediction device 400 shown in FIG. 8 includes a CPU (Central Processing Unit) 4000, a main memory unit 4001, and an auxiliary memory unit 4002.

[0070] Crime occurrence area prediction device 400 is realized by software when CPU 4000 shown in FIG. 8 executes a program that provides the functions of each component.

[0071] That is, the CPU 4000 loads a program stored in the auxiliary storage unit 4002 into the main storage unit 4001, executes it, and controls the operation of the crime occurrence area prediction device 400, thereby realizing each function by software.

[0072] The main memory unit 4001 is used as a data working area and a data temporary saving area, and is, for example, a RAM (Random Access Memory).

[0073] The auxiliary storage unit 4002 is a non-transitory tangible storage medium, such as a magnetic disk, a magneto-optical disk, a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a semiconductor memory.

[0074] In the crime occurrence area prediction device 400 of this embodiment, the auxiliary memory unit 4002 stores programs for realizing the crime occurrence area prediction unit 410 (data acquisition unit 411, past crime history information memory unit 412, third learning unit 413, third prediction unit 414).

[0075] As described above, in this embodiment, the crime area prediction device 400 predicts people flow using satellite images taken by artificial satellites and drone images taken by drones. This improves the accuracy of people flow prediction.

[0076] Furthermore, the crime area prediction device 400 predicts crime areas with a high risk of crime based on the people flow prediction results and crime history information related to past crime histories. This improves the accuracy of predicting crime areas with a high risk of crime. For example, by predicting areas where crimes are likely to occur, it is possible to take crime prevention measures in advance, such as installing surveillance cameras and increasing patrols.

[0077] Next, an overview of the present invention will be described. Fig. 9 is a block diagram showing an overview of a prediction device according to the present invention. A prediction device 10 according to the present invention includes image acquisition means 11 (e.g., data acquisition units 111, 411) that acquires satellite images taken by an artificial satellite and drone images that are images taken by a drone of an area that overlaps at least a portion of the area captured by the satellite images, and people flow prediction means 12 (e.g., people flow prediction unit 110, crime occurrence area prediction unit 413) that predicts people flow using the satellite images and drone images.

[0078] This configuration makes it possible to update dynamic data using regularly acquired satellite and drone images, and to statistically analyze people's behavior patterns by time period, date, and event, thereby improving the accuracy of people flow predictions.

[0079] In addition, the people flow prediction means 12 may include a learning means for extracting people from both the satellite imagery and the drone imagery, and training a learning model for predicting people flow using people flow based on the extracted people and related information about the people flow.

[0080] The prediction device 10 may also include a disaster damage prediction means (e.g., disaster damage prediction unit 120) that predicts disaster damage information related to disaster damage in the event of a disaster based on past disaster damage history information and weather and topographical information, and a human damage prediction means (e.g., disaster human damage prediction unit 130) that predicts human damage in the event of a disaster based on human flow information predicted by the learning model and disaster damage information predicted by the disaster damage prediction means.

[0081] Such a configuration can improve the accuracy of predicting human damage in the event of a disaster.

[0082] The prediction device 10 may also be equipped with a crime occurrence area prediction means (e.g., crime area prediction unit 410) that predicts crime occurrence areas with a high risk of crime occurrence based on people flow information predicted by a learning model and crime history information regarding past crime history.

[0083] Such a configuration can improve the accuracy of predicting crime-prone areas where the risk of crime is high.

[0084] In another embodiment, the prediction device 10 can predict the congestion situation at a specific location based on traffic volume for each time period, day of the week, and event, which can lead to the alleviation of congestion by implementing traffic restrictions, etc.

[0085] Furthermore, the prediction device 10 can predict people's behavior patterns based on traffic volume for each time period, day of the week, and event, which can be used for effective event hosting and marketing.

[0086] Additionally, the results of people flow prediction by the prediction device 10 may be used in fields such as urban planning and infrastructure development, tourism, retail and commercial facilities, and the like.

[0087] Generally, there are two reasons why predicting people flow is difficult. First, there is uncertainty in the data used for the prediction. For example, predicting people flow requires data on population density under normal circumstances, but population density fluctuates significantly depending on the time of day, day of the week, and season, making it difficult to accurately grasp. Second, there is uncertainty in people's behavior patterns. For example, people's behavior is affected by a wide variety of external factors, such as weather, events, transportation conditions, and sudden social phenomena, making it difficult to predict the impact of these factors on people flow.

[0088] In the above embodiment, by integrating satellite and drone images to perform people flow prediction, it is possible to collect wide-area and accurate people flow data and perform people flow prediction based on time series. Specifically, by using satellite images to grasp large-scale population density and its fluctuations, and by using drone images to store detailed local data, it is possible to more precisely capture changes in people flow depending on the time of day and day of the week, thereby reducing data uncertainty. In addition, by utilizing detailed geographical and temporal data obtained from satellite and drone images to analyze the relationship between external factors (e.g., the presence or absence of events) and people flow, it is possible to reduce uncertainty in people's behavior patterns. This improves the accuracy of people flow prediction.

[0089] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0090] (Supplementary Note 1) An image acquisition means for acquiring a satellite image taken by an artificial satellite and a drone image taken by a drone, the drone image being an image of an area overlapping at least a part of the imaged area of ​​the satellite image; and a people flow prediction means for predicting people flow using the satellite image and the drone image. Prediction device.

[0091] (Supplementary Note 2) The people flow prediction means includes a learning means (e.g., a first learning unit 112, a third learning unit 413) that extracts people from each of the satellite image and the drone image, and trains a learning model that predicts people flow using people flow based on the extracted people and related information about the people flow (e.g., day of the week, presence or absence of an event). 2. The prediction device of claim 1.

[0092] (Appendix 3) A disaster damage prediction means for predicting disaster damage information regarding disaster damage when a disaster occurs based on past disaster damage history information and meteorological and topographical information; and a human damage prediction means for predicting human damage in the event of a disaster based on the human flow information predicted by the learning model and the disaster damage information predicted by the disaster damage prediction means. 3. The prediction device of claim 2.

[0093] (Appendix 4) A crime occurrence area prediction means is provided for predicting crime occurrence areas with a high risk of crime occurrence based on the people flow information predicted by the learning model and crime history information regarding past crime history. 3. The prediction device of claim 2.

[0094] (Appendix 5) The computer Acquire a satellite image taken by a satellite and a drone image taken by a drone, the drone image capturing an area that overlaps at least a portion of the area captured by the satellite image; Predicting people flow using the satellite images and the drone images Forecasting methods.

[0095] (Appendix 6) The computer, People are extracted from each of the satellite image and the drone image, and a learning model for predicting people flow is trained using people flow and related information based on the extracted people. Forecasting method described in Appendix 5.

[0096] (Appendix 7) The computer Based on past disaster damage history information and weather and topographical information, we predict disaster damage information in the event of a disaster, Based on the people flow information predicted by the learning model and the disaster damage information predicted by the disaster damage prediction means, human damage is predicted in the event of a disaster. The forecasting method described in Appendix 6.

[0097] (Appendix 8) The computer Predicting crime-prone areas with a high risk of crime based on the people flow information predicted by the learning model and crime history information related to past crime histories. The forecasting method described in Appendix 6.

[0098] (Appendix 9) To the computer, an image acquisition process for acquiring a satellite image taken by a satellite and a drone image taken by a drone, the drone image capturing an area that overlaps at least a portion of the area captured by the satellite image; A people flow prediction process that predicts people flow using the satellite image and the drone image; A prediction program for executing the above.

[0099] (Appendix 10) To the computer, People are extracted from each of the satellite image and the drone image, and a learning model for predicting people flow is trained using people flow and related information based on the extracted people. Attachment 9. The prediction program described in the appended claim 9.

[0100] (Appendix 11) To the computer, A disaster damage prediction process for predicting disaster damage information related to disaster damage in the event of a disaster based on past disaster damage history information and weather and topographical information; and executing a human damage prediction process for predicting human damage in the event of a disaster based on the human flow information predicted by the learning model and the disaster damage information predicted by the disaster damage prediction means. 11. The prediction program according to claim 10.

[0101] (Appendix 12) To the computer, A crime occurrence area prediction process is executed to predict crime occurrence areas with a high risk of crime occurrence based on the people flow information predicted by the learning model and crime history information related to past crime history. 11. The prediction program according to claim 10. [Explanation of symbols]

[0102] 10 Prediction Device 11 Image acquisition method 12 People flow prediction method 100 Human damage prediction device during disasters 110 People flow prediction department 111 Data Acquisition Unit 112 First Learning Section 113 First Prediction Section 120 Disaster Damage Prediction Department 121 Past disaster history information storage unit 122 Weather and terrain information storage unit 123 Second Learning Section 124 Second Prediction Section 130 Disaster Human Casualty Prediction Department 200,500 satellites 210,510 Photography Department 300,600 drones 310,610 Photography Department 400 Crime Area Prediction Device 410 Crime Area Prediction Department 411 Data Acquisition Department 412 Past Criminal History Information Storage Unit 413 Third Learning Section 414 Third Prediction Department 1000,4000 CPU 1001,4001 Main memory 1002,4002 Auxiliary storage

Claims

1. an image acquisition means for acquiring a satellite image taken by an artificial satellite and a drone image taken by a drone, the drone image being an image of an area that overlaps at least a part of the area of ​​the satellite image; and a people flow prediction means for predicting people flow using the satellite image and the drone image. Prediction device.

2. The people flow prediction means includes a learning means for extracting people from each of the satellite image and the drone image, and for learning a learning model for predicting people flow using people flow based on the extracted people and information related to the people flow. The prediction device according to claim 1 .

3. disaster damage prediction means for predicting disaster damage information relating to disaster damage in the event of a disaster based on past disaster damage history information and meteorological and topographical information; and a human damage prediction means for predicting human damage in the event of a disaster based on the human flow information predicted by the learning model and the disaster damage information predicted by the disaster damage prediction means. The prediction device according to claim 2 .

4. crime occurrence area prediction means for predicting crime occurrence areas with a high risk of crime occurrence based on the people flow information predicted by the learning model and crime history information relating to past crime history; The prediction device according to claim 2 .

5. The computer Acquire a satellite image taken by a satellite and a drone image taken by a drone, the drone image capturing an area that overlaps at least a portion of the area captured by the satellite image; Predicting people flow using the satellite images and the drone images Forecasting methods.

6. The computer People are extracted from each of the satellite image and the drone image, and a learning model for predicting people flow is trained using people flow and related information based on the extracted people. The prediction method according to claim 5.

7. On the computer, an image acquisition process for acquiring a satellite image taken by a satellite and a drone image taken by a drone, the drone image capturing an area that overlaps at least a portion of the area captured by the satellite image; A people flow prediction process that predicts people flow using the satellite image and the drone image; A prediction program for executing the above.

8. On the computer, People are extracted from each of the satellite image and the drone image, and a learning model for predicting people flow is trained using people flow and related information based on the extracted people. The prediction program according to claim 7.

Images