Geological feature change detection device and geological feature change detection method

Abstract

To provide a ground object change detection device capable of providing information about ground object change by reducing a difference between a position of a ground object acquired from an image and a position of an actual ground object.SOLUTION: A ground object change detection device includes: a ground object mask estimation part 21 for generating a ground object mask 36 from each of a comparison object image 35 obtained by imaging a ground object from the sky and a past image 31 obtained by imaging the ground object from the sky in the past; a ground object mask correction part 22 for calculating multiplicity each of ground object area information 33 that registers position information on the ground object, a ground mask 36a generated from a past image 31, and a ground object mask 36b generated from the comparison object image 35, with respect to ground objects registered in ground object area information 33, and registering an area corresponding to a ground object whose multiplicity is equal to or more than a threshold in corrected mask information 37; and a ground object change detection part 23 for calculating a difference between corrected mask information 37a acquired from the past image 31 and corrected mask information 37b acquired from the comparison object image 35 to detect time change of the ground object.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a ground object change detection device and a ground object change detection method for detecting changes in ground objects.

Background Art

[0002] Sensing from above by an aircraft or the like and sensing from a satellite are used to grasp the situation over a wide area on the ground, and are utilized, for example, to grasp the damage situation at the time of a disaster. In addition, due to the development of wireless communication technology and the miniaturization and high performance of hardware such as cameras, real-time sensing using drones has become common today.

[0003] At the time of a disaster, it is very important to quickly grasp the damage situation caused by the disaster in order to perform evacuation guidance and rescue activities quickly and accurately. Elements for measuring the damage situation include, for example, information representing the area of earth and sand outflow, information representing the flooded area due to flooding, and information representing the collapse of buildings, breakage / sinking of roads, etc. By quickly and accurately grasping this information, the damage caused by the disaster can be minimized.

[0004] As a conventional method for recognizing the damage situation caused by a disaster, for example, by using a recognition model that has learned image data such as an image taken in advance of a predetermined disaster situation for an image taken from above after the disaster, it is possible to determine whether or not a disaster has occurred in the captured image after the disaster, or to display the damaged location.

[0005] Patent Document 1 describes a device for estimating disaster damage to a building from images taken from above and building polygons acquired before a disaster occurs. Building polygons are data representing the two-dimensional shape of a building (location and area of ​​the building) on ​​the ground surface. This device includes a learning model for estimating the damage status of a building corresponding to a building polygon. The device sets the condition of the damage, such as the state of blue tarpaulin covering, as the target damage status, outputs the estimation results of the learning model as feature images, and estimates the disaster damage occurring to the building by referring to the feature image corresponding to each building polygon. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Japanese Patent Publication No. 2019-95886 [Overview of the project] [Problems that the invention aims to solve]

[0007] Generally, optical images captured from above or at high altitudes by cameras suffer from geometric distortion due to lens aberrations, which can cause discrepancies between the buildings in the image and pre-acquired building polygons. Conventional techniques, even when lens aberrations are eliminated by calibrating the camera according to a predetermined procedure, still have the problem that the positions of buildings in the image and the building polygons will not match unless the captured image is aligned with an actual map. As a result, the location of buildings in feature images will differ from reality, making it difficult to grasp the extent of damage caused by disasters accurately and quickly.

[0008] Therefore, there is a need for technology that can reduce the difference (positional discrepancy) between the position of geographic features (objects on the ground) in images taken from above and the actual position of geographic features, thereby enabling accurate detection of changes in geographic features.

[0009] The object of the present invention is to provide a feature change detection device and a feature change detection method that can reduce the difference between the location of a feature obtained from an image and the actual location of the feature, and provide information about changes in the feature. [Means for solving the problem]

[0010] The feature change detection device according to the present invention includes: a camera that captures a comparison image, which is an image of a feature taken from above or a high place; a feature mask estimation unit that generates a feature mask representing the location information of the feature from a past image, which is an image of the feature taken from above or a high place at a time prior to the time the comparison image was taken, and from the comparison image; feature area information in which the location information of the feature is registered in advance; a feature mask correction unit that calculates the degree of overlap with the feature registered in the feature area information for each of the feature masks generated from the past image and the feature mask generated from the comparison image, and registers the area corresponding to the feature whose degree of overlap with the feature is greater than or equal to a threshold as a corrected feature mask in the corrected mask information; and a feature change detection unit that detects the temporal change of the feature by calculating the difference between the corrected mask information obtained using the feature mask of the past image and the corrected mask information obtained using the feature mask of the comparison image.

[0011] The present invention provides a method for detecting changes in geographic features, which is performed on a geographic feature change detection device equipped with geographic feature area information in which the location information of the geographic feature is pre-registered, and includes a geographic feature mask estimation step in which the geographic feature change detection device generates a geographic feature mask representing the location information of the geographic feature from both a past image, which is an image of the geographic feature taken from the air or a high place at a time prior to the time the comparison image was taken, and the comparison image, and the geographic feature change detection device generates from the geographic feature mask generated from the past image and the comparison image The device includes a feature mask correction step in which, for each of the feature masks obtained, the degree of overlap with the feature registered in the feature area information is calculated, and the area corresponding to the feature whose degree of overlap with the feature is greater than or equal to a threshold is registered in the corrected mask information as the corrected feature mask; and a feature change detection step in which the feature change detection device detects the temporal change of the feature by calculating the difference between the corrected mask information obtained using the feature mask of the past image and the corrected mask information obtained using the feature mask of the comparison target image. [Effects of the Invention]

[0012] According to the present invention, a feature change detection device and a feature change detection method can be provided that can reduce the difference between the location of a feature obtained from an image and the actual location of the feature, and provide information regarding changes in the feature. [Brief explanation of the drawing]

[0013] [Figure 1] A functional configuration diagram of a geological change detection device according to Embodiment 1 of the present invention. [Figure 2] A diagram showing an example of the hardware configuration of a geographical feature change detection device. [Figure 3] This diagram illustrates the process performed by the geological feature change detection device in Example 1. [Figure 4] A diagram illustrating an example of the processing steps in the feature mask correction section. [Figure 5] An example flowchart showing the processes performed by a geological feature change detection device. [Figure 6]This diagram shows an example of a display screen output by a geological feature change detection device. [Figure 7] A functional configuration diagram of a geological change detection device according to Embodiment 2 of the present invention. [Figure 8] This diagram illustrates the process performed by the geological feature change detection device in Example 2. [Modes for carrying out the invention]

[0014] The geological feature change detection device and method according to the present invention detect changes in geological features that occurred between the time points in which the images were taken, using at least two images of geological features taken from the air or a high place at different time points and location information of the geological features acquired in advance. According to the present invention, the difference between the location of the geological feature obtained from the image and the actual location of the geological feature (location of the geological feature acquired in advance) can be reduced, providing information on changes in geological features and enabling accurate detection of changes in geological features.

[0015] Conventional technologies require the acquisition and training of a large number of images of buildings affected by disasters that cause changes in geographic features in order to create a learning model for recognizing geographic features. However, disasters generally occur infrequently, making it extremely difficult to acquire a large number of images suitable for training. In this invention, it is not necessary to acquire a large number of images required for training in order to detect changes in geographic features.

[0016] Furthermore, with conventional technology, it is difficult to determine whether or not a geographical feature (such as a building) has disappeared. This is because information simply obtained from an image cannot distinguish whether or not the feature never existed in the first place or has disappeared. For example, it is impossible to determine whether there was an error in the building polygon information and the building never existed in the first place, or whether the building disappeared due to a disaster. With the present invention, it is possible to easily determine whether or not a building has disappeared.

[0017] The following describes a geological feature change detection device and geological feature change detection method according to an embodiment of the present invention, with reference to the drawings. The geological feature change detection method according to an embodiment of the present invention is performed by the geological feature change detection device according to an embodiment of the present invention.

[0018] In this specification, a ground feature refers to an object existing on the ground, including, for example, artificial objects such as buildings, roads, and power transmission lines, and natural objects such as vegetation. The change of a ground feature refers to, for example, a change in the shape or position of the ground feature, or the disappearance or appearance of the ground feature. The position information of a ground feature is information indicating the position and area (outline) of the ground feature, and is also simply referred to as the position of the ground feature.

[0019] In the drawings used in this specification, the same or corresponding components are denoted by the same reference numerals, and repeated descriptions of these components may be omitted.

Example

[0020] The ground feature change detection device and the ground feature change detection method according to Embodiment 1 of the present invention will be described.

[0021] FIG. 1 is a functional configuration diagram of a ground feature change detection device 10 according to this embodiment. The ground feature change detection device 10 according to this embodiment includes a camera 11 and functional units of a ground feature mask estimation unit 21, a ground feature mask correction unit 22, and a ground feature change detection unit 23. Further, the ground feature change detection device 10 includes a past image 31, a ground feature mask estimation model 32, and ground feature area information 33.

[0022] The camera 11 is an imaging device that captures an image of the ground from above or a high place. The camera 11 may be fixed at an arbitrary position, or may be installed on a controlled moving body such as an aircraft or a drone. Hereinafter, the image captured by the camera 11 is referred to as a comparison target image 35. The comparison target image 35 includes the ground features captured by the camera 11.

[0023] The past image 31 is an image captured from above or a high place of the ground at a time point before the camera 11 captures the comparison target image 35. The position information of the past ground features can be obtained from the past image 31. The past image 31 is stored in the ground feature change detection device 10.

[0024] The feature mask estimation model 32 is a model for generating a feature mask 36 from the comparison image 35 and the past image 31. The feature mask 36 is location information of features obtained from the image, that is, information representing the location and area of ​​features obtained from the image. The feature mask 36 includes masks for one or more features. The feature mask estimation model 32 is, for example, a machine learning model that recognizes the location and area of ​​features from an image, and can be created in advance by machine learning using images provided for training. The feature mask estimation model 32 is stored in the feature change detection device 10.

[0025] The feature mask estimation unit 21 estimates the feature mask 36 using the feature mask estimation model 32. Specifically, the feature mask estimation unit 21 takes the comparison image 35 captured by the camera 11 and past images 31 that overlap with the comparison image 35 in at least part, as input, and uses the feature mask estimation model 32 to generate a feature mask 36 for both the comparison image 35 and the past images 31. The feature mask estimation unit 21 outputs the generated feature mask 36 to the feature mask correction unit 22.

[0026] The feature mask 36 can be represented by the location information of the feature (information indicating the location and area of ​​the feature), or by information from which the location information of the feature can be obtained through any processing. The feature mask 36 can be represented in any format; for example, it may be represented using the vertices of the polygon that forms the outline of the feature, or a polygon that represents the feature.

[0027] The feature area information 33 is pre-registered information indicating the location and area of ​​a feature, and is stored in the feature change detection device 10. The feature area information 33 contains location information for predetermined features. The feature area information 33 is information obtained from publicly available information such as maps, and may also be information created by the operator or administrator of the feature change detection device 10. For example, the feature area information 33 may be bitmap data that distinguishes between the area where a feature exists and other areas, or it may be polygon information representing the outline of a feature, such as the base map information published by the Geospatial Information Authority of Japan.

[0028] Furthermore, the feature area information 33 may also register information about features formed when a single feature is divided into multiple features, or information about features formed when multiple features are merged into a single feature. In this way, the unit of features (the scope treated as one feature) registered in the feature area information 33 can be arbitrarily defined.

[0029] The feature mask correction unit 22 calculates the degree of overlap between the feature mask 36a generated from the past image 31 and the feature mask 36b generated from the comparison image 35, and the feature registered in the feature area information 33. If the feature registered in the feature area information 33 and the feature masks 36a and 36b satisfy predetermined determination conditions using the degree of overlap, the unit determines that the feature exists. The feature mask correction unit 22 then creates and outputs corrected mask information 37.

[0030] Corrected mask information 37 is information obtained by correcting the feature masks 36 generated from the past image 31 and the comparison image 35 using feature area information 33. In the corrected mask information 37, for features represented by the feature mask 36 whose existence has been confirmed by the feature area information 33, the area corresponding to this feature is registered as a corrected feature mask. The corrected mask information 37 also records whether or not each feature registered in the feature area information 33 exists. The corrected mask information 37 can be expressed in any data format, such as text data or image data.

[0031] The feature mask correction unit 22 performs the above processing to create corrected mask information 37a using the feature mask 36a generated from the past image 31, and to create corrected mask information 37b using the feature mask 36b generated from the comparison target image 35.

[0032] The feature mask correction unit 22 will be explained in detail.

[0033] First, the feature mask correction unit 22 obtains the feature mask 36 from the feature mask estimation unit 21. Next, the feature mask correction unit 22 extracts from the feature area information 33 the area in which the features included in the obtained feature mask 36 exist (i.e., the area corresponding to the imaging area of ​​the past image 31 and the comparison image 35). Hereafter, the extracted area of ​​the feature area information 33 will be referred to as the reference area of ​​the feature area information 33.

[0034] In this embodiment, as an example, we will explain the case where the entire imaging area of ​​the past image 31 and the comparison image 35 match each other. If only a portion of the imaging area of ​​the past image 31 and the comparison image 35 match each other, feature area information 33 corresponding to the matching area may be extracted. Even in this case, the same effect as in this embodiment can be obtained.

[0035] Next, the feature mask correction unit 22 aligns the acquired feature mask 36 with the reference area of ​​the feature area information 33. The feature mask correction unit 22 performs this alignment by any method. For example, the imaging area of ​​the camera 11 may be calculated from external parameters representing the installation position and orientation of the camera 11, and the feature area information 33 corresponding to this imaging area may be determined, thereby aligning the position of the feature mask 36b with the reference area of ​​the feature area information 33. Alternatively, the position information of the camera 11 or the mobile object on which the camera 11 is installed (e.g., an aircraft or drone) recorded when the comparison image 35 was captured may be obtained from GPS information, etc., and this position information may be used to align the position of the feature mask 36b with the reference area of ​​the feature area information 33.

[0036] Next, the feature mask correction unit 22 selects one feature from among the features included in the reference area of ​​the feature area information 33. Then, the feature mask correction unit 22 determines whether or not the selected feature exists based on the degree of overlap between the selected feature and the mask corresponding to the selected feature among the feature masks 36. The feature mask correction unit 22 repeats the above process for other features included in the reference area of ​​the feature area information 33 to create corrected mask information 37.

[0037] The feature mask correction unit 22 can use, for example, the overlapping area between the area of ​​the selected feature (the feature included in the feature area information 33) and the mask corresponding to the selected feature in the feature mask 36 as the degree of overlap used to determine whether or not the selected feature exists. Alternatively, the feature mask correction unit 22 may use the ratio of the above-mentioned overlapping area to the area of ​​the selected feature's region as the degree of overlap. These areas may be represented, for example, by the number of pixels that make up the area on the image.

[0038] The feature mask correction unit 22 may redefine a feature formed by dividing a single feature registered in the feature area information 33 into multiple features as a new feature, or it may redefine a feature formed by integrating multiple features into a single feature as a new feature. For example, if the feature is a building, one building may be divided into four parts, and each of these four features may be registered as a new feature. In this way, if only a part of a single building is destroyed by a disaster or the like, the loss of a part of the building can be detected.

[0039] Furthermore, the feature mask correction unit 22 may specify a particular area of ​​the feature and change the method for determining the presence or absence of the feature in this area compared to the method for determining it in other areas. For example, if the feature is a reservoir dam, the vertices of the reservoir dam's outline can often be recognized, but the area in other parts where recognition is possible changes dynamically, for example, depending on the amount of water stored in the reservoir. In such cases, the area other than the vertices of the reservoir dam's outline can be specified as a particular area, and in this particular area, the threshold for the degree of overlap used for determination can be set lower than that of the vertices of the reservoir dam, or dynamically changed according to the amount of water stored in the reservoir dam received from the input unit 12 shown in Figure 2, which will be described later.

[0040] The feature change detection unit 23 detects changes in features over time between the time the past image 31 was captured and the time the comparison image 35 was captured. The feature change detection unit 23 calculates the difference between the corrected mask information 37a obtained using the feature mask 36a of the past image 31 and the corrected mask information 37b obtained using the feature mask 36b of the comparison image 35. The difference calculated by the feature change detection unit 23 (the difference between the corrected mask information 37a and the corrected mask information 37b) represents the features that have changed from the time the past image 31 was captured to the time the comparison image 35 was captured.

[0041] The feature change detection unit 23 extracts the difference between at least two corrected mask information sets 37 created from images captured at different points in time. This difference represents the temporal change of the feature detected by the feature change detection unit 23. The feature change detection unit 23 outputs the obtained difference as feature change information 34.

[0042] The feature change information 34 is information about the temporal changes of features detected by the feature change detection unit 23. The feature change information 34 may include information recording whether or not there have been changes in the features included in the comparison image 35.

[0043] Figure 2 shows an example of the hardware configuration of the feature change detection device 10 according to this embodiment. In addition to the camera 11 shown in Figure 1, the feature change detection device 10 includes an input unit 12, an output unit 13, a CPU (Central Processing Unit) 14, a main memory 15, and an auxiliary memory 16.

[0044] The input unit 12 can be configured with, for example, a keyboard or mouse, and accepts input from the operator of the geological feature change detection device 10. The input unit 12 also receives images and information required by the geological feature change detection device 10, such as the comparison image 35 captured by the camera 11.

[0045] The output unit 13 has a screen, such as a liquid crystal display, and is used to display various information to the operator. The output unit 13 may also be equipped with a storage device, such as a hard disk drive, and may, for example, save the output of the geological change detection device 10 in the form of image data.

[0046] The CPU 14 controls the feature change detection device 10 and performs data calculations. The CPU 14 loads various programs and data into the main memory 15 and executes these programs to realize the functions of the feature change detection device 10 (for example, the feature mask estimation unit 21, the feature mask correction unit 22, and the feature change detection unit 23).

[0047] The main memory 15 can be configured as, for example, RAM (Random Access Memory). The main memory 15 contains programs that implement functions such as the feature mask estimation unit 21, the feature mask correction unit 22, and the feature change detection unit 23.

[0048] The auxiliary storage device 16 can be configured as, for example, a hard disk drive. The auxiliary storage device 16 stores various types of data, such as past images 31, feature mask estimation models 32, and feature area information 33.

[0049] Figure 3 is a diagram illustrating the processing performed by the feature change detection device 10 in Example 1. Figure 3 shows a past image 31, a comparison image 35, a feature mask 36a generated from the past image 31, a feature mask 36b generated from the comparison image 35, corrected mask information 37a obtained by correcting the feature mask 36a, and corrected mask information 37b obtained by correcting the feature mask 36b.

[0050] Figure 3 shows an example where the feature is a building, and the feature change detection device 10 detects changes in the imaged building. The feature mask 36 includes the building mask. The corrected mask information 37 includes the corrected building mask, which is the corrected building mask.

[0051] Past image 31 includes buildings T1, T2, T3, T4, and T5 as features.

[0052] Comparison image 35 was taken after past image 31 was taken. Assume that a landslide occurred between the time past image 31 was taken and the time comparison image 35 was taken. Comparison image 35 shows buildings T1, T2, T4, and T5 after they have been swept away and disappeared due to the mudslide. In comparison image 35, building T3 is still present.

[0053] The feature mask estimation unit 21 receives the past image 31 and the comparison image 35 as input, generates a feature mask 36a for the past image 31, and generates a feature mask 36b for the comparison image 35.

[0054] In the example shown in Figure 3, the feature mask estimation unit 21 uses semantic segmentation, a type of machine learning algorithm that determines whether or not each pixel in an image is included in the recognition target, as the feature mask estimation model 32. For example, the feature mask estimation unit 21 estimates a feature mask 36 in the image, where areas containing buildings are represented by 1 and other areas are represented by 0. Figure 3 shows this feature mask 36 displayed in an image as feature mask 36a and feature mask 36b.

[0055] Generally, estimation results from machine learning algorithms, including semantic segmentation, contain errors. For example, in the example shown in Figure 3, building T4 is actually a rectangle that is long in the north-south direction (vertical direction in Figure 3), as shown in past image 31. However, in feature mask 36a, the shape of the mask corresponding to building T4 is elliptical, and a mask that does not follow the original contour of building T4 without deviation is not obtained. Such errors are common in machine learning algorithms.

[0056] Therefore, if the change in a feature is determined by calculating the difference between feature mask 36a and feature mask 36b, as in conventional techniques, the above-mentioned error (the error between the actual feature and the mask representing this feature in feature mask 36) may be detected as a difference. Thus, conventional techniques have the problem of confusing the above-mentioned error with the changes in the building that should be detected.

[0057] In this embodiment, to solve this problem, the feature mask correction unit 22 corrects the feature mask 36 using the feature area included in the feature area information 33. Specifically, the feature mask correction unit 22 determines the degree of overlap between the feature mask 36a and the features registered in the feature area information 33, and determines that the feature exists if the feature and the feature mask 36a satisfy predetermined determination conditions using the degree of overlap. Then, the feature mask correction unit 22 determines the degree of overlap between the feature mask 36b and the features registered in the feature area information 33, and determines that the feature exists if the feature and the feature mask 36b satisfy predetermined determination conditions using the degree of overlap.

[0058] The feature mask correction unit 22 performs such a determination and corrects the feature masks 36a and 36b by determining the feature masks 36a and 36b corresponding to the existing feature, thereby obtaining corrected mask information 37a obtained by correcting feature mask 36a and corrected mask information 37b obtained by correcting feature mask 36b.

[0059] The feature mask correction unit 22 performs the above processing on each of the features registered in the feature area information 33, thereby creating and outputting corrected mask information 37a and corrected mask information 37b.

[0060] The feature change detection unit 23 extracts and calculates the difference between the corrected mask information 37a and the corrected mask information 37b, and outputs the calculated difference as feature change information 34.

[0061] The following describes the processing details of the feature mask correction unit 22.

[0062] Figure 4 illustrates an example of the processing content of the feature mask correction unit 22. As an example, Figure 4 shows a case where the feature to be detected is a building, and the feature area information 33 is polygon information representing the outline of the feature.

[0063] Figure 4 shows feature mask 36c as a feature mask 36 that represents feature masks 36a and 36b. Feature mask 36c includes four masks M as features (buildings), namely mask M1, mask M2, mask M3, and mask M4.

[0064] Furthermore, Figure 4 shows feature area information 33c as the reference area of ​​feature area information 33 (the area where features included in feature mask 36 exist). Feature area information 33c contains four feature areas R, namely feature area R1, feature area R2, feature area R3, and feature area R4.

[0065] Furthermore, Figure 4 shows an overlay diagram 41, which is a diagram in which the feature mask 36c and the feature area information 33c are superimposed.

[0066] Furthermore, Figure 4 shows a diagram illustrating the corrected mask information 37c obtained from the feature mask 36c and feature area information 33c using the superimposed diagram 41.

[0067] In the example shown in Figure 4, the feature mask correction unit 22 receives the feature mask 36c as input, calculates the degree of overlap between the mask M included in the feature mask 36c and the feature (feature area R) registered in the feature area information 33c, and creates corrected mask information 37c. The feature area R is assumed to be represented by a polygon that represents the outline of a building. Masks M1 to M4 correspond to feature areas R1 to R4, respectively.

[0068] Furthermore, in the example shown in Figure 4, the feature mask correction unit 22 uses the overlap ratio as the overlap degree, which is the criterion for determining whether or not each building exists. The overlap ratio is the ratio of the overlapping area to the area of ​​the feature region R. The overlapping area is the area where the feature region R and the area of ​​the feature mask 36c corresponding to the feature region R (mask M) overlap. A threshold of 50% is used for the overlap ratio. That is, the feature mask correction unit 22 determines that a building exists if the overlap ratio is 50% or more. Note that this threshold is not limited to 50% and can be set to any value.

[0069] In the following explanation, the conditions used by the feature mask correction unit 22 to determine whether or not a feature exists, based on the degree of overlap, are referred to as predetermined determination conditions. In the example above, the predetermined determination condition is that a building is determined to exist if the degree of overlap is greater than or equal to a threshold (overlap rate of 50% or more).

[0070] Furthermore, the criteria for determining whether or not a building (landmark) exists are not limited to the overlap ratio; any other criteria can be used, for example, the absolute value of the overlapping area. As mentioned above, any value can be used for the threshold of the determination criteria. These determination criteria and thresholds may be stored in advance in the auxiliary storage device 16, or they may be set by the operator of the landmark change detection device 10 via the input unit 12.

[0071] In the example shown in Figure 4, when the feature mask 36c and the feature area information 33c are superimposed, as shown in the superimposed figure 41, the ratio of the overlapping area between feature area R1 and mask M1 to the area of ​​feature area R1 (overlap ratio) is 50% or more. Since the overlap ratio is above the threshold (50%), mask M1 has successfully recognized the building corresponding to feature area R1.

[0072] The feature mask correction unit 22 determines that a building exists in feature area R1 because the overlap rate for feature area R1 is greater than or equal to a threshold. Then, in the corrected mask information 37c, the feature mask correction unit 22 designates the area corresponding to feature area R1 as the area where a building exists. In other words, the feature mask correction unit 22 records the feature area R1 indicated by mask M1 as an existing building in the corrected mask information 37c.

[0073] The feature mask correction unit 22 determines that a building exists in feature area R2 because the overlap ratio of mask M2 with respect to feature area R2 is greater than or equal to a threshold (overlay diagram 41), and in the corrected mask information 37c, it designates the area corresponding to feature area R2 as the area where a building exists. In other words, the feature mask correction unit 22 records the feature area R2 indicated by mask M2 as an existing building in the corrected mask information 37c.

[0074] Furthermore, feature area R4 does not overlap with any of the masks M1 to M4 included in feature mask 36c, and the overlapping area and overlapping ratio are zero (overlay diagram 41). The feature mask correction unit 22 determines that there is no building corresponding to feature area R4 because the overlapping ratio for feature area R4 is less than the threshold.

[0075] Furthermore, in the example shown in Figure 4, as shown in the superimposed diagram 41, the mask M3 does not overlap with any of the feature areas R registered in the feature area information 33c. Therefore, it can be inferred that the mask M3 is a mask that appeared because the feature mask estimation unit 21 misidentified a building.

[0076] Furthermore, as shown in the superimposed diagram 41, although mask M4 overlaps with feature area R3, the overlap rate is less than the threshold (50%). Therefore, it can be inferred that mask M4 is a mask that appeared because the feature mask estimation unit 21 misidentified a building. The feature mask correction unit 22 then determines that there is no building corresponding to feature area R3.

[0077] The feature mask correction unit 22 performs the above processing for each of the feature regions R1 to R4 to create corrected mask information 37c. In the corrected mask information 37c, the feature mask 36c is corrected using the feature region information 33c, and corrected building masks M11 and M22 corresponding to masks M1 and M2 are shown. In addition, in the corrected mask information 37c, the contour shapes of the corrected building masks M11 and M22 are corrected by the feature region information 33c, and are changed from masks M1 and M2 to match feature regions R1 and R2, respectively.

[0078] In Figure 4, as an example, the corrected mask information 37c is shown as information output as a bitmap image containing the corrected feature area information. The corrected mask information 37 may also be, for example, polygon information representing the outline of a feature, and can be expressed in any data format.

[0079] As mentioned above, even when using a model that has been thoroughly trained on data, general machine learning algorithms can still produce minute errors in generating the feature mask 36. Therefore, in conventional techniques, when detecting changes in features by calculating the difference between the feature mask 36a and feature mask 36b generated for each of the past image 31 and the comparison image 35, not only the difference due to the actual change in the feature appears as noise, but also the difference due to minute errors in the feature mask 36. As a result, in conventional techniques, false alarms and missed alarms can occur due to this noise.

[0080] In this embodiment, corrected mask information 37, which is information obtained by correcting the feature mask 36 using feature area information 33, is used to compare corrected mask information 37a obtained from past images 31 with corrected mask information 37b obtained from comparison images 35. In the corrected mask information 37a and 37b, the location information of the features is obtained based on the location information of the features that is pre-registered in the feature area information 33 (i.e., the feature mask 36 is corrected). Therefore, in this embodiment, it is possible to detect changes in features that occur over time without being affected by errors that occurred in the generation of the feature mask 36.

[0081] Figure 5 is an example flowchart showing the process performed by the geological feature change detection device 10 according to this embodiment.

[0082] In step S101, the feature mask estimation unit 21 receives the past image 31 and at least one comparison image 35 as input and generates a feature mask 36 for each of the past image 31 and comparison image 35. In the following explanation, for the sake of clarity, we will consider the case where there is only one comparison image 35, and refer to the feature mask 36 corresponding to the past image 31 as feature mask 36a, and the feature mask 36 corresponding to the comparison image 35 as feature mask 36b.

[0083] In step S102, the feature mask correction unit 22 receives the feature mask 36 (36a, 36b) as input and extracts from the feature area information 33 the area in which the features included in the acquired feature mask 36 exist (i.e., the area corresponding to the imaging area of ​​the past image 31 and the comparison image 35). This extraction can be performed, for example, using the positional information (coordinate values) that the past image 31 and the comparison image 35 possess.

[0084] In step S103, the feature mask correction unit 22 selects one feature mask 36 that has not been corrected in the correction process described later. If multiple feature masks 36 are input, the feature mask correction unit 22 performs the correction process on all feature masks 36. In this embodiment, we will describe the case in which the feature mask correction unit 22 first selects feature mask 36a and then selects feature mask 36b.

[0085] Steps S104 to S108 are correction processes for the feature mask 36.

[0086] In step S104, the feature mask correction unit 22 selects one area from the feature area information 33 extracted in step S102 (reference area of ​​feature area information 33) where a determination of whether or not a feature exists (feature existence determination) has not been performed.

[0087] In step S105, the feature mask correction unit 22 calculates the degree of overlap between the feature region selected in step S104 and the feature mask 36. As previously explained, the degree of overlap is a value used to determine whether or not the selected feature exists (a value used in predetermined determination conditions). In this embodiment, the overlap ratio previously explained is used as the degree of overlap. The overlap ratio is the ratio of the overlapping area to the area of ​​the feature region selected in step S104, and the overlapping area is the area where this feature region and the area of ​​the feature mask 36 corresponding to this feature region overlap.

[0088] In step S106, the feature mask correction unit 22 determines, according to predetermined determination conditions, whether the overlap rate calculated in step S105 is equal to or greater than a preset threshold. If the overlap rate is equal to or greater than the threshold, the feature mask correction unit 22 determines that a feature exists corresponding to the feature area selected in step S104 and performs the process in step S107. On the other hand, if the overlap rate is less than the threshold, the feature mask correction unit 22 determines that a feature does not exist corresponding to the feature area selected in step S104 and performs the process in step S108.

[0089] In step S107, the feature mask correction unit 22 determines that a feature exists corresponding to the feature region selected in step S104, and therefore, in the corrected mask information 37, it designates the region corresponding to this feature region as the region where the feature exists. In other words, the feature mask correction unit 22 records the feature region indicated by the feature mask 36 as an existing feature in the corrected mask information 37.

[0090] At least one corrected mask information 37 is created for each feature mask 36. In this embodiment, the feature mask correction unit 22 creates two corrected mask information 37a, which corresponds to feature mask 36a, and corrected mask information 37b, which corresponds to feature mask 36b.

[0091] In step S108, the feature mask correction unit 22 determines whether or not a feature exists (feature existence determination) has been performed for all feature regions registered in the feature region information 33 extracted in step S102. If there are feature regions for which the feature existence determination has not been performed, the feature mask correction unit 22 performs the process in step S104 and selects one new feature region for which the feature existence determination has not been performed. If the feature existence determination has been performed for all feature regions, the feature mask correction unit 22 performs the process in step S109.

[0092] In step S109, the feature mask correction unit 22 determines whether the correction process described above (steps S104 to S108) has been performed on all feature masks 36 generated in step S101. If the correction process has been performed on all feature masks 36, the feature change detection unit 23 performs the process in step S110. If the correction process has not been performed on all feature masks 36, the feature mask correction unit 22 performs the process in step S103. In this embodiment, the feature mask correction unit 22 first selects feature mask 36a, and then selects feature mask 36b. If the correction process has not been performed on feature mask 36b, the feature mask correction unit 22 performs the process in step S103 and selects feature mask 36b.

[0093] In step S110, the feature change detection unit 23 compares the multiple corrected mask information 37 created in step S107 and records the difference. Specifically, the feature change detection unit 23 calculates the difference between the corrected mask information 37a obtained by correcting the feature mask 36a of the past image 31 and the corrected mask information 37b obtained by correcting the feature mask 36b of the comparison image 35, and outputs the calculated difference as feature change information 34. In this way, the feature change detection unit 23 detects changes in features between the time the past image 31 was captured and the time the comparison image 35 was captured.

[0094] For example, if the objective is to detect features that have disappeared between the time the past image 31 was captured and the time the comparison image 35 was captured, the feature change detection unit 23 calculates the result of subtracting the corrected mask information 37b from the corrected mask information 37a and records the obtained result as feature change information 34.

[0095] Furthermore, for example, if the feature is a building and the objective is to detect a building that has been newly constructed between the time the past image 31 was captured and the time the comparison image 35 was captured, the result of subtracting the corrected mask information 37a from the corrected mask information 37b is obtained, and the obtained result is recorded as feature change information 34.

[0096] In this embodiment, we have described an example where one feature mask 36 contains only masks representing one type of feature (for example, the feature mask 36c in Figure 4 contains only masks M1 to M4 representing buildings). A single feature mask 36 can contain masks representing multiple types of features (for example, buildings and roads).

[0097] Furthermore, a feature mask 36 can be generated for each type of feature. For example, if the features are buildings and roads, two types of feature masks 36 may be generated: one feature mask 36 containing a mask representing buildings, and another feature mask 36 containing a mask representing roads. In this case, in step S102, the region containing features of the type represented by the feature mask 36 is extracted from the feature region information 33.

[0098] Figure 6 shows an example of a display screen 60 output by the geological feature change detection device 10 according to this embodiment. The output unit 13 (Figure 2) displays the display screen 60 on a screen such as a liquid crystal display.

[0099] The display screen 60 shown in Figure 6 includes a display area 61 for past images 31, a display area 65 for comparison target images 35, and a display area 63 for an image (detection result image) showing the changes in features detected by the feature change detection unit 23. The display area 61 for past images 31 displays the past image 31 of the area corresponding to the comparison target image 35. The display area 65 for comparison target images 35 displays, for example, the comparison target image 35 captured by the camera 11.

[0100] The display area 63 of the detection result image displays, for example, an image obtained from the difference between the corrected mask information 37a for the past image 31 and the corrected mask information 37b for the comparison target image 35, as the detection result image. That is, the detection result image displays features that have changed between the corrected mask information 37a obtained using the feature mask 36a of the past image 31 and the corrected mask information 37b obtained using the feature mask 36b of the comparison target image 35.

[0101] In the display area 63 of the detection result image in Figure 6, as an example, a building that has collapsed and disappeared is displayed, that is, a building that exists in the past image 31 but does not exist in the comparison image 35. This building is a feature that has changed between the corrected mask information 37a for the past image 31 and the corrected mask information 37b for the comparison image 35.

[0102] Furthermore, the display screen 60 may include a setting area 67. The setting area 67 is an area for the operator of the feature change detection device 10 to set the image to be displayed on the display screen 60.

[0103] The operator can change the information displayed in the display area 63 of the detection result image by manipulating the setting area 67 using the input unit 12 (Figure 2). For example, the operator can select the features (buildings) to display in the display area 63 using checkboxes or radio buttons in the setting area 67. In the example shown in Figure 6, the operator can display either collapsed buildings or existing buildings. Also, for example, the operator can change the threshold value of the overlap used by the feature mask correction unit 22 when determining the presence of features using a slider in the setting area 67.

[0104] Furthermore, the display screen 60 may also display feature area information 33 and feature masks 36 generated by the feature mask estimation unit 21 overlaid on the past image 31 and the comparison image 35 by the operator operating the setting area 67.

[0105] Furthermore, the display screen 60 may, for example, display information about a feature in an additional window such as a pop-up when the operator moves the mouse cursor over the feature in the past image 31 using the mouse on the input unit 12. Examples of such information include the feature ID, which is a serial number for managing the feature, and the result of the feature existence determination (for example, whether the building has disappeared or still exists).

[0106] The feature change detection device 10 according to this embodiment can include location information of objects not registered as features in the feature area information 33. That is, the feature area information 33 may also contain location information of objects not registered as features. Hereinafter, objects that are not registered as features but whose location information is known will be referred to as non-feature objects. The feature area information 33 can contain location information of non-feature objects as map information, etc.

[0107] Non-terrain objects include objects that do not appear in past images 31 or comparison images 35, such as buried water pipes, fiber optic cables, and power lines. Above-ground power lines can also be included as non-terrain objects if they do not appear in past images 31 or comparison images 35.

[0108] The feature change detection unit 23 uses feature area information 33 to refer to the location information of non-feature objects and can estimate changes in non-feature objects from the results of detecting changes in features at the same location as the non-feature object or in the vicinity of the non-feature object. For example, the feature change detection unit 23 estimates that non-feature objects at the same location as or in the vicinity of the feature whose change was detected have changed in the same way as the feature between the time the past image 31 was captured and the time the comparison image 35 was captured.

[0109] For example, suppose the geographical feature is a road, and the non-geographical feature is a water pipe buried underground beneath the road. The geographical feature change detection unit 23 detects an event such as a fracture or subsidence of the road as a geographical feature change. In this case, if the water pipe is located at the same location as the road or in the vicinity of the road, the geographical feature change detection unit 23 estimates that the water pipe is also damaged in the same way as the road. Specifically, when the geographical feature change detection unit 23 detects a fracture in the road, if the water pipe is located at the same location as the fracture or within a certain radius from the fracture, it is highly likely that the water pipe is also damaged, and therefore the unit records the water pipe as damaged or as an object requiring verification.

[0110] Through the above processing, the feature change detection device 10 according to this embodiment can estimate changes in objects that are not registered as features or objects that do not appear in the image.

[0111] As described above, the feature change detection device 10 according to this embodiment can provide information about changes in features by reducing the difference between the location of features obtained from the image and the actual location of features, and by creating corrected mask information 37 by calculating the degree of overlap with features registered in the feature area information 33 for each of the feature mask 36a generated from past image 31 and the feature mask 36b generated from comparison image 35. [Examples]

[0112] This section describes a geological feature change detection device and geological feature change detection method according to Embodiment 2 of the present invention. The following will primarily describe the differences between the geological feature change detection device and geological feature change detection method according to this embodiment and Embodiment 1.

[0113] In Example 1, the feature change detection device 10 generates a feature mask 36 from images taken at different times (past image 31 and comparison image 35), and detects changes in features from the difference between corrected mask information 37 (37a, 37b) created using the feature mask 36.

[0114] In this embodiment, by estimating and generating a ground attribute mask in addition to the feature mask 36 for both the comparison image 35 and the past image 31, it is possible to provide information about changes in features even when the changes in features are ambiguous.

[0115] Ground attributes refer to objects that exist between a feature and the camera 11 and obstruct the imaging of the feature by the camera 11. In other words, in the image captured by the camera 11, the feature overlaps with the ground attributes and is hidden by them, so part or all of it is not captured.

[0116] A feature / ground attribute mask is a general term for both feature masks and ground attribute masks. A ground attribute mask is information representing the location of ground attributes obtained from an image, that is, the location and area of ​​ground attributes obtained from an image. Therefore, a feature / ground attribute mask is information representing the location of features and ground attributes obtained from an image, that is, the location and area of ​​features and ground attributes obtained from an image. Since a feature / ground attribute mask includes both feature masks and ground attribute masks, it contains masks for one or more features and ground attributes.

[0117] Figure 7 is a functional configuration diagram of the feature change detection device 10 according to this embodiment. The feature change detection device 10 according to this embodiment differs from Embodiment 1 in that the configuration of the feature mask estimation unit 21 and the feature mask correction unit 22 is different, and it is equipped with a feature / ground attribute mask estimation model 32a instead of the feature mask estimation model 32.

[0118] The feature / ground attribute mask estimation model 32a is a model for generating feature / ground attribute masks 38 (i.e., both feature masks and ground attribute masks) from the comparison image 35 and the past image 31. The feature / ground attribute mask estimation model 32a can be pre-created, for example, by machine learning and stored in the feature change detection device 10.

[0119] The feature mask estimation unit 21 estimates and generates feature and ground attribute masks 38 for both the comparison image 35 and the past image 31 using the feature and ground attribute mask estimation model 32a. The feature mask estimation unit 21 outputs the generated feature and ground attribute masks 38 to the feature mask correction unit 22.

[0120] The feature mask correction unit 22 calculates the degree of overlap with features registered in the feature area information 33 for each of the feature / ground attribute masks 38a generated from the past image 31 and the feature / ground attribute mask 38b generated from the comparison image 35, and creates and outputs corrected mask information 37. In this embodiment, the corrected mask information 37 is information obtained by correcting the feature / ground attribute masks 38 generated from the past image 31 and the comparison image 35 using the feature area information 33. The corrected mask information 37 includes information on features represented by the feature / ground attribute mask 38 whose existence has been confirmed by the feature area information 33, with the area corresponding to these features registered as a corrected feature mask, as well as information on features that require verification (features requiring verification) based on the feature / ground attribute mask 38 and the feature area information 33.

[0121] The feature mask correction unit 22 determines that a feature exists if the feature registered in the feature area information 33 satisfies predetermined determination conditions (predetermined determination conditions using the degree of overlap, as explained in Example 1) with respect to the feature mask. Furthermore, the feature mask correction unit 22 determines that a feature is a feature with added ground attributes if the feature registered in the feature area information 33 satisfies predetermined determination conditions with respect to the ground attribute mask. A feature with added ground attributes is a feature that, in the image captured by the camera 11, overlaps with the ground attributes and is obscured by the shadow of the ground attributes, so that part or all of it is not captured. In this embodiment, under the predetermined determination conditions in Example 1, the feature mask correction unit 22 determines the degree of overlap between the feature / ground attribute mask 38 (feature mask and ground attribute mask) and the feature registered in the feature area information 33.

[0122] The feature mask correction unit 22 performs the above processing to create corrected mask information 37a using the feature / ground attribute mask 38a generated from the past image 31, and to create corrected mask information 37b using the feature / ground attribute mask 38b generated from the comparison image 35. In addition, if there are features with added ground attributes, the feature mask correction unit 22 outputs information about the features that are features with added ground attributes to the feature change detection unit 23.

[0123] The feature change detection unit 23 extracts and calculates the difference between the corrected mask information 37a and the corrected mask information 37b, and outputs the calculated difference as feature change information 34. In addition, if a feature with ground attributes is present, the feature with ground attributes is designated as a feature requiring verification, and information about this feature is output. A feature with ground attributes is a feature that overlaps with ground attributes in the image, and it may be unclear whether it actually exists or not due to the ground attributes. For this reason, a feature with ground attributes is designated as a feature requiring verification to determine whether it actually exists or not.

[0124] The output unit 13 (Figure 2) can display, for example, a surface with added ground attributes as a surface requiring verification on a screen such as a liquid crystal display.

[0125] Figure 8 is a diagram illustrating the processing performed by the feature change detection device 10 in Example 2. Figure 8 shows the past image 31, the comparison image 35, the feature / ground attribute mask 38a generated from the past image 31, the feature / ground attribute mask 38b generated from the comparison image 35, the corrected mask information 37a obtained by correcting the feature / ground attribute mask 38a, and the corrected mask information 37b obtained by correcting the feature / ground attribute mask 38b.

[0126] Figure 8 shows an example where the feature is a building, the ground attribute is vegetation, and the feature change detection device 10 detects changes in the imaged building. Note that the ground attribute is not limited to vegetation, but may also be clouds, power lines, or flying objects. The feature / ground attribute mask 38 includes a building mask, which is the feature mask, and a vegetation mask, which is the ground attribute mask. The corrected mask information 37 includes a corrected building mask, which is the corrected building mask, and a ground attribute-added feature 39.

[0127] Past image 31 includes buildings T1, T2, T3, T4, and T5 as features, and also includes vegetation P as a ground attribute. In past image 31, buildings T3 and T4 are overlapped with vegetation P and hidden by it, so parts of them are not visible.

[0128] Comparison image 35 was taken after past image 31 was taken. Assume that a landslide occurred between the time past image 31 was taken and the time comparison image 35 was taken. Comparison image 35 shows buildings T1, T2, T4, and T5 after they have been swept away and disappeared due to the mudslide. In comparison image 35, building T3 is still present.

[0129] The feature mask estimation unit 21 receives a past image 31 and a comparison image 35 as input, generates a feature / ground attribute mask 38a for the past image 31, and generates a feature / ground attribute mask 38b for the comparison image 35. The feature / ground attribute mask 38a includes feature masks (building masks) corresponding to buildings T1 to T5 and ground attribute masks (vegetation masks) corresponding to vegetation P. The feature / ground attribute mask 38b includes only the vegetation mask.

[0130] In the example shown in Figure 8, the feature mask estimation unit 21 uses semantic segmentation, a type of machine learning algorithm that determines whether or not each pixel in the image is included in the recognition target, as the feature / ground attribute mask estimation model 32a. For example, the feature mask estimation unit 21 estimates a feature / ground attribute mask 38 in the image, where areas containing buildings are represented by 1, areas containing vegetation are represented by 2, and other areas are represented by 0. Figure 8 shows this feature / ground attribute mask 38 displayed in an image as feature / ground attribute mask 38a and feature / ground attribute mask 38b.

[0131] The feature mask estimation unit 21 may generate multiple types of feature / ground attribute masks 38. For example, the feature mask estimation unit 21 may generate two types of feature / ground attribute masks 38: one in which areas containing buildings are represented by 1 and other areas by 0, and another in which areas containing vegetation are represented by 1 and other areas by 0. In this way, the feature mask estimation unit 21 can generate feature / ground attribute masks 38 represented in any way.

[0132] The feature mask correction unit 22 calculates the degree of overlap between the feature / ground attribute mask 38a and the features registered in the feature area information 33. If the feature and the feature mask satisfy predetermined criteria, it determines that the feature exists. If the feature and the ground attribute mask satisfy predetermined criteria, it determines that the feature is a feature with added ground attributes 39. The feature mask correction unit 22 then calculates the degree of overlap between the feature / ground attribute mask 38b and the features registered in the feature area information 33. If the feature and the feature mask satisfy predetermined criteria, it determines that the feature exists. If the feature and the ground attribute mask satisfy predetermined criteria, it determines that the feature is a feature with added ground attributes 39. Note that the criteria between a feature and a feature mask and the criteria between a feature and a ground attribute mask may differ from each other, for example, in terms of the definition of overlap and the threshold for overlap.

[0133] The feature mask correction unit 22 performs the above processing on each of the features included in the feature / ground attribute mask 38a and the feature / ground attribute mask 38b, thereby creating and outputting corrected mask information 37a obtained by correcting the feature / ground attribute mask 38a and corrected mask information 37b obtained by correcting the feature / ground attribute mask 38b.

[0134] In the example shown in Figure 8, the presence of building T3, which is a ground attribute-added feature 39, in the corrected mask information 37a and 37b is because building T3 is registered in the feature area information 33 and is a feature that overlaps with the ground attribute mask (vegetation mask) of the feature / ground attribute masks 38a and 38b. Building T3, which is a ground attribute-added feature 39, is a feature that requires verification.

[0135] The feature change detection unit 23 extracts and calculates the difference between the corrected mask information 37a and the corrected mask information 37b, and outputs the calculated difference as feature change information 34.

[0136] In the example shown in Figure 8, both the past image 31 and the comparison image 35 include vegetation P in the foreground. Therefore, in both the past image 31 and the comparison image 35, part of the building T3 is not captured due to the ground-level vegetation P, making it difficult to accurately grasp the changes in the building T3.

[0137] In the feature change detection device 10 according to this embodiment, features that cannot be detected because their entirety is not imaged due to the influence of ground attributes are designated as features with added ground attributes 39, thereby indicating to the operator of the feature change detection device 10 that these features cannot be detected due to ground attributes. For example, in the example shown in Figure 8, the feature change detection device 10 may record building T3, which is a feature with added ground attributes 39, as a feature requiring verification and display it on the display screen 60. Building T3 is a feature requiring verification, and although it was not detected by the feature / ground attribute mask 38b due to ground attributes (vegetation P), it may actually be a feature that exists. Therefore, upon seeing that building T3 is displayed as a feature requiring verification on the display screen 60, the operator can, for example, treat building T3 in the same way as a lost building and use this information to formulate operations such as dispatching a rescue team.

[0138] Although the comparison image is not shown, the following example is also conceivable. For instance, suppose that vegetation P and building T3 existed when past image 31 (Figure 8) was captured, but did not exist when the comparison image was captured. Furthermore, suppose that in the comparison image, vegetation P and building T3 are not captured, but the ground at the location of building T3 is captured.

[0139] In this case, the feature corresponding to building T3 is registered as a ground attribute-added feature 39 in the corrected mask information 37a, and as a feature that does not exist in the corrected mask information 37b. At the time the past image 31 was captured, building T3 was registered as a ground attribute-added feature 39 and is a feature that requires verification, so it is unclear whether it exists or not. However, since it is confirmed that building T3 does not exist at the time the comparison image was captured, it is registered as a lost feature. Note that the conditions under which the feature change detection unit 23 extracts the difference are not limited to the above embodiment, but include various examples.

[0140] It should be noted that the present invention is not limited to the embodiments described above, and various modifications are possible. For example, the embodiments described above are explained in detail to make the present invention easier to understand, and the present invention is not necessarily limited to embodiments having all the configurations described. Furthermore, it is possible to replace parts of the configuration of one embodiment with the configuration of another embodiment. It is also possible to add configurations from other embodiments to the configuration of one embodiment. Furthermore, it is possible to delete parts of the configuration of each embodiment, or to add or replace other configurations. [Explanation of symbols]

[0141] 10...Feature change detection device, 11...Camera, 12...Input unit, 13...Output unit, 14...CPU, 15...Main memory, 16...Auxiliary memory, 21...Feature mask estimation unit, 22...Feature mask correction unit, 23...Feature change detection unit, 31...Past image, 32...Feature mask estimation model, 32a...Feature / ground attribute mask estimation model, 33, 33c...Feature area information, 34...Feature change information, 35...Comparison target image, 36, 36a, 36b, 36c...Ground Object mask, 37, 37a, 37b, 37c... Corrected mask information, 38, 38a, 38b... Feature / ground attribute mask, 39... Feature with added ground attribute, 41... Overlay map, 60... Display screen, 61... Display area of ​​past images, 63... Display area of ​​detection result image, 65... Display area of ​​comparison target image, 67... Setting area, M, M1~M4... Mask, M11, M22... Corrected building mask, P... Vegetation, R, R1~R4... Feature area, T1~T5... Building.

Claims

1. A camera that captures comparison images, which are images of geographical features taken from above or at a high altitude, A feature mask estimation unit generates a feature mask representing the location information of the feature from a past image, which is an image of the feature taken from above or at a high altitude at a time prior to the acquisition of the comparison image, and from the comparison image. The location information of the aforementioned feature is registered in advance as feature area information, A feature mask correction unit calculates the degree of overlap with the feature registered in the feature area information for each of the feature masks generated from the aforementioned past image and the feature mask generated from the aforementioned comparison image, and registers the area corresponding to the feature whose degree of overlap with the feature is greater than or equal to a threshold as the corrected feature mask in the corrected mask information. A feature change detection unit detects the temporal change of a feature by calculating the difference between the corrected mask information obtained using the feature mask of the past image and the corrected mask information obtained using the feature mask of the comparison target image, A geological feature change detection device characterized by comprising the following:

2. Equipped with a feature mask estimation model created using machine learning, The feature mask estimation unit generates the feature mask using the feature mask estimation model. The geological feature change detection device according to claim 1.

3. Objects existing between the camera and the feature are defined as ground attributes. The feature mask estimation unit generates a feature / ground attribute mask representing the positional information of the feature and the ground attribute from the past image and the comparison image, respectively. The geological feature change detection device according to claim 1.

4. The feature / ground attribute mask comprises the feature mask and a ground attribute mask representing the location information of the ground attribute. The aforementioned feature mask correction unit is For each of the feature / ground attribute masks generated from the aforementioned past images and the feature / ground attribute masks generated from the aforementioned comparison images, the degree of overlap with the features registered in the feature area information is calculated. The region corresponding to the feature whose degree of overlap with the aforementioned ground attribute mask is greater than or equal to a threshold is registered in the corrected mask information as a feature with added ground attributes. The geological feature change detection device according to claim 3.

5. It has an output unit with a screen, The output unit displays on the screen the features that have changed between the corrected mask information obtained using the feature mask of the past image and the corrected mask information obtained using the feature mask of the comparison image. The geological feature change detection device according to claim 1.

6. It has an output unit with a screen, The output unit displays the ground attribute-added feature on the screen. The geological feature change detection device according to claim 4.

7. The aforementioned feature area information includes location information of objects that are not registered as features. The feature change detection unit estimates that an object located at the same or nearby position as the feature whose time change was detected has changed between the time the past image was captured and the time the comparison image was captured. The geological feature change detection device according to claim 1.

8. Images of geographical features taken from above or from a high vantage point are used as comparison images. The location information of the aforementioned feature is executed by a feature change detection device that has pre-registered feature area information, The feature change detection device performs a feature mask estimation step, which generates a feature mask representing the location information of the feature from a past image, which is an image of the feature taken from above or at a high altitude at a time prior to the time the comparison target image was taken, and from the comparison target image. The feature change detection device performs a feature mask correction step in which it calculates the degree of overlap with the feature registered in the feature area information for each of the feature masks generated from the past image and the feature mask generated from the comparison target image, and registers the area corresponding to the feature whose degree of overlap with the feature is greater than or equal to a threshold as the corrected feature mask in the corrected mask information. The feature change detection device performs a feature change detection step in which it detects the temporal change of the feature by calculating the difference between the corrected mask information obtained using the feature mask of the past image and the corrected mask information obtained using the feature mask of the comparison target image, A method for detecting changes in geological features, characterized by comprising the following:

9. Objects existing between the camera that captured the comparison image and the feature are defined as ground attributes. In the feature mask estimation step, a feature / ground attribute mask representing the positional information of the feature and ground attributes is generated from the past image and the comparison image, The feature / ground attribute mask comprises the feature mask and a ground attribute mask representing the location information of the ground attribute. In the feature mask correction step, For each of the feature / ground attribute masks generated from the aforementioned past images and the feature / ground attribute masks generated from the aforementioned comparison images, the degree of overlap with the features registered in the feature area information is calculated. The region corresponding to the feature whose degree of overlap with the aforementioned ground attribute mask is greater than or equal to a threshold is registered in the corrected mask information as a feature with added ground attributes. The method for detecting changes in geological features according to claim 8.

Images