JP7544669B2 - Rockfall monitoring system and method - Google Patents

Description

The present invention relates to a rockfall monitoring system and a rockfall monitoring method.

There are techniques for detecting rock falls and collapses that occur on cliffs and slopes (see, for example, Patent Documents 1 to 3).
Patent Document 1 describes a technology in which sensor optical fibers are laid in each section to be monitored for rockfalls, and the sensor optical fibers for each section are connected in series or parallel to form an optical fiber system, and rockfalls are detected from changes in the light transmitted through this optical fiber system (see, for example, paragraph 0013).
Patent Document 2 describes a technology in which falling rocks are guided and collected in an accumulation area using a rockfall protection net stretched over the slopes of a monitored area, and the change in volume of the accumulation area or the change in weight of the collected fallen rocks is converted into a change in strain in an optical fiber, and the presence or absence of falling rocks is detected from the change in the amount of strain in the optical fiber (see, for example, paragraph 0006).
Patent Document 3 describes a rockfall detection device that is attached to rockfall protection facilities, such as rockfall protection nets, installed on slopes and immediately below the slopes to detect falling rocks, landslides, etc. (See, for example, claim 1.) This rockfall detection device is equipped with semiconductor sensor elements that detect geomagnetism, pressure, inclination, acceleration, etc., and judges the occurrence of a rockfall or landslide based on the results of comparing waveform information, such as the amplitude and period of vibration caused by a falling rock or landslide, with pre-stored information.

JP 2000-180219 A JP 2002-138418 A JP 2011-047252 A

The techniques disclosed in Patent Documents 1 to 3 make it possible to give warnings and cautions about falling rocks that have already occurred, but they do not allow for warnings and cautions that take into account rockfalls that will occur in the near future.
Furthermore, the techniques of Patent Documents 1 to 3 have the problem that optical fiber systems, rockfall protection nets, etc. must be installed in advance in the range where rockfall is to be detected. For example, if the range where rockfall is expected is wide, optical fiber systems, rockfall protection nets, etc. must be installed in the entire range, which requires a great deal of effort for advance preparation.
From this perspective, the present invention provides a rockfall monitoring system and a rockfall monitoring method that can contribute to improving management and countermeasures against rockfalls.

The rockfall monitoring system according to the present invention is a system for monitoring the occurrence of a rockfall, and includes an imaging unit for imaging a monitoring area, a rockfall analysis unit, and a rockfall prediction unit.
The rockfall analysis unit obtains quantitative first information regarding a rockfall that has occurred in the monitoring area from the images captured by the imaging unit through image analysis using a frame difference method.
The rockfall prediction unit predicts a new rockfall that will occur in the monitoring area based on the first information and second information related to the environment at the time the captured image is captured. The rockfall prediction unit has a trained model that has been machine-learned to input a plurality of units of the first information and the second information that are consecutive in time series, with a predetermined time being one unit, and output third information related to the new rockfall in the next unit of time .
The first information is, for example, any one of a source of a rockfall, a size of the rockfall, and a characteristic of the rockfall movement. The vehicle may further include a notification unit that notifies the occurrence of a new rockfall.
This makes it possible to provide warnings and alerts that take into account rockfalls that may occur in the near future, thereby contributing to improved management and countermeasures against rockfalls.
The image capturing device may further include a rockfall detection unit that detects a rockfall in an image captured by the imaging unit by a frame difference method. In this case, the rockfall analysis unit may obtain the first information from a frame difference image in which a rockfall is detected by the rockfall detection unit.
In this way, rockfall detection and quantitative first information regarding rockfall can be obtained by analyzing captured images, thereby reducing the effort required for advance preparations and making it easier to monitor rockfalls than before.

The rockfall monitoring method according to the present invention is a method for monitoring the occurrence of a rockfall, and includes an imaging step of imaging a monitoring area, a rockfall analysis step, and a rockfall prediction step.
In the rockfall analysis step, first quantitative information regarding a rockfall occurring in the monitoring area is obtained from the captured images captured in the imaging step by image analysis using a frame difference method.
In the rockfall prediction step, a new rockfall occurring in the monitoring area is predicted based on the first information and second information related to the environment at the time when the captured image was captured. In this rockfall prediction step, a prediction is made using a trained model that has undergone machine learning so as to input a plurality of units of the first information and the second information that are consecutive in time series , with a predetermined time being one unit, and output third information related to the new rockfall in the next unit of time .
This makes it possible to provide warnings and alerts that take into account rockfalls that may occur in the near future, thereby contributing to improved management and countermeasures against rockfalls.

This invention can contribute to improving management and countermeasures against falling rocks.

1 is a conceptual diagram of a rockfall monitoring system according to an embodiment of the present invention; 1 is a schematic configuration diagram of a rockfall monitoring system according to an embodiment of the present invention. FIG. 2 is a block diagram of an image processing device that is a part of a monitoring unit. FIG. 2 is a flow chart showing an operation image of the rockfall monitoring system according to the embodiment of the present invention. FIG. 1 is an image diagram of image analysis using the frame difference method. 11 is a diagram for explaining a process of extracting quantitative information (first information) regarding falling rocks. FIG. 11 is a diagram for explaining a process for estimating the source of a rockfall. FIG. FIG. 2 is a diagram illustrating an example of observation data on which learning data is based. FIG. 13 is a diagram illustrating an example of a learning model. FIG. 2 is a diagram illustrating an example of observation data on which learning data is based. FIG. 13 is a diagram illustrating an example of a learning model.

Hereinafter, the embodiment of the present invention will be described in detail with reference to the drawings. Each drawing is merely a schematic illustration to allow a sufficient understanding of the present invention. Therefore, the present invention is not limited to the illustrated examples. In each drawing, common or similar components are denoted by the same reference numerals, and duplicated explanations thereof will be omitted.
<Configuration of rockfall monitoring system according to embodiment>
The configuration of a rockfall monitoring system 1 according to an embodiment will be described with reference to Fig. 1 and Fig. 2. Fig. 1 is a conceptual diagram of the rockfall monitoring system 1. Fig. 2 is a schematic diagram of the rockfall monitoring system 1.

The rockfall monitoring system 1 has a function of detecting rockfalls from the captured video (consisting of successive images) and of notifying the detection of a rockfall. The rockfall monitoring system 1 also has a function of predicting new rockfalls that will occur in the near future based on the detected rockfalls.
The rockfall monitoring system 1 can be used in various situations requiring monitoring of rockfalls, and the rockfalls to be monitored are not particularly limited in terms of shape, size, color, etc. Here, as shown in Fig. 1, it is assumed that the rockfall monitoring system 1 is used to monitor cliffs and slopes, and at least a part (or the entirety) of the cliffs and slopes is the monitoring area.
The rockfall monitoring system 1 uses artificial intelligence (AI) technology to learn the tendency of rockfalls from images of rockfalls, and predicts rockfalls that will occur in the near future based on rockfalls that have already occurred. In the learning and prediction, quantitative information about rockfalls is extracted from images of rockfalls by image analysis using a frame difference method (also called "interframe difference method"), and the quantitative information is used. In the frame difference method, a frame difference image between two consecutive frames is generated, and a moving object is identified by threshold processing. Details will be described later.

As shown in Fig. 1, the rockfall monitoring system 1 includes an imaging unit 2, a monitoring unit 3, and an alarm unit 4. As shown in Fig. 2, the imaging unit 2, the monitoring unit 3, and the alarm unit 4 receive power from a power source 5. The type and number of the power sources 5 are not particularly limited. For example, the imaging unit 2, the monitoring unit 3, and the alarm unit 4 can each be configured as an independent device.
As shown in Fig. 1, the imaging unit 2 is installed so as to be able to capture an image of a location where a rockfall 8 may occur (here, a cliff face 9). In Fig. 1, the imaging unit 2 is installed on a bridge located away from the cliff face 9. The notification unit 4 is installed in a location where it can be seen by a person to whom an alarm is to be issued (e.g., a person passing near the cliff face 9 or a person performing work near the cliff face 9, who will receive a report of the occurrence or prediction of a rockfall). In Fig. 1, the notification unit 4 is installed on a road leading to the cliff face 9. The installation location of the monitoring unit 3 is not particularly limited, and it may be installed, for example, near the notification unit 4 or in a facility where a manager who manages the rockfall monitoring system 1 is located.
The imaging unit 2 and the monitoring unit 3 are communicatively connected, for example, by wireless communication, and an image of the cliff face 9 is transmitted from the imaging unit 2 to the monitoring unit 3. The monitoring unit 3 and the notification unit 4 are communicatively connected, for example, by a communication cable, and the monitoring unit 3 notifies the notification unit 4 of the detection results and prediction results of the rockfall 8 (whether or not a rockfall 8 has been detected, whether or not a rockfall will occur after a predetermined time, etc.). Note that the configuration of the rockfall monitoring system 1 is not limited to that shown here, and for example, the monitoring unit 3 may have the functions of the imaging unit 2 and the notification unit 4 (that is, the monitoring unit 3 may include the imaging unit 2 and the notification unit 4).

The imaging unit 2 is a means for imaging the monitoring area (here, the cliff face 9). The imaging unit 2 has a resolution that allows the recognition of the falling rocks 8, and is capable of continuous shooting to the extent that the movement of the falling rocks 8 can be seen. The imaging unit 2 is, for example, a digital video camera, and captures several tens of images (for example, 40 images) per second (hereinafter, sometimes referred to as "frames"). In this embodiment, since it is assumed that the imaging unit 2 will be used to monitor the cliff face 9, the imaging unit 2 is installed to capture the cliff face 9. The images captured by the imaging unit 2 are sent to the monitoring unit 3. It is preferable that the imaging unit 2 has specifications that allow it to withstand long periods of outdoor use.
The imaging unit 2 may have a lighting means (not shown). The lighting means is a light source that emits light, such as an LED (Light Emitting Diode) light or an organic electroluminescent light (OLED). The lighting means is installed so as to illuminate the area in which the occurrence of rockfall 8 is monitored (i.e., the monitoring area). The number and shape of the lighting means are not particularly limited. The lighting means is a light source that includes at least a wavelength component corresponding to the sensitivity (camera sensitivity) of the imaging unit 2, and may be, for example, visible light, infrared light, etc. The imaging unit 2 may have a zoom function.

The alarm unit 4 is a means for notifying a danger when a rockfall 8 is detected. The alarm unit 4 may also notify a danger when it predicts the occurrence of a rockfall 8 in the near future (e.g., several minutes to several hours later). The alarm unit 4 notifies the person to be warned, for example, that a rockfall 8 has been detected or predicted, visually or aurally. The alarm unit 4 may be, for example, a display board, a rotating light, or a buzzer. In addition, in the case where work is assumed on a cliff face 9, a display function or a buzzer that sounds a warning sound may be provided on something worn by the person to be warned (e.g., a helmet).
The alarm unit 4 is installed at a location related to the object of monitoring (in FIG. 1, it is installed on a road passing in front of a cliff face 9). When notifying a prediction of a rockfall 8, the alarm unit 4 may issue an alarm according to the probability and scale of the rockfall 8 occurring, and the time until the rockfall 8 occurs. When the alarm unit 4 is installed outdoors, it is preferable that the specifications are such that it can withstand long-term outdoor use.

The monitoring unit 3 will be described with reference to Fig. 2 (and Fig. 1 as appropriate). The monitoring unit 3 performs image processing on the images captured by the imaging unit 2 (see Fig. 1) to detect rockfalls 8. The monitoring unit 3 also extracts quantitative information (first information) related to rockfalls from the images captured by the imaging unit 2 by image analysis using a frame difference method, and inputs the quantitative information into a trained model to predict the occurrence of future rockfalls 8.
2, the monitoring unit 3 is composed of, for example, an image storage device 3A and an image processing device 3B. The image storage device 3A and the image processing device 3B may be a PC (Personal Computer) or a server. The monitoring unit 3 may be a single device that combines the image storage device 3A and the image processing device 3B.
The image storage device 3A stores images of the cliff face 9. The images stored in the image storage device 3A are used for learning the learning model. The image storage device 3A may store images of a plurality of cliff faces 9 taken at different locations.
The image processing device 3B is a device that predicts whether a new rockfall will occur in the near future based on quantitative information about rockfalls extracted by image analysis using a frame difference method. The image processing device 3B has a trained model, and predicts the occurrence of a rockfall by inputting quantitative information about rockfalls into the trained model. The image processing device 3B may have a function of training the training model itself, or may acquire and use a trained model trained by another device (e.g., the image storage device 3A).

The configuration of the image processing device 3B will be described with reference to Fig. 3. Fig. 3 is a block diagram of the image processing device 3B. The image processing device 3B includes, for example, a storage unit 31 and a control unit 32.
The memory unit 31 stores information required for detecting and predicting rockfall 8 (e.g., shooting condition information 31a, detection target information 31b, rockfall movement information 31c, etc.). The shooting condition information 31a is, for example, information on the imaging unit 2 (angle of view, resolution, frame rate, etc.) and information on the distance L from the cliff face 9 to the imaging unit 2. The detection target information 31b is, for example, information on the size (minimum and maximum) and falling speed of the rockfall 8 to be detected. The rockfall movement information 31c is information on the characteristics of the rockfall movement, and is, for example, information on the movement of the rockfall 8 for each fall pattern (fall type) such as jumping movement, sliding movement, and free fall movement.
The control unit 32 is realized by a program execution process by a CPU (Central Processing Unit), a dedicated circuit, etc. When the control unit 32 is realized by a program execution process, the memory unit 31 stores a program (a rockfall monitoring program) for realizing the functions of the control unit 32.

The control unit 32 mainly includes an image acquisition unit 33, a rockfall detection unit 34, a rockfall analysis unit 35, a rockfall prediction unit 36, and a rockfall notification unit 37. Note that the functions of the control unit 32 shown in Fig. 3 are separated for the sake of convenience of explanation and do not limit the present invention. An overview of the process will be described here, and the details of the process will be described in the operation described later.
The image acquisition unit 33 acquires the image captured by the imaging unit 2. The image acquisition unit 33 here receives the image from the imaging unit 2 via wireless communication, but the communication means for acquiring the image is not particularly limited.
The rockfall detection unit 34 performs image processing on the image captured by the imaging unit 2, and detects the falling rocks 8 from the processed image. The method of image processing by the rockfall detection unit 34 is not particularly limited, and for example, the falling rocks 8 are detected using a frame difference method. In the frame difference method, a frame difference image between two consecutive frames is generated, and a moving object is identified by performing threshold processing. A method of detecting falling rocks using the frame difference method is described in, for example, "JP 2019-203288 A".
The rockfall analysis unit 35 extracts quantitative information (first information) regarding the rockfall 8 by image analysis using a frame difference method. The quantitative information extracted by the rockfall analysis unit 35 is sent to the rockfall prediction unit 36 and used for learning the learning model and for predicting rockfalls 8 that will occur in the near future.
The rockfall prediction unit 36 has a trained model, and inputs the quantitative information extracted by the rockfall analysis unit 35 into the trained model to predict the occurrence of a future rockfall 8. The rockfall prediction unit 36 may use the quantitative information extracted by the rockfall analysis unit 35 to train the trained model or re-train the trained model.
When the rockfall notification unit 37 detects a rockfall 8 or predicts the occurrence of a rockfall 8 in the near future (for example, several minutes to several hours later), it notifies the alarm unit 4 of that information. When the alarm unit 4 receives information about the occurrence of a rockfall 8 or a prediction thereof, it notifies the alarm unit 4 of that fact.

<Operation of rockfall monitoring system according to embodiment>
The operation of the rockfall monitoring system 1 according to the embodiment will be described with reference to Fig. 4 (and Figs. 1 to 3 as appropriate). Fig. 4 is a flow diagram showing an image of the operation of the rockfall monitoring system 1 according to the embodiment. As shown in Fig. 4, the operation of the rockfall monitoring system 1 mainly includes a "rockfall image recording process (S10)", an "environmental data recording process (S20)", a "rockfall analysis process S30", a "rockfall learning process (S40)", and a "rockfall prediction process (S50)".
(Rockfall image recording process (S10))
The imaging unit 2 starts photographing the cliff face 9, and if a rockfall 8 occurs during photographing, the imaging unit 2 stores the successive images at the time of the rockfall 8 and the time of photographing in the image storage device 3A. The image captured by the imaging unit 2 is sent to the image processing device 3B, and the rockfall detection unit 34 determines whether or not the image contains a falling rock 8, and the image determined to contain a falling rock 8 may be sent to the image storage device 3A and stored therein.

(Environmental data recording process (S20))
An institution not shown (for example, the Japan Meteorological Agency) stores environmental data (second information) related to the environment of the cliff face 9 (or the vicinity) being monitored. The environmental data may be, for example, meteorological data related to weather and earthquake data related to earthquakes. For example, the Japan Meteorological Agency constantly conducts meteorological and earthquake observations, and records the meteorological and earthquake data in association with the time when the measurements were made. The meteorological data may be, for example, the amount of precipitation, temperature, average wind speed, and maximum wind speed for each specified period of time. The earthquake data may be, for example, seismic intensity and type of earthquake.
(Rockfall analysis process S30)
The rockfall analysis unit 35 of the image processing device 3B acquires the images stored in the image storage device 3A and extracts quantitative information (first information) regarding the rockfall 8 by image analysis using the frame difference method. Image analysis using the frame difference method will be described with reference to Fig. 5. Fig. 5 is a conceptual diagram of image analysis using the frame difference method.
In the frame difference method, frame difference images F of two consecutive captured images E in time series are generated for the captured images E of the past N frames (N≧3, for example, N=10 is preferable ₎ , and a region of a moving object (here, a falling rock 8) is extracted from each frame difference image F by performing threshold processing (see the upper and middle parts of FIG. 5). FIG. 5 shows a case where a frame difference image _F1 is generated from captured images _E1 and _E2 , a frame difference image _F2 is generated from captured images E2 and _E3 , and a frame difference image _F3 is generated from captured images _E3 and _E4 .
Further, the frame difference images F are synthesized to generate a frame difference synthetic image G, and the union of the moving object regions is obtained (see the lower part of FIG. 5). FIG. 5 shows a case where the frame difference synthetic image G is generated from the frame difference images F ₁ , F ₂ , F ₃ , . . .

The rockfall analysis unit 35 of the image processing device 3B obtains quantitative information (first information) regarding the rockfall 8, such as (a) the source of the rockfall, (b) the scale of the rockfall, and (c) the characteristics of the rockfall movement.
(A) Estimation of the source of a rockfall Estimation of the source of a rockfall will be described with reference to Fig. 6 and Fig. 7. Fig. 6 is a diagram for explaining the extraction process of quantitative information regarding a rockfall 8. Fig. 7 is a diagram for explaining the estimation process of the source of a rockfall 8.
As shown in FIG. 6, when the frame difference images F are arranged in a time series, the frame difference images F are composed of only black pixels in the period _T1 before the occurrence of the rockfall and the period _T3 when the rockfall has ended. On the other hand, in the period _T2 during which the rockfall is occurring, the area of the falling rock 8 (symbol K) is marked with white pixels, and the frame difference images F ₁ to F _n are composed of black and white pixels. The rockfall analysis unit 35 identifies the source of the falling rock 8 by using the frame difference image F in which white pixels appear in the time series (frame difference image F ₁ in FIG. 6). The rockfall analysis unit 35 may determine the position of the white pixel as the source of the falling rock 8, or may divide the frame difference image F into a predetermined number of parts to set an area H for distinguishing the source on the frame difference image F as shown in FIG. 7, and identify which area H the area of the falling rock 8 (symbol K) corresponds to. In FIG. 7, the identification information of the area H is marked in the area H with a two-digit number.

(A) Estimation of the Scale of a Rockfall The rockfall analysis unit 35 estimates the approximate scale of the rockfall 8 by determining the area of the rockfall 8 (i.e., the area of the white pixels indicated by the symbol K) on the frame difference image F. The scale of the rockfall 8 may be estimated using, for example, the following formula (1).
The value obtained by the formula (1) may be further graded to represent the scale of the rockfall 8. For example, the scale S of the rockfall 8 may be assigned to "1 square meter or more" as a scale rating of "5", to "0.8 square meters or more and less than 1 square meter" as a scale rating of "4", to "0.6 square meters or more and less than 0.8 square meters" as a scale rating of "3", to "0.4 square meters or more and less than 0.6 square meters" as a scale rating of "2", to "0.2 square meters or more and less than 0.4 square meters" as a scale rating of "1", and to "0 square meters or more and less than 0.2 square meters" as a scale rating of "0".

(C) Estimation of Characteristics of Falling Rocks The rockfall analysis unit 35 estimates the characteristics of the falling rocks' movement from the trajectory of the falling rocks 8 shown in the frame difference composite image G shown in Fig. 5. The characteristics of the falling rocks' movement are, for example, falling patterns (fall types) such as jumping movement, sliding movement, and free fall movement. Jumping movement is when the falling rocks 8 hit the cliff face 9 and bounce while falling. Sliding movement is when the falling rocks 8 slide along the cliff face 9 while falling. Free fall movement is when the falling rocks 8 falls without making much contact with the cliff face 9. The rockfall analysis unit 35 holds, for example, the characteristics of these movements as information in advance, and estimates the characteristics of the falling rocks' movement by comparing the held information with the trajectory of the falling rocks 8 shown in the frame difference composite image G.
The quantitative information regarding the rockfall 8 extracted by the rockfall analysis unit 35 is sent to the rockfall prediction unit 36 and used for training the learning model and for predicting rockfalls 8 that will occur in the near future.

(Rockfall learning process (S40))
The rockfall learning process is a process of learning the relationship between (A) quantitative information (first information) and environmental data (second information) related to the rockfall 8, and (B) information related to the occurrence of a new rockfall 8 in the near future (third information). The rockfall prediction unit 36 performs machine learning using, for example, at least one of "the source of the rockfall, the size of the rockfall, and the characteristics of the rockfall movement" obtained as quantitative information related to the rockfall 8, and at least one of "precipitation amount per predetermined time, temperature, average wind speed, maximum wind speed, seismic intensity, and earthquake type" obtained as environmental data. The number and type of data used in learning are preferably selected in consideration of the geographical characteristics of the location where the cliff face 9 is located.
The rockfall prediction unit 36 has a learning model, and the learning model is trained by machine learning using learning data that is a combination of (A) quantitative information (first information) and environmental data (second information) related to the rockfall 8, and (B) information related to the occurrence of new rockfalls 8 in the near future (third information). The learning model is a learning system in machine learning, and can derive good results by comparing the results of classification, prediction, and judgment based on given data with the actual results that are correct answers, and adjusting various parameters. The learning model is, for example, a neural network, and the present embodiment will be described assuming a neural network. The learning model is also called a "learner" or "artificial intelligence (AI)".

As is well known, neural networks are an information processing technique that was created with the aim of having a computer mimic the workings of the human brain. Neural networks excel at complex nonlinear processing, and by using a learning function, they have the advantage that there is no need to formulate the relationship between explanatory variables and objective variables. Neural networks are generally composed of an input layer, a hidden layer, and an output layer. Explanatory variables are input to the input layer, and objective variables are output from the output layer. The hidden layer plays the role of relating the two. There are no restrictions on the number of hidden layers and the number of neurons in each hidden layer, and they can be set arbitrarily.
The method of training the learning model is not particularly limited, and the learning model is trained by, for example, the error backpropagation method. Specifically, (A) quantitative information (first information) and environmental data (second information) on the rockfall 8 constituting the training data are input to the input layer, and the intermediate layer is adjusted based on the error between the result output from the output layer and (B) information on the occurrence of a new rockfall 8 in the near future (third information) constituting the training data. In other words, the rockfall prediction unit 36 trains the learning model by machine learning so that (B) information on the occurrence of a new rockfall 8 in the near future (third information) is output by inputting (A) quantitative information (first information) and environmental data (second information) on the rockfall 8. The number of training data to be trained is not particularly limited, and the learning is terminated when, for example, an expected accuracy is reached.

(Rockfall prediction process (S50))
The rockfall prediction process is a process of predicting the occurrence of a new rockfall 8 in the near future (B) from (A) quantitative information (first information) related to the rockfall 8 and environmental data (second information). The rockfall prediction unit 36 predicts the occurrence of a rockfall in the near future using (A) the quantitative information and environmental data related to the rockfall 8 acquired in real time.
The rockfall prediction unit 36 has a trained learning model (trained model). The trained model is a learning model that has been machine-learned by the rockfall learning step (S40). In other words, the trained model is trained to output information (B) regarding the occurrence of a new rockfall 8 in the near future (third information) by inputting (A) quantitative information (first information) regarding the rockfall 8 and environmental data (second information). The rockfall prediction unit 36 inputs (A) quantitative information and environmental data regarding the rockfall 8 acquired in real time, and outputs (B) information regarding the occurrence of a new rockfall 8 in the near future that has not yet occurred (for example, "whether or not there is a rockfall" and "the time, source, scale, etc. of the rockfall to occur").

Next, the construction of the learning data and the learning model will be described. Fig. 8 shows an example of the observation data on which the learning data is based. The observation data shown in Fig. 8 was measured every hour. Each observation data consists of "date", "time", "number of rockfalls", "source of rockfall", "scale of rockfall", "precipitation [mm]", "temperature [℃]", "average wind speed [m/s]", "maximum wind speed [m/s]", and "earthquake (seismic intensity)".
The number of rockfalls here is the number of rockfalls that occurred in one hour. The source of the rockfall is the identification information of area H in FIG. 7. The scale of the rockfall is a further step-by-step evaluation of the value calculated by the above-mentioned formula (1). The amount of precipitation is the amount per hour (hourly rainfall), the average temperature and average wind speed are the average values per hour, and the maximum wind speed is the maximum instantaneous wind speed in one hour. The earthquake is the seismic intensity of earthquakes that occurred within one hour.
FIG. 9 shows an example of the learning model. For example, the learning model is trained using a set of information on the rockfall 8 at a certain first time point (for example, information shifted by one hour, such as 10 hours, 9 hours, 8 hours, . . . 1 hour before the time of prediction) and information on the rockfall 8 after a time has elapsed from the first time point (for example, information on the time of prediction). As a result, by inputting information on the rockfall 8 that has occurred and the environment into the trained model, it is possible to predict the "rockfall occurrence time", "rockfall source", and "rockfall scale" of the rockfall 8 that will occur in the near future. Note that, for example, "rockfall occurrence time", "rockfall source", "rockfall scale", and "precipitation" are given high priority as items of the learning data, and by including these items, it is possible to estimate with high accuracy the "rockfall occurrence time", "rockfall source", and "rockfall scale" of the rockfall that will occur in the near future.

An example of the observation data on which the learning data is based is shown in Fig. 10. The observation data shown in Fig. 10 was measured every hour. Each observation data consists of "date", "time", "presence or absence of rockfall", "number of rockfalls", "scale of rockfall", "precipitation [mm]", "temperature [℃]", "average wind speed [m/s]", "maximum wind speed [m/s]", and "earthquake (seismic intensity)".
Here, the presence or absence of rockfall is information on whether or not a rockfall 8 has occurred within one hour, and the number of rockfalls is the number of rockfalls that have occurred within one hour. The scale of rockfall is a further step-by-step evaluation of the value calculated by the above-mentioned formula (1). The amount of precipitation is the amount per hour (hourly rainfall), the average temperature and average wind speed are the average values per hour, and the maximum wind speed is the maximum instantaneous wind speed in one hour. Earthquakes are the seismic intensity of earthquakes that have occurred within one hour.
FIG. 11 shows an example of a learning model. For example, the input data is the number of rockfalls, the scale of rockfalls, the amount of precipitation, the temperature, the wind speed, and the earthquake, and the output data is whether a rockfall has occurred (whether a rockfall has occurred or not). For example, one hour is set as "one unit," the past 10 hours are the input data (see symbol "a" in FIG. 10), and the immediately following hour is the output data (see symbol "b" in FIG. 10). When multiple rockfalls 8 have occurred at the same location in one hour, the scale is set as, for example, a sum value. This makes it possible to predict whether a rockfall 8 will occur in the next hour from the observation data of the past 10 hours. It is also possible to learn to predict the probability that a rockfall 8 will occur in one hour. It is also possible to input several hours' worth of observation data into the learning model by performing addition processing, statistical processing, or the like.

As described above, according to the rockfall monitoring system 1 of the embodiment, the rockfall analysis unit 35 obtains quantitative first information on a rockfall 8 that has occurred in the monitoring area from the captured image by image analysis using the frame difference method, and the rockfall prediction unit 36 predicts a new rockfall 8 that will occur in the monitoring area using the first information. As a result, it is possible to issue a warning or alert taking into account a rockfall 8 that will occur in the near future, which contributes to improving management and measures against rockfalls 8.
Although the embodiment of the present invention has been described above, the present invention is not limited to this, and can be practiced within the scope of the claims.

REFERENCE SIGNS LIST 1 Rockfall monitoring system 2 Imaging unit 3 Monitoring unit 3A Image storage device 3B Image processing device 4 Notification unit 5 Power supply 8 Rockfall 9 Cliff face (monitoring area)
33 Image acquisition unit 34 Rockfall detection unit 35 Rockfall analysis unit 36 Rockfall prediction unit 37 Rockfall notification unit

Claims (5)

A rockfall monitoring system for monitoring the occurrence of rockfall, comprising:
An imaging unit that images a monitoring area;
a rockfall analysis unit that obtains quantitative first information regarding a rockfall occurring in the monitoring area from the captured image captured by the imaging unit by image analysis using a frame difference method;
a rockfall prediction unit that predicts a new rockfall that will occur in the monitoring area based on the first information and second information related to an environment at the time when the captured image is captured,
the rockfall prediction unit has a trained model that has been machine-learned to input a plurality of units of the first information and the second information that are consecutive in time series, with a predetermined time being one unit, and output third information regarding the new rockfall in the next unit of time ;
A rockfall monitoring system. The rockfall monitoring system according to claim 1, characterized in that the first information is any one of the source of the rockfall, the size of the rockfall, and the characteristics of the rockfall movement. a rockfall detection unit that detects a rockfall in an image captured by the imaging unit by a frame difference method;
The rockfall analysis unit obtains the first information from a frame difference image in which a rockfall is detected by the rockfall detection unit.
3. A rockfall monitoring system according to claim 1 or 2. The rockfall monitoring system according to any one of claims 1 to 3, further comprising an alarm unit that notifies the occurrence of a new rockfall. A rockfall monitoring method for monitoring the occurrence of a rockfall, comprising the steps of:
an imaging step of imaging a monitoring area;
a rockfall analysis step of obtaining quantitative first information regarding a rockfall occurring in the monitoring area from the captured image captured in the imaging step by image analysis using a frame difference method;
a rockfall prediction step of predicting a new rockfall that will occur in the monitoring area based on the first information and second information related to the environment at the time when the captured image is captured,
In the rockfall prediction step, a prediction is made using a trained model that has undergone machine learning so as to input a plurality of units of the first information and the second information that are consecutive in time series, with a predetermined time being one unit , and output third information regarding the new rockfall in the next unit of time .
A method for monitoring falling rocks.

Images