Fire detection system and installation location detection system

Abstract

Reduce false fire detections by fire detection systems. [Solution] A fire determination system according to one embodiment of the present invention comprises: a first determination means that determines whether or not there is a fire in a monitoring area based on information other than images obtained in the monitoring area; an access means that accesses a trained model that has been machine-trained using training data in which images are included as explanatory variables and information relating to the correctness of the determination result by the same type of determination means as the first determination means in the area where the image was taken is included as an objective variable; a shooting means that photographs the monitoring area; and a second determination means that the access means inputs the image taken by the shooting means at the time the determination by the first determination means was made as an explanatory variable and determines whether or not there is a fire based on the information obtained from the trained model.

Description

Technical Field

[0001] The present invention relates to a fire determination system and an installation location determination system.

Background Art

[0002] When the occurrence of a fire is detected by sensing the heat of the hot air flow generated by the fire, a technology that emits a sound to notify (report) the occurrence of the fire to the surroundings is known.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The fire determination system may erroneously determine that a fire has occurred due to the influence of disturbances such as sunlight, heat sources other than fire, and dust, even though a fire has not actually occurred.

[0005] The present invention provides a technology for reducing the false determination of a fire by a fire determination system.

Means for Solving the Problems

[0006] One aspect of the present disclosure provides a fire detection system comprising: a first determination means for determining whether or not there is a fire in a monitoring area based on information other than an image obtained in the monitoring area; an access means for accessing a trained model that has been machine-trained using training data in which an image is included as an explanatory variable and information relating to the correctness of the determination result by a determination means of the same type as the first determination means in the area where the image was taken is included as an objective variable; a shooting means for photographing the monitoring area; and a second determination means for determining whether or not there is a fire based on information obtained from the trained model, with the access means inputting an image taken by the shooting means at the time the determination by the first determination means was made as an explanatory variable.

[0007] Another aspect of this disclosure provides an installation location determination system comprising: one fire determination means for determining the presence or absence of a fire in a monitoring area based on information other than images obtained in the monitoring area; an access means for accessing a trained model for determining installation locations that has been machine-learned using training data which includes, as explanatory variables, attribute information indicating the attributes of the installation location of a fire determination means of the same type as the one fire determination means, and as a target variable, information regarding the correctness of the determination result by the fire determination means installed at the said installation location; and an installation location determination means which determines the suitability of a candidate installation location based on information obtained from the trained model for determining installation locations, after the access means inputs attribute information indicating the attributes of a candidate installation location for a fire determination means of the same type as the one fire determination means as explanatory variables. [Effects of the Invention]

[0008] According to the present invention, the number of false fire detections by the fire detection system is reduced. [Brief explanation of the drawing]

[0009] [Figure 1] A diagram showing the overall configuration of the fire detection system according to the embodiment. [Figure 2] A diagram showing the hardware configuration of a fire detection device, detector, and camera according to an embodiment. [Figure 3] A diagram showing the functional configuration of a fire detection device, detector, and camera according to an embodiment. [Figure 4] A sequence diagram showing the processing procedure for a fire detection system according to an embodiment. [Figure 5] This diagram shows the hardware configuration of the fire detection device included in a fire detection system and installation location determination system according to one modified example. [Figure 6] A diagram showing the functional configuration of a fire detection system and installation location determination system in one modified example. [Modes for carrying out the invention]

[0010] 1. Examples Figure 1 shows the overall configuration of the fire detection system 1. The fire detection system 1 comprises a fire detection device 10, a detector 20, and a camera 30. The fire detection device 10 is connected to the detector 20 and the camera 30 by wire or wireless. Although Figure 1 shows one detector 20 and one camera 30, this number may be two or more.

[0011] The fire detection device 10 is a device that determines the correctness of the detection result when the detector 20 detects a fire, that is, when it determines that there is a fire. When the detector 20 detects a fire in the monitored space (hereinafter referred to as the "monitoring area"), the fire detection device 10 receives a fire detection notification from the detector 20 that it has detected a fire. The fire detection device 10 also receives images of the monitoring area taken by the camera 30 from the camera 30. When the fire detection device 10 receives a fire detection notification from the detector 20, it determines whether the fire detected by the detector 20 is actually occurring, based on the images taken by the camera 30 and received from the camera 30 at the same time, that is, at the time the detector 20 determined that there was a fire. If the fire detection device 10 determines that there is actually a fire, it sends an alarm execution instruction to the detector 20 to execute the alarm. If it determines that there is no actual fire, it sends an alarm non-execution instruction to the detector 20 to not execute the alarm.

[0012] The sensor 20 is a device that senses a fire when it occurs in the monitored area. The monitored area of the sensor 20 is, for example, collective housing such as ordinary houses, condominiums, hotels, office buildings, commercial facilities, etc.

[0013] As long as the method by which the sensor 20 senses a fire is a method of determining the presence or absence of a fire in the monitored area based on information other than the image obtained in the monitored area, any method may be used. That is, the sensor 20 may be a flame sensor that senses a flame based on the output signal of a sensor that senses electromagnetic waves of a specific wavelength, a smoke sensor that senses smoke based on the output signal of a sensor that senses light diffusely reflected by floating substances in the air, a heat sensor that senses the temperature rise rate or temperature by thermal expansion of air, a bimetal, or a thermistor, a CO sensor that senses CO in the air based on the current value flowing through an electrolyte solution, or the like.

[0014] When the sensor 20 senses a fire, it transmits a fire detection notification to the fire determination device 10.

[0015] The camera 30 is a device that photographs the monitored area. The camera 30 transmits the photographed image to the fire determination device 10.

[0016] FIG. 2 is a diagram showing the hardware configuration of the fire determination device 10, the sensor 20, and the camera 30.

[0017] The fire determination device 10 is a computer including a processor 151, a communication IF (Interface) 152, and a memory 153.

[0018] The processor 151 controls the operations of other components of the fire determination device 10 by executing a program for the fire determination device 10 stored in the memory 153. The communication IF 152 communicates wirelessly or wired with each of the sensor 20 and the camera 30. The memory 153 continuously stores various information including the program executed by the above-described processor 151.

[0019] The sensor 20 includes a processor 251, a communication IF 252, a memory 253, a sensor 254, and a speaker 255.

[0020] The processor 251 controls the operations of the other components of the sensor 20 by executing the programs stored in the memory 253. The communication IF 252 communicates wirelessly or wired with the fire detection device 10. The memory 253 continuously stores various data including the programs executed by the above-mentioned processor 251.

[0021] The sensor 254 measures the physical quantity for detecting a fire in the monitoring area. The physical quantity measured by the sensor 254 differs depending on the method by which the sensor 20 detects a fire. For example, when the sensor 20 is a flame sensor, the sensor 254 is a sensor that senses electromagnetic waves of a specific wavelength emitted by the flame.

[0022] The speaker 255 emits sound.

[0023] The camera 30 includes a processor 351, a communication IF 352, a memory 353, a lens 354, and an image sensor 355.

[0024] The processor 351 controls the operations of the other components of the camera 30 by executing the programs stored in the memory. The communication IF 352 communicates wirelessly or wired with the communication partner device. The memory 353 continuously stores various data including the programs executed by the above-mentioned processor 351.

[0025] The lens 354 condenses the light emitted from the subject and projects the condensed light onto the image sensor 355. The image sensor 355 includes a sensor group in which a large number of elements sensitive to light in a predetermined wavelength band are two-dimensionally arranged, and generates an image of the subject from the light projected from the lens 354. The images continuously generated by the image sensor 355 are arranged in time series to form a moving image.

[0026] Figure 3 is a diagram showing the functional configuration of the fire detection system 1.

[0027] When the processor 151 of the fire detection device 10 (Figure 2) executes a program stored in the memory 153, the fire detection device 10 functions as a device equipped with communication means 101, storage means 102, image recognition means 103, access means 104, and determination means 105 (an example of a second determination means).

[0028] The communication means 101 communicates various types of information between the detector 20 and the camera 30. The information communicated by the communication means 101 includes fire detection notifications transmitted from the detector 20, images transmitted from the camera 30, and instructions to activate and deactivate an alarm transmitted to the detector 20.

[0029] The memory means 102 stores various types of information. The information stored by the memory means 102 includes images (videos) that the communication means 201 continuously receives from the camera 30, and the trained model M (an example of a trained model for fire detection) described below.

[0030] The trained model M is generated in a device different from the fire detection system 1 (Figure 1) and stored in the fire detection device 10. The trained model M is a machine learning model that uses training data in which images are included as explanatory variables and information regarding the correctness of the judgment result by the detection means 202 of the detector 20 and the same type of detection means in the area where the image was captured is included as the objective variable.

[0031] The training data used to generate the trained model M is generated using information obtained when a detector of the same type as detector 20 installed in various different monitoring areas detects a fire, and is divided into the following two types. (1) Training data in which images taken by a camera within the monitoring area of ​​a detector at the time the detector detected a fire are used as explanatory variables, and the label "Fire Present" indicating that a fire was actually occurring is used as the dependent variable. (2) Training data in which images taken by a camera of the monitoring area of ​​a detector at the time a fire was detected by the detector are used as explanatory variables, and the label "No Fire" indicating that no fire actually occurred is used as the dependent variable.

[0032] Furthermore, the images used as explanatory variables for the training data used to generate the trained model M are videos captured by a camera, that is, a set of images extracted from a series of images arranged in time, showing parts of specific objects recognized by known image recognition techniques (hereinafter referred to as "videos without background"). Here, specific objects include people, animals, puddles, automobiles, warning lights on construction vehicles, flames, etc.

[0033] Thus, instead of using the images of the monitored area directly as explanatory variables for the training data used to generate the trained model M, images extracted from those images that show specific objects that affect fire detection are used. This reduces the influence of images of different background areas in images taken of different monitored areas. As a result, training data for training the trained model M can be obtained using information from various different monitored areas, and the same trained model M can be used even if the monitored area of ​​the fire detection system 1 is different.

[0034] Furthermore, by using multiple images arranged in a time series instead of a single image as explanatory variables in the training data used to generate the trained model M, the trained model M outputs different information depending on the time-series information of the images of the monitored area, that is, how the color, shape, size, etc. of people, flames, etc., contained in the images of the monitored area change over time. As a result, the accuracy of the information output by the trained model M is higher than that of a trained model generated using training data with a single image as an explanatory variable.

[0035] The method used to generate the trained model M can be any machine learning method that uses supervised data. A typical example of such a machine learning method is deep learning, but the trained model M may also be generated using a neural network with a small number of hidden layers, or a machine learning method other than a neural network (such as Naive Bayes, decision trees, or support vector machines).

[0036] The trained model M, which was created using the aforementioned training data, outputs information indicating whether or not a fire is actually occurring when it receives a background-free video extracted from a video of a specific object taken in the monitored area. Hereafter, the trained model M will output either the label "Fire Present," which indicates that a fire is actually occurring, or the label "No Fire," which indicates that a fire is not actually occurring.

[0037] When the communication means 101 receives a fire detection notification from the sensor 20, the image recognition means 103 reads the video received by the communication means 101 from the camera 30 at the same time from the storage means 102, recognizes a specific object from the read video, and generates a background-free video by extracting the portion of the recognized object from the video.

[0038] When the image recognition means 103 generates a video without a background, the access means 104 accesses the trained model M, inputs the video without a background generated by the image recognition means 103 into the trained model M as an explanatory variable, and obtains the information output from the trained model M, namely, "fire present" or "no fire present".

[0039] The determination means 105 determines whether or not there is a fire based on the information obtained by the access means 104 from the trained model M. Specifically, if the access means 104 obtains "fire present" from the trained model M, the determination means 105 determines that a fire has actually occurred, and if the access means 104 obtains "no fire present" from the trained model M, the determination means 105 determines that a fire has not actually occurred.

[0040] If the determination means 105 determines that a fire has actually occurred, it generates an alarm activation instruction and passes it to the communication means 101. In that case, the communication means 101 transmits the alarm activation instruction to the detector 20. On the other hand, if the determination means 105 determines that a fire has not actually occurred, it generates an alarm non-activation instruction and passes it to the communication means 101. In that case, the communication means 101 transmits the alarm non-activation instruction to the detector 20.

[0041] The above is a description of the functional configuration of the fire detection device 10. Next, the functional configuration of the detector 20 will be explained.

[0042] When the processor 251 of the sensor 20 (Figure 2) executes a program stored in the memory 253, the sensor 20 functions as a device comprising a communication means 201, a sensing means 202, a determination means 203 (an example of a first determination means), and an alarm means 204.

[0043] The communication means 201 communicates various information with the fire detection device 10. The information communicated by the communication means 201 includes a fire detection notification to be sent to the fire detection device 10, and instructions to activate and deactivate an alarm to be sent from the fire detection device 10.

[0044] The sensing means 202 continuously measures physical quantities in the monitoring area and outputs information to the determination means 203 for determining whether or not there is a fire in the monitoring area. For example, if the detector 20 is a flame detector that detects flames by the intensity of infrared rays, the sensing means 202 measures, for example, the intensity of infrared rays in a predetermined wavelength band and outputs a signal indicating the measurement result to the determination means 203.

[0045] The determination means 203 determines whether or not there is a fire in the monitoring area based on the measurement results indicated by the signal output from the sensing means 202.

[0046] The alarm generation means 204 performs the process for generating an alarm in any of the following cases.

[0047] (1) If the communication means 201 receives an alarm activation instruction from the fire detection device 10 before a predetermined time has elapsed after the sensing means 202 has detected a fire. (2) If, after the sensing means 202 has detected a fire, the communication means 201 has not received either an alarm activation instruction or an alarm deactivation instruction from the fire detection device 10 within a predetermined time.

[0048] Furthermore, even if the sensing means 202 detects a fire, if the communication means 201 receives an instruction from the fire detection device 10 not to issue an alarm before a predetermined time has elapsed, the alarm issuing means 204 will not perform the necessary processing for issuing an alarm.

[0049] The alarm activation means 204, as part of the process for activating an alarm, pronounces an audio message such as "There is a fire."

[0050] The above is a description of the functional configuration of the sensor 20. Next, we will explain the functional configuration of the camera 30.

[0051] When the processor 351 of camera 30 (Figure 2) executes a program stored in memory 353, camera 30 functions as a device equipped with communication means 301 and imaging means 302.

[0052] The communication means 301 communicates various information with the fire detection device 10. The information communicated by the communication means 301 includes video footage of the monitored area captured by the imaging means 302.

[0053] The shooting means 302 continuously photographs the monitored area and generates a video.

[0054] The above is a description of the functional configuration of the fire detection system 1. Next, we will explain the operation of the fire detection system 1.

[0055] Figure 4 shows a sequence diagram illustrating the steps taken by the fire detection system 1 to determine whether or not there is a fire in the monitored area.

[0056] The series of processes shown in Figure 4 are executed each time the detector 20 detects a fire. Images are continuously transmitted from the camera 30 to the fire detection device 10 regardless of whether the detector 20 has detected a fire or not. The fire detection device 10 stores the images received from the camera 30 sequentially, associating them with the time information of receipt. The fire detection device 10 then sequentially deletes older images from its stored data as a predetermined time has elapsed since their receipt.

[0057] In such a state, when the detector 20 detects a fire (step S101), the detector 20 sends a fire detection notification to the fire detection device 10 (step S102).

[0058] When the fire detection device 10 receives a fire detection notification from the detector 20, it generates a video without a background from a video of multiple images stored at that time arranged in chronological order (step S103).

[0059] Next, the fire detection device 10 inputs the generated background-free video into the trained model M (step S104).

[0060] Next, the fire detection device 10 determines whether or not a fire has occurred based on the information output from the trained model M (step S105).

[0061] If the fire detection device 10 determines that a fire has occurred (step S105; Yes), it sends an alarm activation instruction to the detector 20 (step S106).

[0062] On the other hand, if the fire detection device 10 determines that no fire has occurred (step S105; No), it sends an instruction to the detector 20 not to activate the alarm (step S107). Once the fire detection device 10 has sent either an alarm activation instruction or an alarm non-activation instruction, it terminates the process according to the sequence shown in Figure 4.

[0063] If the detector 20 receives an instruction to activate the alarm from the fire detection device 10, it activates the alarm (step S108). On the other hand, if the detector 20 receives an instruction not to activate the alarm from the fire detection device 10, it does not activate the alarm and terminates the process according to the sequence in Figure 4.

[0064] However, for example, if a communication failure occurs between the fire detection device 10 and the detector 20, the detector 20 may not receive either an alarm activation instruction or an alarm deactivation instruction from the fire detection device 10 as a response to the fire detection notification after sending the fire detection notification in step S102. In such a case, the detector 20 will start activating the alarm.

[0065] Specifically, after the detector 20 transmits a fire detection notification to the fire detection device 10 in step S102, it determines at sufficiently short intervals whether the elapsed time since the transmission of the fire detection notification has reached a predetermined time (step S109). If it determines that the elapsed time since the transmission of the fire detection notification has reached a predetermined time (step S109; Yes), the detector 20 determines whether it has received a response from the fire detection device 10 to the transmission of the fire detection notification, i.e., an instruction to activate the alarm or an instruction not to activate the alarm (step S110).

[0066] If the detector 20 has received a response from the fire detection device 10 to the transmission of a fire detection notification (step S110; Yes), it terminates the process according to the sequence in Figure 4. If the detector 20 has received an alarm activation instruction as a response to the transmission of the fire detection notification from the fire detection device 10, it continues the alarm activation that was started in step S108. On the other hand, if the detector 20 has received an alarm deactivation instruction as a response to the transmission of the fire detection notification from the fire detection device 10, it will terminate the process according to the sequence in Figure 4 without activating an alarm.

[0067] In step S110, if the detector 20 determines that it has not received a response from the fire detection device 10 to the transmission of a fire detection notification (step S110; No), the detector 20 starts to emit an alarm (step S111). After that, the detector 20 terminates the process according to the sequence in Figure 4.

[0068] Furthermore, the alarm activated by the sensor 20 in step S108 or S111 will continue until someone performs an operation to stop the alarm on the sensor 20.

[0069] The above is a description of the operation of the fire detection system 1.

[0070] According to the fire detection system 1 described above, if the detector 20 falsely detects a fire, the fire detection device 10 detects the false detection, and a false alarm by the detector 20 is avoided.

[0071] 2. Variations The embodiment described above is one embodiment of the present invention, and various modifications are possible. Modifications are described below. Note that two or more of the following modifications may be combined.

[0072] (Fire detection based on images) In addition to detecting fire with the detector 20, the presence or absence of a fire may also be determined based on images captured by the camera 30. In one modified embodiment, the fire detection device 10 includes a determination means (an example of a third determination means, hereinafter referred to as the "image fire determination means") that determines the presence or absence of a fire based on video received by the fire detection device 10 from the camera 30, in addition to the determination means 105 that determines the presence or absence of a fire based on information output from the trained model M. The image fire determination means determines, for example, that a fire has occurred when the image recognition means 103 recognizes flames. The determination means 105 determines whether or not a fire has actually occurred based on information output from the trained model M, in addition to when it receives a fire detection notification from the detector 20 and when the image fire determination means determines that a fire has occurred.

[0073] (Regarding object recognition) In the above embodiment, the explanatory variables of the trained model M are images of a specific object (image without background) recognized and extracted from the image of the monitored area (image with background) captured by the camera. However, instead, images with background may be used as explanatory variables of the trained model M. In that case, the effect of using images without background as described above will not be obtained, but false alarms will be reduced compared to the case where false alarm avoidance using the trained model M is not performed.

[0074] (About the video) In the above embodiment, video was used as an explanatory variable for the trained model M. However, instead, a still image (i.e., a single image, or multiple images not arranged in time series) may be used as an explanatory variable for the trained model M. In that case, the effects of using video as described above will not be obtained, but false alarms will be reduced compared to the case where false alarm avoidance using the trained model M is not performed.

[0075] (Regarding the type of sensor) As previously described, the method by which the detector 20 detects a fire can be any method, as long as it is based on information other than images. However, when the determination means 203 of the detector 20 determines the presence or absence of a fire based on time-changing information obtained in the monitoring area by the sensing means 202 (sensor 254), it is particularly desirable that video, i.e., time-changing images, be used as explanatory variables for the trained model M. This is because, since the detector 20 detects a fire based on time-changing information, using a trained model M trained on training data that includes time-changing images as explanatory variables increases the likelihood that the trained model M can detect a false detection of a fire with high accuracy if the detector 20 misdetects one.

[0076] (Regarding image types) In the embodiments described above, the camera 30 is assumed to be a camera that generates an image represented by visible light, but the method by which the camera 30 captures images is not limited to this. For example, the camera 30 may be an infrared camera that senses infrared light and generates an image, a thermographic camera that senses heat and generates an image, or a distance image camera (distance image sensor) that generates an image based on the distance calculated from the time it takes for a test wave such as infrared light to reflect off the surface of an object and return.

[0077] (Regarding where trained models are stored) In the embodiments described above, the trained model M is assumed to be stored in the fire detection device 10 of the fire detection system 1, but the location where the trained model M is stored is not limited to this. For example, the trained model M may be stored in a system different from the fire detection system 1 (for example, an external server device that can communicate with the fire detection device 10), and the access means 104 of the fire detection device 10 may access the trained model M stored in that system via the communication means 101.

[0078] (Regarding updating trained models) The trained model M may be updated using training data generated from information acquired during the operation of the fire detection system 1 or a system of the same type. For example, in the fire detection system 1 and similar systems, when a fire is detected by a detector, a human observes the image captured by the camera and labels it as "fire present" or "no fire present." This data is then used as an explanatory variable to generate training data with the label as the objective variable. The trained model M may then be updated using this generated training data. Alternatively, the trained model M may be updated using training data generated from information collected from multiple fire detection systems on a server device in the cloud.

[0079] (Regarding the format of information output by the trained model) In the embodiment described above, the trained model M outputs information such as "fire present" or "no fire" as the target variable. However, the information output by the trained model M may be in any format, as long as it indicates the likelihood of a fire occurring. For example, the trained model M may output a numerical value indicating the probability that a fire has occurred. In that case, the determination means 105 of the fire determination device 10 may determine that a fire has occurred if the numerical value output from the trained model M is above a predetermined threshold (or greater than a predetermined threshold), and determine that no fire has occurred otherwise.

[0080] (Regarding the method of issuing a warning) In the embodiments described above, the fire alarm is triggered by the sound emitted by the detector 20, but the method of triggering the alarm is not limited to this. For example, in addition to or instead of emitting a sound, the detector 20 may notify people in the vicinity of the fire by lighting a lamp or displaying a message such as "A fire has broken out." Alternatively, the fire detection device 10 may notify people in the vicinity of the fire, either in place of or in addition to the detector 20.

[0081] Furthermore, the fire detection system 1 may also include multiple sets of fire detection devices 10, detectors 20, and cameras 30, a receiver that receives signals transmitted from each of the multiple fire detection devices 10, and a bell, emergency broadcasting equipment, smoke control equipment, etc., connected to the receiver. In that case, when a fire is detected, the bell and emergency broadcasting equipment will operate according to instructions from the receiver, and a fire alarm will be issued.

[0082] Furthermore, the fire detection system 1 may notify an external party that a fire has occurred as part of the process for issuing an alarm. For example, if the receiver described above is connected to a higher-level system, and a fire is detected, the higher-level system may be notified that a fire has occurred.

[0083] (Regarding the arrangement of components in the fire detection system) In the embodiment described above, the components of the fire detection system 1 are arranged in one of the fire detection device 10, the detector 20, and the camera 30 as shown in Figure 3. However, this arrangement is just an example, and other arrangements may be adopted. For example, the detection means 203 provided in the detector 20 may be arranged in the fire detection device 10. In that case, the detector 20 transmits the information acquired by the sensing means 202 (sensor 254) to the fire detection device 10, the fire detection device 10 detects a fire, and the correctness of the detection result is verified using a trained model M. Furthermore, two or three of the fire detection device 10, the detector 20, and the camera 30 may be configured as a single device. Also, one or more of the fire detection device 10, the detector 20, and the camera 30 may be configured as multiple devices.

[0084] (Regarding the determination of the suitability of the installation location) The trained model M used in the above-described embodiment outputs information for determining the presence or absence of a fire. The fire detection system 1 may be configured as an installation location determination system that determines the suitability of the installation location of a detector 20 (an example of the first determination means) by using a trained model that outputs information for determining the suitability of the installation location of a detector of the same type as the detector 20, in addition to or instead of the trained model M.

[0085] Figure 5 shows the hardware configuration of the fire detection device 10, which is included in the fire detection system 1 that also serves as an installation location determination system, according to this modified example. In this modified example, the hardware configuration of the detector 20 and the camera 30 is the same as in the embodiment described above. In this modified example, the fire detection device 10 includes a touchscreen 154 in addition to the components of the fire detection device 10 in the embodiment described above. The touchscreen 154 accepts input of characters, etc., by touch operation by the user and displays characters, etc., to the user.

[0086] Figure 6 shows the functional configuration of the fire detection system 1 according to this modified example. In this modified example, the fire detection device 10 includes, in addition to the functional components of the fire detection device 10 in the above-described embodiment, an attribute information acquisition means 106, an access means 107 (an example of a second access means), an installation location determination means 108, an intermediate layer information acquisition means 109 (an example of an acquisition means), and a presentation means 110. Furthermore, in this modified example, the storage means 102 of the fire detection device 10 stores a learned model N (an example of a learned model for installation location determination) in addition to the learned model M.

[0087] The attribute information acquisition means 106 acquires attribute information entered by the user. The attribute information acquired by the attribute information acquisition means 106 is information indicating the attributes of candidate installation locations for the detector 20. Examples of information indicating the attributes of candidate installation locations for the detector 20 include, but are not limited to, the height of the candidate installation location for the detector 20 from the ground, whether or not there is a busy road within a predetermined distance from the candidate installation location for the detector 20, whether or not there are structures such as trees or fences that are easy for people to climb within a predetermined distance from the candidate installation location for the detector 20, whether or not there is a kindergarten or the like within a predetermined distance from the candidate installation location for the detector 20, whether or not it is exposed to direct sunlight, etc.

[0088] The access means 107 accesses the trained model N, inputs the attribute information acquired by the attribute information acquisition means 106 as explanatory variables into the trained model N, and acquires the information output by the trained model N. The information output from the trained model N may be any of the following: information indicating the degree of fit, such as "Goodness of Fit: 93", or information such as "Not suitable" or "Suitable".

[0089] The installation location determination means 108 determines whether a candidate installation location for the sensor 20 is suitable based on the information acquired by the access means 107 from the learned model N.

[0090] If the installation location determination means 108 determines that a candidate installation location for the sensor 20 is unsuitable, the intermediate layer information acquisition means 109 acquires, for example, the information generated in the intermediate layer by the trained model N when attribute information of the candidate installation location is input as an explanatory variable, via the access means 107.

[0091] If the installation location determination means 108 determines that a candidate installation location for the detector 20 is unsuitable, the presentation means 110 presents information regarding a suitable installation location for the detector 20 based on the information acquired by the intermediate layer information acquisition means 109.

[0092] The trained model N is a pre-trained model developed using machine learning with training data that includes attribute information indicating the installation location of detectors of the same type as detector 20 as explanatory variables, and information regarding the accuracy of the judgment results (fire detection results) from detectors installed at those locations as the objective variable. Furthermore, the trained model N is a neural network type pre-trained model.

[0093] The training data used to generate the trained model N is, for example, the following:

[0094] (Explanatory variables) Installation height: 200cm, presence of easily climbable structures: none, presence of busy roads: none, presence of kindergartens, etc.: none. Presence of direct sunlight: none. (Dependent variable) No false detections.

[0095] (Explanatory variables) Installation height: 100cm, presence of easily climbable structures: none, presence of busy roads: yes, presence of kindergartens, etc.: none. Presence of direct sunlight: yes. (Dependent variable) False detection occurred.

[0096] The above training data includes, for each detector of the same type as the fire detection system 1, which is deployed in various monitoring areas, attribute information of the installation location of the detector as an explanatory variable, and information on whether or not the detector has previously made a false alarm as an objective variable.

[0097] The storage means 102 of the fire detection device 10 stores a trained model N that has been machine-learned using the training data described above.

[0098] A user, such as a worker, who intends to install a new detector 20 in any location within the monitoring area, inputs the attribute information of the candidate locations for the detector 20, i.e., the locations where they wish to install the detector 20, into the touchscreen 154 of the fire detection device 10.

[0099] The fire detection device 10 inputs attribute information entered by the user as explanatory variables into a trained model N, and determines the suitability of the candidate installation location based on the information output from the trained model N.

[0100] If the fire detection device 10 determines that a candidate installation location is suitable, it displays a message such as "Suitable" on the touchscreen 154.

[0101] On the other hand, if the fire detection device 10 determines that a candidate installation location is unsuitable, it acquires information generated in the intermediate layer by the trained model N. The information generated in the intermediate layer by the trained model N includes information on which items of the attribute information of the candidate installation location contribute most to the determination that the candidate installation location is unsuitable. For example, if the installation height is 100 cm or less and many false detections are occurring due to children's mischief, then if the installation height of the candidate installation location is 100 cm or less, information is generated in the intermediate layer to lower the suitability of that candidate installation location.

[0102] Based on the information acquired as described above and generated in the intermediate layer by the trained model N, the fire detection device 10 displays information on the touchscreen 154 regarding a suitable location for installing the detector 20, such as, "The installation height is too low. Is there a location where it can be installed at a height of 200 cm or more?"

[0103] According to this modified example, a user who intends to install a new detector 20 can install it in a location with fewer false detections. As a result, false detections of fires by the fire detection system 1 are reduced.

[0104] In this modified example, the attribute information input to the trained model N as explanatory variables may include images taken at the installation location or candidate installation locations. In this case, the trained model N is trained to estimate the features of images of objects that cause false fire detection (e.g., easily climbable structures, busy roads, etc.) from the image features input as explanatory variables in the training data, and to assign a larger weight to such features. Then, the trained model N determines whether the installation location for the detector 20 is suitable based on the features extracted from the images input as explanatory variables during operation. [Explanation of Symbols]

[0105] 1...Fire detection system, 10...Fire detection device, 20...Detector, 30...Camera, 101...Communication means, 102...Storage means, 103...Image recognition means, 104...Access means, 105...Determination means, 106...Attribute information acquisition means, 107...Access means, 108...Installation location determination means, 109...Intermediate layer information acquisition means, 110...Presentation means, 151...Processor, 152...Communication IF, 153...Memory, 154...Touchscreen, 201...Communication means, 202...Sensing means, 203...Determination means, 204...Alarming means, 251...Processor, 252...Communication IF, 253...Memory, 254...Sensor, 255...Speaker, 301...Communication means, 302...Shooting means, 351...Processor, 352...Communication IF, 353...Memory, 354...Lens, 355...Image sensor.

Claims

1. A first determination means for determining whether or not there is a fire in the monitoring area based on information other than images obtained in the monitoring area, An access means for accessing a trained model that has been machine-trained using training data that includes an image as an explanatory variable and information regarding the correctness of the result of a determination by the same type of determination means as the first determination means in the region where the image was captured as an explanatory variable, A means for photographing the aforementioned monitoring area, The access means inputs the image captured by the photography means as an explanatory variable at the time the determination by the first determination means is made, and the second determination means determines whether or not there is a fire based on the information obtained from the trained model. A fire detection system equipped with the following features.

2. A signaling means that performs a process for issuing a signal according to the result of the determination by the second determination means. The fire detection system according to claim 1, comprising:

3. The system includes a third determination means for determining whether or not there is a fire in the monitoring area based on an image captured by the aforementioned photographic means, The second determination means determines whether or not there is a fire based on the determination result of the third determination means. The fire detection system according to claim 1.

4. The aforementioned trained model is a machine learning model that uses training data in which explanatory variables include images extracted from captured images that represent parts of specific objects recognized by image recognition. The system includes an image recognition means that recognizes a specific object from an image captured by the aforementioned shooting means, The access means takes an image of a specific part of an object recognized by the image recognition means as an explanatory variable and obtains information from the trained model. The fire detection system according to claim 1.

5. The aforementioned image is a video, The first determination means determines whether or not there is a fire based on the time-varying information obtained in the monitoring area. The fire detection system according to claim 1.

6. (Determination of suitability of installation location) A system for determining the suitability of the installation location of the first determination means, which is included in the fire determination system according to claim 1, When the trained model accessed by the access means is a trained model for fire detection, A second access means accesses a trained model for determining installation locations, which is machine-trained using training data that includes attribute information indicating the attributes of the installation location of a determination means of the same type as the first determination means as an explanatory variable, and information regarding the correctness of the determination result by the determination means installed at the said installation location as an objective variable. The second access means inputs attribute information indicating the attributes of the candidate installation location of the first determination means as explanatory variables, and the installation location determination means determines the suitability of the candidate installation location based on the information obtained from the trained model for installation location determination. A system for determining the installation location, equipped with the following features.

7. An access means for accessing a pre-trained model for determining installation location, which is machine-trained using training data that includes, as an explanatory variable, attribute information indicating the attributes of the installation location of a fire determination means of the same type as the fire determination means, and as an objective variable, information regarding the correctness of the determination result by the fire determination means installed at the said installation location. The access means inputs attribute information indicating the attributes of candidate installation locations for a fire detection means of the same type as the first fire detection means as explanatory variables, and the installation location determination means determines the suitability of the candidate installation location based on the information obtained from the trained model for installation location determination. A system for determining the installation location, equipped with the following features.

8. The aforementioned attribute information is an image taken at the installation location or a candidate installation location. The installation location determination system according to claim 6 or 7.

9. The aforementioned pre-trained model for determining the installation location is a neural network type pre-trained model. If the installation location determination means determines that the candidate installation location is unsuitable, the acquisition means acquires information generated in the intermediate layer of the pre-trained model for installation location determination when attribute information indicating the attributes of the candidate installation location is input as an explanatory variable. Based on the information generated in the intermediate layer acquired by the acquisition means, a presentation means presents information regarding a suitable location for installation. The installation location determination system according to claim 6 or 7, comprising:

Images