Flood information generation device, flood information generation method, and program

Abstract

To provide a water immersion information generation device, a method for generating water immersion information, and a program which can more easily generate water immersion information.SOLUTION: A water immersion information generation system includes: an analysis server; an image data server; a weather information distribution server; a people flow information provision server; an area information data server; a feature information data server; an imaging device having an in-vehicle terminal, a member terminal, and a fixation camera; and a distribution destination terminal. The water immersion information generation device includes: image acquisition means for acquiring a road image of a road R and information of the taken position of the road image; region detection means for detecting a road edge object region as an object located at an edge of the road R, a road region between a pair of road edge object regions, and region detection means for detecting a submergence region F of a submergence part of the road R from the road image; and extraction means for extracting the position which is the nearest to the center of the road R in the direction crossing the road region Pf of the outer periphery of the submergence region F of the road region as the water immersion evaluation point.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] This invention relates to a flooding information generation device, a flooding information generation method, and a program.

Background Art

[0002] Flood damage often occurs in various places due to heavy rain and the like. By quickly grasping the flooded area, water depth, and water volume when flooding occurs, disaster countermeasures can be promptly formulated and implemented. Conventionally, there is a technique for specifying the flooded area using a photographed image from above using an aircraft such as a helicopter. However, an aircraft may not be able to fly or may not be able to take appropriate photographs due to heavy rain or strong winds, and thus information may not be obtained promptly. On the other hand, in recent years, there is a technique for estimating the flooded area by using posted images on SNS (Social Network Service) for estimating the flooded area and estimating the flooded area by the level flooding method based on the highest flooded point and the surrounding terrain (Non-Patent Document 1).

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, there is a problem that it takes a great deal of effort to obtain elevation information related to the flooded position from a large number of posted images taken at unspecified positions and select and extract the necessary information, which hinders the rapid generation of flooding information.

[0005] The purpose of this disclosure is to provide a flood information generation device, a flood information generation method, and a program that can more easily obtain the information necessary to generate flood information. [Means for solving the problem]

[0006] To achieve the above objectives, this disclosure is intended to An image acquisition means for acquiring road images of roads and information on the location where said road images were taken, From the aforementioned road image, an object located at the edge of the road Corresponding A region detection means for detecting a road edge object region, a road region sandwiched between a pair of road edge object regions, and a submerged region which is the submerged part of the road, Of the outer edge of the submerged area in the road area, in the direction that crosses the road area Based on the gradient The aforementioned road The highest elevation An extraction method for extracting the location closest to the flood assessment point, This is a flood information generation device characterized by having the following features: [Effects of the Invention]

[0007] According to the present invention, it is possible to obtain the information necessary to generate flood information more easily. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing the configuration of the flood information generation system. [Figure 2] This figure shows an example of road surface flooding. [Figure 3] This figure shows an example of detecting road flooding from road images. [Figure 4] This diagram illustrates the identification of the edges and center of a road. [Figure 5] This diagram schematically illustrates an example of changes in road flooding conditions. [Figure 6] This figure shows an example of a steep slope area where a danger level notification is issued. [Figure 7]This flowchart shows the control procedure for the flood information generation control process. [Figure 8] This is a flowchart showing the control procedure for the region detection process. [Figure 9] This flowchart shows the control procedure for identifying flood inundation assessment points. [Figure 10] This is a flowchart showing the control procedure for the flood risk assessment process. [Modes for carrying out the invention]

[0009] Hereinafter, embodiments of the present invention will be described based on the drawings. Figure 1 is a block diagram showing the configuration of the flood information generation system 100 of this embodiment.

[0010] The flood information generation system 100 includes an analysis server 10, which is a flood information generation device in this embodiment, an image data server 20, a weather information distribution server 30, a human flow information provision server 40, an area information data server 50, a topographic information data server 60, a shooting device 70 including an in-vehicle terminal 71, a member terminal 72, and a fixed camera 73, and a distribution destination terminal 80. These are connected via networks N1 and N2. Networks N1 and N2 can be the Internet or a LAN (Local Area Network), and do not need to be separate.

[0011] The analysis server 10 acquires and analyzes road images, which are image data of roads, to estimate the extent of flooding and calculate the degree of risk of flooding (flood risk). The analysis server 10 distributes the obtained information, such as the extent of flooding and the flood risk, to the designated destination.

[0012] The analysis server 10 includes a CPU 11 (Central Processing Unit), a RAM 12 (Random Access Memory), a storage unit 13, a communication unit 14, and the like. The CPU 11 is a hardware processor that performs arithmetic processing and overall controls the operation of the analysis server 10. The CPU 11 does not have to be single, and a plurality of CPUs may perform arithmetic processing in parallel or independently for each application.

[0013] The RAM 12 provides a working memory space for the CPU 11 and stores temporary data. The RAM 12 is, for example, but not limited to, DRAM. At least the CPU 11 and the RAM 12 are included in the computer of the present embodiment.

[0014] The storage unit 13 is a non-volatile memory that stores programs, setting data, and the like. The non-volatile memory is, for example, a flash memory, a HDD (Hard Disk Drive), or the like. The storage unit 13 stores the learned model 131 as model storage means. Further, a program 132 related to the control of the generation of flooding information using the learned model 131 is stored in the storage unit 13.

[0015] The communication unit 14 controls operations related to communication (data transmission and reception) with external devices. The communication operation is controlled according to a predetermined communication standard. The communication standard may include, for example, the standard of a LAN (Local Area Network) or the standard of a wireless LAN. The analysis server 10 may also include a display unit, an operation reception unit, and the like, and it may be possible for an operator to check the processing status or perform an input operation.

[0016] The image data server 20 acquires and stores road image data from the in-vehicle terminal 71, member terminal 72, and fixed camera 73, along with camera parameter information including the date and time of capture and at least the location where the image was taken. The road images may be classified and stored according to the date and time of capture and location where the image was taken. The image data server 20 may delete the road image data that has been sent to the analysis server 10.

[0017] The weather information distribution server 30 distributes weather information, in this case rainfall information, specifically data on the areas, time periods, and amount of rainfall, to external parties. The pedestrian flow information server 40 distributes pedestrian flow information, including location information of pedestrians and vehicles obtained from mobile devices such as smartphones and car navigation systems, and / or information on abnormalities in pedestrian flow obtained by analyzing this location information, such as areas where people are concentrated (congested) or cars are stuck in traffic, or conversely, areas where these are being avoided.

[0018] The weather information distribution server 30 and the human movement information provision server 40 may distribute information generated by organizations or individuals, including for-profit companies, and the organizations or individuals distributing the information may be different from the organizations or individuals operating and using the analysis server 10 and the image data server 20. The distributed information may be made publicly available without restriction, or it may be provided to contractors or members on a limited basis, either for a fee or free of charge.

[0019] The area information data server 50 stores area information 51, such as areas to which the level flooding method described later is applied and steep slope areas (corresponding areas) that are subject to evaluation of flood risk, within the analysis range of the analysis server 10. Areas to which the level flooding method is applied are geographical areas to which the level flooding method can be applied, and are set for areas where flooding may occur due to heavy rain water accumulating because it cannot be drained completely (internal flooding) or flooding due to flooding from rivers (external flooding). Steep slope areas can be, for example, warning areas established based on the Act on Promotion of Measures to Prevent Sediment-Related Disasters in Sediment-Related Disaster Warning Areas, etc. (Sediment-Related Disaster Prevention Act). In this embodiment, a warning area that includes a steep slope and an area adjacent to the steep slope on the lower side (flood-prone area; that is, an area where flood damage is expected due to water flow on the steep slope) is set as a steep slope area. Note that warning areas may include areas adjacent to the steep slope on the upper side, and these areas may suffer damage such as isolation due to flooding of the area adjacent to the lower side. In light of this, it is acceptable to designate areas that are at risk of flooding, including areas that are adjacent to steep slopes at an upper level, as areas with a high risk of flooding. Furthermore, the area information data server 50 stores associated destinations (recipients) for distributing information indicating the occurrence of flooding or an increase in the risk level, for each area subject to the Level Flood Control Law and for steep slope areas. In addition, even if an area is neither an area subject to the Level Flood Control Law nor a steep slope area, the local government or other entity to which the area belongs may be associated and stored as a recipient.

[0020] The terrain information data server 60 stores terrain data 61 of roads and their surrounding areas within the analysis range of the analysis server 10. The terrain data 61 includes, for example, DEM data (Digital Elevation Model). DEM data is defined by assigning three-dimensional coordinates to each unit area (grid, cell, or mesh) divided into a two-dimensional matrix of an appropriate size. In addition, the road surface data may also include elevation data based on unevenness information obtained by light sectioning or laser scanning (laser point cloud measurement). Alternatively, information such as the gradient in the direction of road extension, the gradient in the transverse direction perpendicular to the direction of extension, and the width of the road may be acquired in advance based on the road register and stored as road surface data. These area information 51 and terrain data 61 may be stored in the storage unit 13 of the analysis server 10.

[0021] The imaging device 70 photographs the road and generates and outputs road image data. The imaging device 70 generates and outputs information of at least the date and time of shooting and the shooting location along with the road image. The shooting location includes at least latitude and longitude, and preferably also altitude. The shooting location is not particularly limited, but may be acquired by the imaging device 70 itself through satellite positioning or the like. If the in-vehicle terminal 71 and the fixed camera 73 can acquire camera parameters other than the shooting location (such as shooting direction and field of view), these camera parameters are also output. The in-vehicle terminal 71 and the fixed camera 73 may, for example, periodically (the in-vehicle terminal 71 may be limited to when the vehicle is in motion or when the engine is running) take pictures and automatically transmit the image data to the image data server 20. Such in-vehicle terminal 71 and fixed camera 73 may be pre-contracted imaging devices.

[0022] A member terminal 72 is a terminal device of a user who has entered into a contract or has agreed to provide road images obtained primarily through shooting operations on their own device to the flood information generation system 100. The member terminal 72 transmits road image data obtained through the shooting operations of the user of the member terminal 72 (collectively referred to as a member) to the image data server 20. A member terminal 72 includes mobile terminals with shooting capabilities, such as smartphones and pod terminals, and does not need to be a dedicated terminal device related to the above contract. Therefore, the images captured by each member terminal 72 are not limited to road images, and road images are selectively transmitted to the image data server 20. The transmission of road images can be performed, for example, in response to the uploading of the road images to a specific SNS, or through a direct transmission operation by the member. An example of a member is a member of a navigation service. In addition, users of application programs (apps) that utilize the functions of the member terminal 72, such as activity (exercise amount, vital information) management (those who have agreed to the license and provision of road images), and employees of flood information distribution service companies may also be included as members. Alternatively, all terminal devices that post road images to a specific social networking service whose terms of service include consent to the provision of road images may be included in member terminal 72.

[0023] The distribution terminal 80 is a terminal device that serves as the recipient (destination) for the flood information and flood risk information obtained by the analysis server 10. This distribution terminal 80 may be, for example, a PC or various general-purpose mobile terminals. If the information is distributed by email, the email address is the virtual recipient. The distribution terminal 80 may include terminal devices of organizations that deal with flood damage, such as local governments, designated government offices, and logistics companies. The distribution terminal 80 may acquire the distribution information by email as described above, or it may be received by push notification via a dedicated application, etc., and notification may be performed immediately. The flood information and flood risk information may be distributed to recipients corresponding to areas where flooding has occurred or the flood risk has increased, for example, and there may be organizations or individuals that have contracts to receive all information (information corresponding to multiple areas) generated by the analysis server 10.

[0024] Next, we will explain how to calculate the flooded area in this embodiment. Since the water surface is static and locally spread over an area of ​​the same elevation, this kind of flooded surface spread is likely to occur within each area that is divided into limited areas such as depressions and riverbanks. The method of estimating the highest point of the submerged ground surface elevation according to the amount of water that flows in under these conditions and the volume that is stored is called the level flooding method.

[0025] If the elevation of the localized flooded surface can be determined from images taken of the actual submerged area, the extent of the flooding can be estimated based on this elevation and topographic data 61. When determining the elevation of the flooded surface in this way, the road surface, especially the surface of paved roads, is easy to determine the elevation of, so measuring the flooding status of the road surface is suitable for estimating the surrounding flooded surface. The height of the road surface is usually continuous in the direction of extension. At the same time, the road surface also has a gradient in the transverse direction that intersects (in this case perpendicular to) the direction of extension. The gradient in the transverse direction of the road surface is a gentle slope that is highest in the center of the road surface and lower at both ends, so that water does not accumulate in the center of the road and is easily drained to both sides.

[0026] Figure 2 shows an example of road surface flooding. As shown in Figure 2(a), on road R, the lower elevation areas are flooded and submerged first, so the flooded area Af is narrower in the center in the transverse direction than at the ends. Note that in this figure, the center line is located in the center in the transverse direction of road R, but this is not the only case.

[0027] If the flooding is less severe than described above, as shown in Figure 2(b), the center of the road R does not flood in the transverse direction, and only a portion of the road width W from both ends is flooded (area Af).

[0028] Next, we will explain how to calculate the flooded area. Figure 3 shows an example of detecting flooding on road R from a road image.

[0029] As shown in Figure 3(a), when road surface flooding first occurs, a localized flooded area Af is created in the relatively low elevation portion of the sloping road R. To obtain the elevation of the flooded area from the road image, the analysis server 10 processes the road image to segment it into regions. For example, semantic segmentation in deep learning is used as the image recognition technology for region segmentation. Here, as shown in Figure 3(b), the regions (classes) to be classified are defined as the submerged region F corresponding to the flooded area Af, the road edge object region E corresponding to the road edge object Ae, the road surface region Rn corresponding to the unflooded portion of road R, and other regions. Furthermore, if other vehicles located on the road are included in the image, the vehicle region may be classified separately.

[0030] A pre-trained machine learning model 131 is used to detect each region in road images using semantic segmentation. The generation of a pre-trained model that detects and outputs at least the submerged region F and the road edge object region E from the input of a road image is performed in advance on any computer using training data. Multiple training images (example images) are prepared by appropriately combining road images in which road R is photographed with the flooded area Af and road images in which the flooded area Af is not photographed, and ground truth data is prepared that indicates which region each pixel of the training image belongs to. The pre-trained model is obtained by updating the parameters of the machine learning model so that the output obtained from inputting the training images to the machine learning model matches the ground truth data.

[0031] Of the regions classified in this way, the road edge object region E is a region that possesses the characteristics of road edge objects Ae, which are characteristic objects (three-dimensional objects, i.e., objects with height in a direction perpendicular to the road surface) located at the edge of the road. Such road edge objects Ae include, for example, sidewalks, guardrails, curbs, and tunnel side walls. Even when the road R is flooded, road edge objects Ae are likely to be partially exposed above the flooded surface.

[0032] The flooded area F is the area of ​​the road surface that has the characteristics of the flooded area Af (submerged). The road surface area Rn is the area of ​​the road surface that has the characteristics of the part that is not flooded (not submerged) (including when the road surface is dry and when it is wet). Areas that are not the road surface or road edge objects Ae are considered other areas, that is, buildings along the roadside, vegetation, ground, walls, fences, traffic lights, signs, billboards, utility poles and power lines, and bodies of water outside the road (including rivers and ponds) are all included in the other areas. Furthermore, even on the road surface, parts where the characteristics of the road surface cannot be obtained due to obstacles or light reflection may be classified as other areas.

[0033] Figure 4 illustrates the identification of the edges and center of a road. Road edge objects Ae are not always fully visible in road images due to submersion, signs, vehicles, pedestrians, and roadside vegetation (including planted areas). Furthermore, guardrails may have incomplete sections. Taking these cases into consideration, the road edge object region E is divided into two groups, left and right, and the boundary line Le within each group (the road side, closer to the opposite group and the submerged region F) is approximated.

[0034] For example, adjacent pixels recognized as road edge object regions E are grouped together to generate groups of pixels in road edge object regions E. For each group thus obtained, the average position (coordinates) of the pixels included in the group is calculated, and a line segment is obtained connecting the calculated average positions. If a pixel on the line segment contains a submerged region F or a road surface region Rn, the group of pixels connected by the line segment corresponds to road edge objects Ae on different sides of road R. If a pixel on the line segment does not contain either a submerged region F or a road surface region Rn, the group of pixels connected by the line segment corresponds to road edge objects Ae on the same side of road R. As a result, groups of three or more are consolidated into two groups along both ends of the road. Note that groups of pixels with fewer than the standard number of pixels may be excluded from the above process and converted into other regions. In captured images, if optical distortion is ignored and the road gradient is uniform, the boundary line Le will be a straight line. However, captured images of partially flooded roads often include cases where the gradient changes along the way. Therefore, the boundary line Le may be approximated by a curve, for example, a quadratic curve.

[0035] The boundary line Le is obtained, for example, as a curve approximating the sequence of pixels on the outer perimeter of the road edge object region E that face the road, for each of the two aggregated groups. Each pixel on the outer perimeter of the road edge object region E is connected by a line segment to the nearest pixel in the submerged region F or road surface region Rn. If there are no other pixels of the road edge object region E on this line segment, it is determined that this pixel faces the road. The boundary line Le is obtained by finding a quadratic regression curve for the coordinates of the group of pixels determined to face the road using the least squares method or similar. In the case of road edge objects Ae located at a predetermined height from the road surface, such as guardrails, the apparent road region is defined as being wider than it actually is, so the obtained boundary line Le may be shifted inward according to the height of the guardrail. If both the left and right sides are the same type of road edge object Ae, such as guardrails, the apparent boundary line Le may be used as is.

[0036] The area between the pair of boundary lines Le (inner boundary) of the road edge object area E obtained as described above is the road area in the flood information generation system 100 of this embodiment. The submerged area F within this road area is identified as having been recognized correctly, and the area other than the submerged area F is identified as the road surface area that is not flooded. However, if there is an area that could not be correctly recognized by image recognition that remains within the range surrounded by the recognized submerged area F and road surface area Rn, these may be included in the submerged area and road surface area, respectively. Also, if the submerged area F and road surface area Rn are recognized as being outside the road area, these classifications may be canceled.

[0037] The boundary line Le may be asymmetrical depending on the direction in which the road image was taken. The direction in which a line passes through a point on the road and crosses it can be asymptotically determined by, for example, finding the angles θR and θL between a suitable straight line Lc passing through that point and the two boundary lines Le, and then changing the direction of the straight line Lc so that the angle difference |θR-θL| is minimized (local minimum) depending on the magnitude of the angles θR and θL. Alternatively, the direction of the straight line Lc can be simply changed by small angles in one direction, and the direction in which the minimum angle difference is obtained can be identified as the crossing direction. Furthermore, since the crossing direction is often perpendicular to lines on the road (such as the center line or lane boundary lines), the crossing direction can be estimated based on the angle with respect to these lines, and the direction in which the difference between angles θR and θL is minimized can be identified by changing the direction by small angles within a narrow range near the estimated direction. By finding the midpoint of the line segment in the identified direction, the center position C of the road can be obtained even if the entire edge of the road is not visible.

[0038] The intersection of this transversely extending straight line Lc and the outer edge of the submerged area F is the outer edge point Pb, and the distance D between the outer edge point Pb and the central position C can be determined. For each pixel located on the outer edge of the submerged area F, excluding the points facing the boundary line Le, the distance D can be obtained by determining the transversely extending straight line Lc passing through each of the remaining outer edge points Pb and the central position C on that straight line Lc. The minimum value of the obtained distances D is taken as the distance Dmin, and the outer edge point Pb that makes this distance Dmin, i.e., the closest to the central position C, is determined as the inundation evaluation point Pf. As shown in Figure 2, when the road is partially flooded, the wider the flooded area Af becomes, the closer the inundation evaluation point Pf is to the center of the road, and the closer the distance Dmin approaches 0. When the entire width of the road is flooded, the inundation evaluation point Pf is located in the center, and the distance Dmin = 0. In the example in Figure 4, there are two inundation evaluation points Pf where the distance Dmin = 0. Road width (carriageway width W, lane width W / 2), etc., can be stored separately in advance as data based on road registers or similar documents.

[0039] Camera parameter information, including at least the shooting position, is stored in association with the road image. The geographical location (three-dimensional coordinates) of the flood inundation assessment point Pf is identified and extracted using these camera parameters and the topographic data of the captured road area. For example, by slightly changing the direction and field of view from the shooting position, the extent of the road and, if possible, its surroundings (buildings, signs, surrounding terrain, etc.) are virtually defined based on the topographic data. The combination of direction and field of view that maximizes the degree of agreement with the obtained road image is identified as the shooting direction and field of view of the road image. The degree of agreement regarding the road extent can be, for example, the degree of overlap between the area where the appearance of the road area represented by the road surface data is virtually defined in the topographic data and the road area extracted from the road image. Based on this identification result, the geographical location of the flood inundation assessment point Pf is determined by inversely calculating the three-dimensional coordinates projected onto the flood inundation assessment point Pf within the road image. If the received shooting location does not include altitude, the elevation of the ground surface can be determined using latitude, longitude, and topographic data, and the altitude can be supplemented by adding the height of the shooting device 70 from the ground surface, which is predetermined according to the type of shooting device 70 (for example, the average shoulder height for a member terminal 72). Furthermore, for road images where the shooting direction and field of view are obtained as camera parameters, the determination of the shooting direction and field of view in the above method can be omitted. In addition, if the degree of agreement regarding the road range obtained by the above method does not reach a predetermined threshold, the operator may manually determine the three-dimensional coordinates of the flood point Pf based on the road image and the topographic data in which the shooting location has been determined. Furthermore, if the road images are in video format or consist of multiple still images taken at close intervals, the Structure from Motion (SfM) method can be applied to multiple frame images or multiple still images to determine the geographical location of the flood inundation assessment point Pf. For example, SfM can be used to identify the line of sight to the flood inundation assessment point Pf in the frame images or still images, and the geographical location of the flood inundation assessment point Pf can be determined from the shooting location of the image, the line of sight from the shooting location to the flood inundation assessment point Pf, and topographic data. Alternatively, SfM can be used to calculate a 3D point cloud of the road captured in the road image, calculate the flood inundation assessment point Pf in the coordinate system of the 3D point cloud, and then fit the 3D point cloud to the topographic data to determine the geographical location of the flood inundation assessment point Pf. If the shooting location or flood assessment point Pf is within the area where the level flooding method is applied, the elevation value, which is the height component of these coordinates, becomes the elevation of the flooded surface (operation as a calculation method). Based on the topographic data of the area where the level flooding method is applied, the area where the ground surface elevation is lower than the calculated elevation is estimated as the flooded area. Flood information corresponding to the estimated flooded area is generated and output.

[0040] On the other hand, roads that cross slopes such as the inclines of mountains and hills are not typically included in the scope where the level flooding method applies, but water flow can descend from above along the slope, causing flooding of buildings and other structures on the lower side. In such roads, especially in locally low-lying areas, if the water flow increases, the side adjacent to the upper slope will begin to be partially flooded according to the transverse gradient, and then, once the flooded area exceeds the highest point in the center (and depending on the road structure, may also overflow curbs and sidewalks on the lower side of the slope), the water flow will descend the slope. Furthermore, if the road is completely flooded, it will hinder vehicle traffic.

[0041] Therefore, the analysis server 10 tracks the change in the location of flood assessment point Pf on such roads to evaluate the degree of flood risk associated with the generation of water flow down the slope.

[0042] Figure 5 is a schematic diagram illustrating an example of changes in road flooding conditions. As shown in Figure 5(a), in the initial stages of road flooding, the flood assessment point Pf is located closer to the edge of the road than to the center, and the distance Dmin is close to half of the road width W, W / 2.

[0043] As flooding progresses, the flood assessment point Pf approaches the center of the road. In the situation shown in Figure 5(b), the flood assessment point Pf is just before the center of the road, and the distance Dmin is close to zero. When the distance Dmin becomes zero, the water flows over the highest point in the center of the road to the opposite side. In other words, the closer the flood assessment point Pf is to the center in the transverse direction of the road (the closer the submerged area F extends to the center), the higher the risk. For this risk assessment, the distance between the flood assessment point Pf and the center may be used directly, or a value Rmin = Dmin / W (an index representing the extent of submersion), which is this distance normalized by the road width W, may be used.

[0044] Figure 6 shows an example of a steep slope area where a risk level notification is issued. If a road R extends along the middle of a slope (steep incline), and a facility H such as a residence is located adjacent to a portion of the road R on the lower side of the slope (lower elevation side), then when the road R becomes flooded due to water flowing from the upper part of the slope, the water can flow further down the slope and cause flooding of the facility H. Therefore, based on the photographic data of the road R, the risk of flooding in the area including the facility H (flood-prone area) is evaluated, and this risk is distributed to the receiving terminal 80 at the facility H. This enables early evacuation or withdrawal from the facility H. Note that the distribution of flood risk information is not limited to the receiving terminal 80 at the facility H within the flood-prone area below the road. For example, it may also include the receiving terminal 80 of the local government to which this steep incline area, especially the flood-prone area, belongs, the department of the government agency responsible for the management and disaster countermeasures of the area, and transportation companies and bus companies that regularly use the road R itself or regularly travel to and from the facility H and its surroundings.

[0045] Figure 7 is a flowchart showing the control procedure by the CPU 11 for the flood information generation control process executed in the analysis server 10 of this embodiment. Figures 8, 9, and 10 are flowcharts showing the control procedures for the area detection process, the flood evaluation point identification process, and the flood risk assessment process, respectively, which are executed within the flood information generation control process. This flood information generation control process, including the flood information generation method of this embodiment, is started by an input operation related to a user's start command and is executed continuously.

[0046] The CPU 11 (image acquisition means, image acquisition step) acquires data of camera parameters, which include at least a road image and the shooting location (step S101). The CPU 11 acquires detection target determination information to determine whether or not the shooting location is a target for detection of a currently submerged area (step S102). This detection target determination information includes the above-mentioned weather information, pedestrian flow information, and area information.

[0047] The CPU 11 determines whether the shooting location is within the detection area for a submerged area (step S103). Specifically, the CPU 11 determines whether the shooting location is within a rainfall area (which may include a predetermined time after the rain has stopped, or may include an area downstream of the river basin from the actual rainfall area) or an area with abnormal human flow (information related to man-made abnormalities such as construction or accidents may be obtained separately and excluded), and whether it is an area where the level flooding method is applied, a steep slope area, or an area where flood damage is anticipated, and whether it is an area for which a distribution destination has been set.

[0048] If it is determined that the shooting location is not within the detection area (NO in step S103), the CPU 11 returns to step S101.

[0049] If the shooting location is determined to be within the detection target area (YES in step S103), the CPU 11 (area detection means, area detection step) executes area detection processing (step S104). As shown in Figure 8, in the area detection processing of step S104, the area detection means inputs the road image to be analyzed into the trained model 131 and obtains the recognition result of the range of each area in the image output from the trained model 131 (step S141). The area detection means determines whether or not the submerged area F has been recognized and detected (step S142). If it is determined that the submerged area F has not been detected (YES in step S142), the area detection means terminates the area detection processing and returns the process to the flood information generation control processing.

[0050] If it is determined that a submerged area F has been detected (YES in step S142), the area detection means determines whether two groups (pairs) of road edge object areas E have been detected (step S143). The two groups referred to here mean, as described above, each of the road edge objects Ae on both sides of the road.

[0051] If it is determined that two groups of road edge object regions E have not been detected ("NO" in step S143), the region detection means terminates the region detection process and returns the process to the flood information generation control process. If it is determined that two groups of road edge object regions E have been detected ("YES" in step S143), the CPU 11 estimates the inner boundary of each of the two groups of road edge object regions E, that is, the boundary on the side facing the flooded region F or the road surface region Rn (step S144). The region detection means estimates the boundary line Le by fitting it with a quadratic curve as described above.

[0052] The region detection means identifies the area between the two estimated boundary lines Le as the road region (step S145). The submerged region F outside this road region is excluded, and the area within the road region other than the submerged region F is identified as the road surface region Rn. The region detection means then terminates the region detection process and returns the process to the flood information generation control process.

[0053] Returning to the flood information generation control process in Figure 7, the CPU 11 determines whether the flooded area F and the pair of road edge object areas E have been detected (step S105). If it is determined that at least one of the flooded area F and the pair of road edge object areas E has not been detected ("NO" in step S105), the CPU 11 returns to step S101.

[0054] If it is determined that a submerged area F and a pair of road edge object areas E have been detected (YES in step S105), the CPU 11 (extraction means, extraction step) executes the inundation evaluation point identification process (step S106).

[0055] In the flood assessment point identification process shown in Figure 9, the extraction means selects a pixel (perimeter point) Pb on the outer perimeter of the identified flooded area F (step S161). The extraction means estimates a straight line Lc extending transversely through the selected pixel Pb based on the angle it makes with two boundary lines Le, and calculates the distance D between the center (center position C) and the selected pixel Pb in the transverse direction of the road area (step S162). The extraction means determines whether the calculated distance D is the minimum value among the distances D calculated for the outer perimeter of the flooded area F in this road image, that is, whether it is smaller than the minimum distance Dmin set at this point (step S163). If it is determined that it is the minimum value of the calculated distance D ("YES" in step S163), the extraction means sets the selected pixel Pb as the flood assessment point Pf and sets its distance D to the minimum distance Dmin from the center position C (step S164). If the flood assessment point Pf and the minimum distance Dmin from the center position C have already been set, the settings are overwritten. The extraction means determines whether or not all pixels on the outer perimeter of the submerged area F have been selected (step S165). If it is determined that not all pixels have been selected ("NO" in step S165), the extraction means returns to step S161. If it is determined that all pixels have been selected ("YES" in step S165), the extraction means terminates the flood evaluation point identification process and returns the process to the flood information generation control process.

[0056] If the determination process in step S163 determines that the calculated distance is not the minimum value ("NO" in step S163), the extraction means proceeds to step S165.

[0057] When the process returns from the flood assessment point identification process to the flood information generation control process, it returns to Figure 7, and the CPU 11 obtains area information 51 from the area information data server 50 (step S107).

[0058] The CPU 11 determines whether the shooting location is within the area where the level flooding method is applied (step S108). If it is determined that the shooting location is within the area where the level flooding method is applied ("YES" in step S108), the CPU 11 (terrain information acquisition means) obtains terrain data 61 from the terrain information data server 60, including the area where the level flooding method is applied related to the shooting location (step S109). Based on the terrain data 61, the CPU 11 determines the shooting direction of the road image (step S110). At this time, the CPU 11 may further specify and adjust the shooting location based on its positional relationship with the surrounding area being photographed, even more precisely than the shooting location determined by satellite positioning or the like.

[0059] The CPU 11 (calculation means) identifies the coordinates of the inundation assessment point Pf from the terrain data 61 and calculates its elevation (step S111). The calculation means applies the level flooding method to estimate the closed area below the calculated elevation as the inundation area (step S112). The calculation means generates inundation information indicating the estimated inundation area (step S113). The CPU 11 selects a distribution destination according to the inundation area and outputs the generated inundation information to the distribution destination terminal 80 selected by the communication unit 14 (step S114). After that, the processing of the CPU 11 returns to step S101.

[0060] If the determination process in step S108 determines that the shooting location is not an area to which the level flooding method applies ("NO" in step S108), the CPU 11 executes the flood risk assessment process (step S121).

[0061] As shown in Figure 10, in the flood risk assessment process, the CPU 11 (index calculation means, index calculation step) calculates a value Rmin = Dmin / W, which is obtained by normalizing the distance Dmin from the central position C of the flood assessment point Pf by the road width W (step S201).

[0062] The CPU 11 (risk determination means, risk determination step) determines whether the value Rmin is less than the reference value T1 (step S202). If it is determined that the value Rmin is less than the reference value T1 ("YES" in step S202), the risk determination means sets the risk of flooding to "high" (step S211). Then, the risk determination means proceeds to step S221.

[0063] If it is determined that the value Rmin is not less than the reference value T1 (i.e., greater than or equal to it) ("NO" in step S202), the risk determination means determines whether the value Rmin is less than or equal to a reference value T2 which is greater than the reference value T1 (step S203). If it is determined that the value Rmin is less than the reference value T2 ("YES" in step S203), the risk determination means sets the flood risk level to "medium" (step S212). Then, the risk determination means proceeds to step S221.

[0064] If it is determined that the value Rmin is not less than the reference value T2 (i.e., "YES" in step S203), the risk determination means sets the risk of flooding to "low" (step S213). Then, the risk determination means proceeds to step S221.

[0065] When the process moves from steps S211 to S213 to step S221, the CPU 11 (area information acquisition means) determines whether the shooting position is within a steep slope area (the area associated with the shooting position) (step S221). If it is determined that the shooting position is within a steep slope area ("YES" in step S221), the danger determination means generates and outputs flood risk information for the steep slope area (especially the flood risk area) and distributes it to a pre-set destination (step S222). Then, the danger determination means terminates the flood risk evaluation process and returns the process to the flood information generation control process. Returning to Figure 7, the CPU 11's process returns to step S101.

[0066] If it is determined that the shooting location is not within a steep slope area ("NO" in step S221), the CPU 11 generates and outputs flood information (inundation information) for the road at the shooting location and distributes it to a predetermined destination (step S223). Then, the CPU 11 terminates the flood risk assessment process and returns the process to the flood information generation control process. Returning to Figure 7, the CPU 11's process returns to step S101.

[0067] Furthermore, the processing in steps S101 to S103 may be performed on the image data server 20 as needed, and the image data to be sent to the analysis server 10 may be filtered in advance. That is, the analysis server 10 may acquire only road image data from the image data server 20 for the areas and periods for which detection processing of the submerged area F is necessary. This reduces the increase in load on the analysis server 10 due to pre-processing. In addition, the image data server 20 can quickly delete image data that is clearly unnecessary for the analysis of the flooding situation, thereby freeing up storage space. Moreover, the captured data from the fixed camera 73 may be sent to the image data server 20 only under specific weather conditions such as rain.

[0068] [Differentiation] This disclosure is not limited to the embodiments described above, and various modifications are possible. For example, in the above, flood information was generated for each road image, but it is also possible to generate flood information after obtaining flood assessment points Pf from multiple road image data within the same area (a certain geographical area), such as within an area where the level flooding method is applied. For example, if the process in step S111 is performed a standard number of times (e.g., 2 times) or more within the same area, the processes in steps S112 to S114 may be performed according to the elevation of the highest elevation flood assessment point within that same area.

[0069] If the elevation of flood assessment points is not obtained for a standard number of times or more within the first standard time, steps S112 to S114 may be performed based on the highest elevation of flood assessment points that are less than the standard number of times, or steps S112 to S114 may be skipped and the obtained flood assessment points may be deleted.

[0070] Furthermore, if flood assessment points are obtained, road image data for the same area may be preferentially acquired in step S101.

[0071] Furthermore, in the above embodiment, the elevation of the flood inundation assessment point was automatically calculated based on the terrain data, but it may also be configured so that the operator inputs the data. In this case, the CPU 11 of the analysis server 10 displays the road image, the flood inundation assessment point, and the terrain data on the display unit, and the operator reads the elevation of the flood inundation assessment point from this display content and inputs the said elevation via the operation reception unit. In this case, narrowing down the display target is effective from the viewpoint of reducing the burden on the operator. Specifically, the CPU 11 selects a predetermined number (1 to several) of road images from among multiple road images taken within the same level flooding method application area (a certain geographical area) in descending order of elevation at the shooting location, and displays the selected road images, their flood inundation assessment points, and terrain data near the shooting location.

[0072] Furthermore, in the above embodiment, the transverse direction of the road was defined as a direction perpendicular to the road's extension direction. However, other directions intersecting the road's extension direction, such as a direction parallel to one side of the edge of the road image, may also be defined. This makes it possible to further reduce the difference in distance from the shooting position to the surface being photographed.

[0073] Furthermore, in the above embodiment, if the road edge object area E on both sides of the road cannot be recognized, the process related to estimating the flood extent and determining the degree of risk is terminated. However, in such cases only, the operator may be allowed to manually set the boundary line Le. Conversely, if two boundary lines Le are identified, the specific location may be displayed and the operator may be required to approve it. In addition, the operator may be required to select some road images for approval. This can prevent the distribution of incorrect flood information.

[0074] Furthermore, in the above embodiment, semantic segmentation is used to recognize the submerged area F and the road edge object area E, but other image recognition techniques may be used. For example, the submerged area F and the road edge object area E may be recognized using a trained model modeled with gradient boosting, random forest, etc. In this case, the CPU 11, as a region detection means, for example, divides the image into meshes and calculates the mean and standard deviation of the RGB brightness values ​​in each mesh, and uses the obtained values ​​as explanatory variables. The CPU 11 generates a trained model by inputting the explanatory variables obtained from each of the many training images and updating the parameters so that the output value obtained matches the ground truth data of the training image.

[0075] Furthermore, although the above embodiment was described as performing both the estimation of the flooded area using the level flooding method and the determination of the flood risk in steep slope areas, it is not limited to this. The latter determination process does not have to be performed.

[0076] Furthermore, while the relative position of the flood assessment point Pf in the transverse direction of the road area was explained as an indicator for determining the flood risk, the indicator is not limited to this. For example, a histogram of the range (number of pixels) in which the submerged area F is detected in the transverse direction of the road area may be generated, and the flood risk may be evaluated according to the shape of the histogram (which may be converted into parameters that indicate the characteristics of the histogram and used).

[0077] Furthermore, while the above explanation described conducting flood risk assessments on steep slopes included in warning zones designated under the Landslide Disaster Prevention Act, this is not limited to such areas. Even on gentle slopes where landslides are unlikely, assessments may be set appropriately according to water flow and other factors. In addition, the distribution of flood risk information related to risk assessments may be separate from the distribution of disaster risk information such as landslides and debris flows, with only flood risk information being distributed, or it may be distributed together with information on other disaster risks.

[0078] Furthermore, while the above description uses a storage unit 13 consisting of non-volatile memory such as an HDD or flash memory as an example of a computer-readable medium for storing the program 132 related to the generation and control of flood information, the invention is not limited to these. Other computer-readable media include other non-volatile memories such as MRAM, and portable recording media such as CD-ROMs and DVD discs. In addition, a carrier wave can also be used as a medium for providing program data according to the present invention via a communication line.

[0079] As described above, the analysis server 10, which is a flood information generation device in this embodiment, is equipped with a CPU 11. The CPU 11, as an image acquisition means, acquires road images of roads and information on the location where the road images were taken. As a region detection means, it detects from the road images road edge object regions E, which are objects located at the edge of the road, road regions sandwiched between pairs of road edge object regions E, and flooded regions F, which are the flooded parts of the road. As an extraction means, it extracts the position closest to the center of the road in the direction crossing the road region from the outer edge of the flooded region F within the road region as a flood evaluation point Pf. By setting the evaluation point for generating flood information on the outer edge of the road's flooded area, and at the position closest to the center of the road, the elevation of the flooded surface can be determined more easily and accurately. This enables the analysis server 10 to generate flood information more quickly. In particular, since the road area is defined by pairs of road edge object areas E, it is easier to appropriately identify the extent of the road R even when flooded, such as on single-lane roads or roads where the center line is not located in the middle. Furthermore, since the center of the road area can be preferentially designated as the flood evaluation point Pf, the calculation of elevation becomes easier. In addition, even if the flood evaluation point Pf is not in the center of the road area, it is determined by the distance from the center, so the flooded surface can be defined while appropriately considering the gradient in the transverse direction of the road.

[0080] Furthermore, the CPU 11, as a region detection means, estimates the road edge by approximating the boundary line Le on the side of the road edge object region E that is closer to the submerged region F with a curve. In this way, by estimating the road edge based on characteristic objects that extend above the road surface, it is easier to identify the road edge even when the road is flooded. In addition, even if the entire road edge cannot be identified, the boundary line between the road edge and the road region can be appropriately determined by approximation.

[0081] Furthermore, the analysis server 10 includes a storage unit 13 as a model storage means that stores a pre-trained model that has been trained to detect and output submerged areas F and road edge object areas E in response to input road images, using multiple example images including road images having submerged areas F. The CPU 11, as a region detection means, inputs the road image acquired by the image acquisition means into the pre-trained model to detect submerged areas F and road edge object areas E in the road image. In this way, by using machine learning models to detect desired areas from image data, the amount of manual processing by operators can be significantly reduced. Furthermore, since the limitations and variability of operator skill can be avoided, it is possible to obtain the information necessary for estimating the extent of flooding more stably and quickly from a large amount of image data.

[0082] Furthermore, the CPU 11 acquires terrain data 61, which includes the three-dimensional coordinates of the road R and the surrounding area of ​​the road R, as a terrain information acquisition means, and calculates the elevation of the flood inundation assessment point Pf based on the acquired terrain data 61 and the location where the road image was taken, as a calculation means. As described above, the elevation of the flood inundation assessment point Pf, which is easily and appropriately determined by utilizing the characteristics of the road surface gradient, can be easily identified based on the terrain data 61, making it possible to estimate the extent of flooding more easily than before.

[0083] Furthermore, the CPU 11, as a calculation means, applies the level flooding method based on the elevation of the flood assessment point Pf and topographic information to generate flood information for the area surrounding the road R (flooded area F) related to the captured image. According to the level flooding method, within an appropriate set range, the elevation of the flooded surface is constant, so by easily and appropriately identifying the elevation of the water surface in the flooded area F on the road R, the flooded area below that elevation can be easily identified from the topographic data. Therefore, the analysis server 10 can easily generate and distribute flood information related to the estimation of the flooded area based on the elevation of the flooded surface quickly obtained from a large amount of image data.

[0084] Furthermore, the CPU 11 may, as an image acquisition means, acquire multiple road images and information on the shooting locations of said road images within a geographic area such as an area where the level flooding method is applied, and as a calculation means, apply the level flooding method based on the highest elevation among the elevations of flood evaluation points Pf extracted from the multiple road images and the topographic data 61 within the geographic area to generate flood information for said geographic area. Depending on the rate of water inflow into the flooded area, the elevation of the flooded surface may not always be sufficiently obtained from the shooting data of a single location, but by using road images from multiple locations in this way, more accurate flood information can be generated. In particular, by using the water surface at the highest elevation, it is possible to suppress underestimation of the flooded area and reduce the occurrence of obstacles and accidents due to estimation errors.

[0085] Furthermore, the flood information generation method of this embodiment includes an image acquisition step of acquiring a road image of a road and information on the location where the road image was taken; a region detection step of detecting from the road image a road edge object region E which is an object located at the edge of the road, a road region sandwiched between pairs of road edge object regions E, and a flooded region F which is the submerged portion of road R; and an extraction step of extracting the position closest to the center of road R in the direction crossing the road region from the outer edge of the flooded region F in the road region as a flood evaluation point Pf. This flood information generation method allows for easier and more accurate identification of the elevation of the flooded surface by setting the evaluation point for generating flood information on the outer edge of the road's flooded area, and at the position closest to the center of the road. This makes it possible to generate flood information more quickly.

[0086] Furthermore, by installing the program 132 related to the above-mentioned flood information generation method into a computer and executing it using the CPU 11, it becomes possible to determine the elevation of the flooded surface more quickly and accurately than before. This enables the generation and output of flood information with greater accuracy and speed.

[0087] The specific configurations, processing operations, and procedures shown in the above embodiments can be modified as appropriate without departing from the spirit of the present invention. The scope of the present invention includes the scope of the invention described in the claims and its equivalents. [Explanation of Symbols]

[0088] 10 Analysis Server 11 CPU 12 RAM 13 Storage section 131 Pre-trained models 132 Programs 14 Communications Department 20 Image Data Server 30 Weather information distribution server 40 People flow information server 50 Area Information Data Server 51 Area Information 60. Topographic Information Data Server 61. Topographic data 70 Imaging device 71 In-vehicle terminals 72 Member terminals 73 Fixed Cameras 80 Destination devices 100 Flood Information Generation System Af flooded area Ae Road edge object E Road edge object area F Submerged area Le boundary line Pf Flood Inundation Assessment Points Rn road surface area

Claims

1. An image acquisition means for acquiring road images of roads and information on the location where said road images were taken, A region detection means for detecting, from the aforementioned road image, a road edge object region corresponding to an object located at the edge of the road, a road region sandwiched between a pair of road edge object regions, and a submerged region which is the submerged part of the road, An extraction means for extracting the point closest to the highest elevation of the road, based on the gradient in the direction crossing the road area, from the outer edge of the submerged area within the road area, as a flood evaluation point. A flood information generation device characterized by being equipped with the following features.

2. The flood information generating device according to claim 1, characterized in that the area detection means estimates the edge of the road by approximating the boundary line on the side of the road edge object area that is closer to the submerged area with a curve.

3. The system includes a model storage means that stores a pre-trained model that uses multiple example images, including road images having the aforementioned submerged region, to detect and output the submerged region and the road edge object region in response to a road image input. The region detection means inputs the road image acquired by the image acquisition means into the trained model to detect the submerged region and the road edge object region in the road image. The flood information generating device according to claim 1 or 2, characterized in that it is a flood information generating device.

4. A means for acquiring terrain information that acquires terrain information including the three-dimensional coordinates of the terrain of the road and the surrounding area of ​​the road, A calculation means for calculating the elevation of the flood assessment point based on the topographic information and the shooting location, A flood information generating device according to any one of claims 1 to 3, characterized by comprising:

5. The inundation information generating device according to claim 4, characterized in that the calculation means generates inundation information for the surrounding area by applying the level flooding method based on the elevation of the inundation evaluation point and the topographic information.

6. The image acquisition means acquires information on multiple road images and the location where the road images were taken within a certain geographical area. The calculation means generates flood information for the geographic area by applying the level flooding method based on the highest elevation among the elevations of the flood assessment points extracted by the extraction means from the plurality of road images and the topographic information within the geographic area. The flood information generating device according to claim 4.

7. Image acquisition step: Obtain road images of roads and information on the location where the road images were taken. A region detection step that detects, from the aforementioned road image, a road edge object region corresponding to an object located at the edge of the road, a road region sandwiched between a pair of road edge object regions, and a submerged region which is the submerged part of the road. Extraction step of extracting the point closest to the highest elevation of the road based on the gradient in the direction crossing the road area, among the outer edges of the submerged area in the road area, as the inundation evaluation point. A method for generating flood information, characterized by including the following:

8. Computers, Image acquisition means for acquiring road images of roads and information on the location where said road images were taken. Region detection means for detecting, from the aforementioned road image, a road edge object region corresponding to an object located at the edge of the road, a road region sandwiched between a pair of road edge object regions, and a submerged region which is the submerged part of the road. Extraction means for extracting the point closest to the highest elevation of the road, based on the gradient in the direction crossing the road area, from the outer edge of the submerged area within the road area, as a flood evaluation point. A program designed to function as such.

Images