On-device artificial intelligence-based fire detection device and method applied thereto

Abstract

Disclosed are an on-device artificial intelligence-based fire detection device and a method applied thereto. An on-device artificial intelligence-based fire detection device according to the present invention comprises: a camera; a memory; and at least one processor, wherein the processor drives the camera to capture an image of the surroundings, generates a result obtained by analyzing the captured image in comparison with dedicated fire detection data learned in advance and stored in the memory, and activates one or more preset fire alarm output means when the generated comparative analysis result is a sign of a fire or a fire has occurred.

Description

On-device artificial intelligence-based fire detection device and method applied thereto

[0001] The present invention relates to an on-device artificial intelligence-based fire detection device and a method applied thereto, and more specifically, to a universally usable on-device artificial intelligence-based fire detection device and a method applied thereto.

[0002] Typically, fires often occur at unexpected times and in unexpected places.

[0003] Recently, the increasing number of fires involving electric vehicles has become a major issue. As fires occur while charging in underground parking lots and initial detection fails, the damage extends not only to the electric vehicles and charging stations but also to the entire underground parking lot and apartments above ground.

[0004] An electric vehicle is a vehicle that generates driving force by supplying electrical energy from a high-voltage battery to an electric motor; because it does not use fossil fuels, it is an eco-friendly vehicle with almost no exhaust gas emissions or noise.

[0005] As such, electric vehicles offer many eco-friendly advantages, and numerous policies are being established to expand their use. Technologically, however, active development is being pursued regarding both the functional aspects of electric vehicles and the infrastructure supporting them. Nevertheless, it can be said that we are still in a transitional phase across all sectors, and we are currently in the process of addressing the resulting issues and making improvements.

[0006] The batteries mainly used in electric vehicles are lithium-ion batteries. If the separator that separates the positive and negative electrodes, which are the main components of the battery, is damaged and the two electrodes come into contact, excessive current flows and heat is generated, which can lead to a fire.

[0007] For example, fires or explosions are occurring in parking lots while electric vehicles are being charged due to issues with the battery, internal short circuits connecting the battery, and chargers. Since electric vehicle fires have the potential to cause widespread damage to surrounding vehicles, facilities, and buildings, significant efforts and urgent improvements are required to protect valuable lives and property.

[0008] What requires even more urgent improvement is the need to establish facilities that enable early detection and intervention in the event of a fire involving an electric vehicle.

[0009] However, installing fire detection equipment requires not only equipment costs and expert fees for installation, but also, given that the usage of conventional engine vehicles and electric vehicles is mixed and conventional engine vehicles still account for a larger share, it is not easy to secure the necessary funds for fire detection in electric vehicles by obtaining the consent of facilities and buildings such as apartments. Even if such funds are secured through policy support, there is a possibility that neglect of equipment management will result from a decline in judgment regarding the continuous utility value of fire detection equipment, as people become desensitized to occasional fires in electric vehicles.

[0010] A solution is needed to resolve these problems.

[0011] Accordingly, the present invention was created to solve the above problems, and the objective of the present invention is to provide an on-device artificial intelligence-based fire detection device that is easy to build and can substantially provide continuous utility value, and a method applied thereto.

[0012] The purpose of the invention is not limited to the purposes mentioned above, and other unmentioned purposes will be clearly understood by those skilled in the art from the description below.

[0013] An on-device artificial intelligence-based fire detection device according to the first aspect of the present invention for achieving the above objective comprises a camera, a memory, and at least one processor, wherein the processor drives the camera to capture images of the surroundings, generates a result of comparing and analyzing the captured images with dedicated fire detection data that has been pre-learned and stored in the memory, and executes one or more pre-set fire alarm output means when the generated result of comparison and analysis indicates a fire occurrence precursor state or a fire is occurring.

[0014] The thermal imaging camera adapter may further include a thermal imaging camera function or be connected to the thermal imaging camera adapter through an external interface.

[0015] An on-device AI-based fire detection device can be equipped as a smartphone or tablet PC.

[0016] The above dedicated fire detection data can be updated via a network or with dedicated fire detection data for updating received through an external interface.

[0017] The recording of the above video operates in CCTV mode, and when an object appearing in the video moves and then stops, the CCTV mode can be switched to fire detection mode.

[0018] In the above fire detection mode, fire-related key point data of the image can be extracted and then compared with the dedicated fire detection data.

[0019] In the above CCTV mode, when an object appears in the video or when the object that appeared moves, the captured video can be stored in the memory or transmitted through a connected network.

[0020] The captured video stored in the memory can be automatically deleted after a predetermined period has elapsed.

[0021] It is linked with a management server via a network, and the management server can centrally manage notifications and measures in the event of a fire.

[0022] If video analysis indicates that the above-mentioned fire is in a precursor state or that a fire is in progress, corresponding fire occurrence information is generated, and the authenticity of the generated fire occurrence information can be authenticated through the management server.

[0023] A fire detection method based on on-device artificial intelligence according to a second aspect of the present invention for achieving the above objective comprises, in a fire detection device, the step of driving a camera to capture images of the surroundings; the step of generating a comparison analysis result by comparing and analyzing the captured images with dedicated fire detection data that has been pre-learned and stored; and the step of executing one or more pre-set fire alarm output means when the comparison analysis result indicates a fire occurrence precursor state or a fire is occurring.

[0024] The on-device artificial intelligence-based fire detection method may further include the step of switching the CCTV mode to a fire detection mode when the recording of the video operates in CCTV mode and an object appearing in the video moves and then stops.

[0025] Accordingly, the present invention has the advantage of easily establishing fire detection facilities in areas where fire is a concern (e.g., parking and charging areas of electric vehicles) through a fire detection device that is easy to provide universally (e.g., smartphones, tablet PCs, etc.).

[0026] In addition, the present invention has the advantage that even non-experts can directly install fire detection equipment in areas where fire is a concern (e.g., parking areas and charging areas of electric vehicles).

[0027] In addition, the present invention is not a facility that prepares only for the occurrence of a fire, but operates in CCTV mode before a fire occurs to create continuous utility value, and thereby has the advantage of enabling continuous management of the fire detection device.

[0028] In addition, the present invention has the advantage of enabling active initial response measures in the event of an actual fire, as it can verify attempts to maliciously cause a fire to malfunction.

[0029] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description in the claims.

[0030] FIG. 1 is a configuration diagram showing a fire detection device according to one embodiment of the present invention.

[0031] Figure 2 is an example diagram showing an example of a fire detection device of Figure 1 installed at the site.

[0032] Figure 3 is a configuration diagram showing the connection configuration between a smartphone and a thermal imaging camera adapter that can be used as a fire detection device of Figure 1.

[0033] Figure 4 is an example diagram showing an example of detecting a fire in an electric vehicle through the fire detection device of Figure 1.

[0034] Figure 5 is an example diagram showing a state in which a fire alarm is notified to an administrator by linking with a management server after a fire is detected through the fire detection device of Figure 1.

[0035] Figure 6 is an example diagram showing a smartphone screen that has been notified of the fire alarm of Figure 5.

[0036] Figure 7 is an example diagram showing a smartphone screen that has been notified of the fire alarm of Figure 5 as another example.

[0037] Figure 8 is an illustrative diagram showing an example of a measure taken following the fire detection of Figure 5.

[0038] FIG. 9 is an example diagram showing an example of installing the fire detection device of FIG. 1 at a site.

[0039] FIG. 10 is an example diagram showing an example of the fire detection device of FIG. 9 operating in CCTV mode.

[0040] FIG. 11 is an example diagram showing an example of a fire detection device of FIG. 9 operating in a fire detection mode to detect a fire.

[0041] And, FIG. 12 is a flowchart illustrating a fire detection method according to one embodiment of the present invention.

[0042] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0043] Furthermore, the embodiments described herein will be explained with reference to cross-sectional views and / or schematic drawings, which are exemplary illustrations of the present invention. Accordingly, the form of the exemplary drawings may be modified due to manufacturing techniques and / or tolerances, etc. Additionally, in each drawing of the present invention, each component may be depicted slightly enlarged or reduced for convenience of explanation.

[0044] The on-device artificial intelligence-based fire detection device and method of the present invention are configured to enable even non-experts to easily and universally establish fire detection facilities wherever there is a concern for fire occurrence, and to operate in a manner that can derive continuous utility value.

[0045] FIG. 1 is a configuration diagram showing a fire detection device according to one embodiment of the present invention.

[0046] As illustrated in FIG. 1, an on-device artificial intelligence-based fire detection device (100) includes a camera (110), a memory (120), and at least one processor (130). The processor (130) drives the camera (110) to capture images of the surroundings, generates a result of comparing the captured images with dedicated fire detection data that has been pre-learned and stored in the memory (120), and if the generated result of the comparison analysis indicates a fire occurrence precursor state or a fire is occurring, it executes one or more pre-set fire alarm output means.

[0047] Here, the dedicated fire detection data is comparison data for determining whether the video captured by the camera (110) contains data related to a fire occurrence precursor state or a fire occurrence situation, and is data that has been specialized and learned in advance in relation to fire occurrence.

[0048] The fire detection device (100) can store dedicated fire detection data in memory (110) and analyze the received video through the operation of the processor (130), thereby enabling real-time detection of whether a fire has occurred at the point where the camera (110) is directed, based on on-device artificial intelligence.

[0049] This can be operated, for example, by downloading and installing an application that operates based on on-device artificial intelligence on a fire detection device (100), or by internally configuring software that operates based on on-device artificial intelligence.

[0050] More specifically, in the case of an on-device artificial intelligence application / software that distinguishes between the aforementioned fire occurrence precursor state and the aforementioned fire occurrence situation, it can first distinguish and secure a focus area where an object suspected of being a fire is detected through the YOLO (You Only Look Once) algorithm or the like for the captured area, and then distinguish between fire and non-fire (e.g., a cigarette burn in a normal situation, a normal fire while cooking on a kitchen stove, etc.) or between the fire occurrence precursor state and the fire occurrence state through a neural network algorithm such as CNN (Convolutional Neural Network) on the image of the focus area.

[0051] That is, if the surrounding environment where the fire detection device (100) is installed is a normal fire (e.g., a fire lit by a cooking appliance in a kitchen) or a temporary fire even if it is not normal (e.g., a cigarette butt lit near an electric vehicle in an outdoor parking lot), it may be classified as a non-fire and not generate fire occurrence information for a fire alarm.

[0052] In addition, since it is possible to distinguish between the fire precursor state and the fire in progress, it is possible to execute the action process for the fire precursor state and the action process for the fire in progress separately.

[0053] For example, the action process for fire precursors focuses on initial fire suppression by immediately notifying the manager in charge of the area of ​​fire precursors, thereby enabling rapid suppression of the initial fire.

[0054] As another example, the response process for an ongoing fire involves reporting the situation to specialized agencies, such as a nearby fire station, when the fire has already escalated to a level that is difficult to suppress in its early stages and is therefore difficult for the manager or other personnel responsible for the area to handle. This allows for the minimization of damage caused by the fire.

[0055] The aforementioned YOLO (You Only Look Once) algorithm can be replaced and applied with Faster R-CNN, ResNet (Residual Network), or equivalent focus feature extraction algorithms.

[0056] In addition, the application of neural network algorithms such as CNN (Convolutional Neural Network) can proceed with a process of augmenting data by adjusting the input images included in the captured images to fit the size of the CNN's input layer, and applying rotation, inversion, brightness and contrast adjustments to the adjusted images.

[0057] Subsequently, CNNs can be used to automatically extract fire-related features, such as flames and cigarette burns, contained within images, and an effective feature extraction network can be constructed by combining convolution layers, activation functions, and pooling layers.

[0058] Based on the features extracted in this way, a fully connected layer is constructed to precisely classify fire forms, such as flames and cigarette embers.

[0059] At this time, a specific loss function can be designed when distinguishing between a flame and a cigarette burn using a CNN, and weights reflecting characteristic differences such as between a flame and a cigarette burn can be added to the cross-entropy loss as shown in Equation 1 below.

[0060] [Formula 1]

[0061] L = -Σ(y_i * log(p_i)) +λ*Σ(w_i * |f_i - t_i|)

[0062] (Here, y_i is the actual label, p_i is the predicted probability, f_i is the feature extracted by the model, t_i is the target feature, w_i is the weight representing the importance of each feature, and λ is the total weight for the feature difference)

[0063] Additionally, to emphasize the morphological differences of fire, such as flames and cigarette embers, outline features can be incorporated into the loss function as shown in Equation 2 below.

[0064] [Equation 2]

[0065] L = L_class + α * L_edge

[0066] (Here, L_class is the classification loss, L_edge is the edge feature loss, and α is the weight)

[0067] Additionally, it is possible to design a loss function that reflects characteristics at various scales by considering the size difference between flames and cigarette embers, as shown in Equation 3 below.

[0068] L =Σ(w_s * L_s)

[0069] (Here, w_s is the weight for each scale, and L_s is the loss at each scale)

[0070]

[0071] The dedicated fire detection data described above can be updated and managed by being updated via a network or by being updated with dedicated fire detection data for updating received through an external interface.

[0072] Figure 2 is an example diagram showing an example of a fire detection device of Figure 1 installed at the site.

[0073] As illustrated in FIG. 2A) and B), the fire detection device (100) may be provided in a way that it is installed within a fire hydrant facing an area where fire is a concern (e.g., a parking area where electric vehicles are parked and an electric charging area, etc.) or attached to a wall or pillar located around an area where fire is a concern (e.g., a parking area where electric vehicles are parked and an electric charging area, etc.).

[0074] In addition, since the fire detection device (100) of the present invention can be installed by a non-expert, it can be easily installed not only in public areas such as parking areas and electric charging areas where electric vehicles are parked, but also in areas within an apartment where individual fires are a concern (e.g., living rooms within apartments where winter heating products are located, around each tent at a campsite, etc.). For a more detailed explanation, a fire detection device applied to the parking area and charging area of ​​an electric vehicle in an underground parking lot will be described as an example below.

[0075] Figure 3 is a configuration diagram showing the connection configuration between a smartphone and a thermal imaging camera adapter that can be used as a fire detection device of Figure 1.

[0076] As shown in FIG. 3, the fire detection device (100) can be provided via a smartphone (200), and can also be provided via a tablet PC or an electronic product equivalent thereto.

[0077] For example, in the case of a fire detection device (100) equipped via a smartphone (200), a thermal imaging camera function can be further added by connecting a thermal imaging camera adapter (300) to the smartphone (200) through the external interface of the smartphone (200).

[0078] In addition, the fire detection device (100) may be configured to include a thermal imaging camera function of the thermal imaging camera adapter (300), unlike the connection method described above.

[0079] Through this, the fire detection device (100) can additionally obtain source data to detect whether a fire has occurred based on thermal imaging in addition to detecting a fire by analyzing an image obtained through an image-based camera function, so even if there is no source to determine whether a fire has occurred within the image captured through the image-based camera function, it is possible to additionally determine whether a fire has occurred based on thermal imaging.

[0080] Additionally, the fire detection device (100) may also include an acoustic-based fire detection function in the same context as adding a thermal image-based fire detection function.

[0081] Here, the term "acoustic-based fire detection function" refers to detecting whether a fire has occurred through a unique sound generated when a fire occurs. To achieve this, it is desirable that the aforementioned dedicated fire detection data includes not only learning data related to image-based fire detection, but also at least one of learning data related to thermal image-based fire detection and learning data related to acoustic-based fire detection.

[0082] Figure 4 is an example diagram showing an example of detecting a fire in an electric vehicle through the fire detection device of Figure 1.

[0083] As illustrated in FIG. 4, the fire detection device (100) can detect a fire that occurs in a parking area and an electric charging area where an electric vehicle is parked, and can execute one or more fire alarm output means as a measure.

[0084] Here, one or more fire alarm output means may include a speaker of a fire detection device that outputs a fire alarm sound, a message transmission system that can notify of the occurrence of a fire through a network connection, and a management system that can automatically execute measures regarding the occurrence of a fire through a network connection.

[0085] Additionally, the fire detection device (100) can detect whether a fire has occurred in a single parking area and electric charging area where an electric vehicle is parked, as shown in FIG. 4.

[0086] Furthermore, as the fire detection device (100) is equipped with a camera orientation point to view multiple parking areas and electric charging areas, it is possible to detect fire in one or more of the multiple parking areas and electric charging areas through image analysis of the multiple parking areas and electric charging areas.

[0087] Accordingly, depending on the performance and function of the fire detection device (100), the fire detection area can be reduced or expanded.

[0088] Figure 5 is an example diagram showing a state in which a fire alarm is notified to an administrator by linking with a management server after a fire is detected through the fire detection device of Figure 1.

[0089] As shown in FIG. 5, the fire detection device (100) can be linked with a management server (not shown) through a network, and the management server (not shown) can manage notification and measures in the event of a fire.

[0090] For example, when a management server (not shown) receives fire occurrence information from a fire detection device (100), it may notify the manager in charge of the area of ​​the fire occurrence information to take action regarding the fire occurrence, or notify the fire department in charge of the area of ​​the fire occurrence information to enable dispatch to the fire source.

[0091] To this end, it is desirable that the aforementioned fire occurrence information include location information regarding the point where the fire occurred. This can be achieved by utilizing the location information of the smartphone (200) when the fire detection device (100) is equipped with a smartphone. Alternatively, when the fire detection device (100) is installed in the parking and charging areas of electric vehicles within an underground parking lot, it is also possible to receive location information corresponding to the point that the camera of the fire detection device is facing (e.g., identification information such as '2D12' indicating location 12 of area D on the second basement floor) and transmit the stored information.

[0092] Referring to FIGS. 7 and 8, fire occurrence information can be transmitted to a manager terminal, such as a smartphone, of a person (e.g., a manager) involved in receiving the fire occurrence information described above.

[0093] For example, the transmission of fire occurrence information may be processed via text message transmission, and the content of the transmitted message may vary depending on the type of fire occurrence information. In the case of a pre-fire condition, such as the occurrence of smoke, a corresponding notification may be provided, and in the case of an ongoing fire, a notification corresponding to the ongoing fire condition may be provided.

[0094] In addition, in cases where a fire is occurring or the fire detection results are unclear, it is also possible to transmit photos or videos of the fire area to enable the person receiving the fire information (e.g., an administrator) to verify and take action.

[0095] The transmission of text messages can support smoother execution of initial measures not only by sending a single message to a person involved in receiving fire occurrence information (e.g., an administrator), but also by sending messages to multiple people including the person involved in receiving fire occurrence information (e.g., an administrator).

[0096] Figure 8 is an illustrative diagram showing an example of a measure taken following the fire detection of Figure 5.

[0097] As shown in FIG. 8, through automatic notification from a management server (not shown), a fire is reported to a fire station located near the point where the fire occurred, allowing a fire truck to be dispatched immediately to the point where the fire occurred. Through this, it is possible to significantly reduce the time required for initial fire detection and response.

[0098] FIG. 9 is an example diagram showing an example of installing the fire detection device of FIG. 1 at a site.

[0099] As shown in FIG. 9, the underground parking lot can be configured with heterogeneous fire detection devices (100: a fire detection device with an image-based camera function, 200: a fire detection device with a thermal image-based fire detection function) for each parking area and charging area of ​​an electric vehicle.

[0100] For example, a fire detection device capable of detecting fires for each electric vehicle parked in each parking area and charging area may be configured as a fire detection device (100) with an image-based camera function, and a fire detection device (200) with a thermal image-based fire detection function capable of detecting fires in both electric vehicles and charging stations located within this area may be configured together.

[0101] FIG. 10 is an example diagram showing an example of the fire detection device of FIG. 9 operating in CCTV mode.

[0102] As illustrated in FIG. 10, each fire detection device (100, 200) operates in CCTV mode when no fire occurs, and can switch from CCTV mode to fire detection mode when an object appearing in the video moves and then stops.

[0103] In fire detection mode, fire detection functions can be performed by extracting fire-related key point data from the video and conducting a comparative analysis with dedicated fire detection data stored in advance.

[0104] Additionally, in CCTV mode, when an object appears in the video or moves, the captured video can be stored in memory or transmitted via a connected network. In this case, if the counterpart of the connected network is a management server (not shown), the video can be received, stored, and managed.

[0105] When storing the corresponding video in memory (110) in CCTV mode, it is also possible to manage the storage capacity by automatically deleting it after a predetermined period has elapsed.

[0106] FIG. 11 is an example diagram showing an example of a fire detection device of FIG. 9 operating in a fire detection mode to detect a fire.

[0107] As illustrated in FIG. 11, a fire detection device (100) with an image-based camera function can be provided for each parking area (parking areas 1 to 4), and can record video of the area in CCTV mode, and if an object such as an electric vehicle appears and moves, the recorded footage can be saved and processed.

[0108] Afterwards, when two electric vehicles are parked in the first parking area and the second parking area, the fire detection device corresponding to the first parking area and the second parking area, respectively, switches from CCTV mode to fire detection mode.

[0109] Among the fire detection devices operating in fire detection mode, the fire detection device (100, Q) corresponding to the second parking area detects a fire that occurred in the second parking area.

[0110] In addition, since the fire detection device (200) with a thermal imaging-based fire detection function operates based on thermal imaging, it is possible to detect fires in all of the first to fourth parking zones. Through this, when fire occurrence information is generated by the fire detection device (100, Q) corresponding to the second parking zone, it is possible to authenticate the authenticity of the fire occurrence information generated by the fire detection device (100, Q) corresponding to the second parking zone through the fire detection device (200) with the thermal imaging-based fire detection function.

[0111] As a specific operation, fire detection device (100, Q) corresponding to the second parking area generates fire occurrence information and transmits the generated fire occurrence information to a management server (not shown). Subsequently, the management server (not shown) requests thermal image-based fire detection information for the area from a verification fire detection device (200) having a thermal image-based fire detection function, which is a verification fire detection device located within the area related to the received fire occurrence information, and can authenticate the authenticity of the previously received fire occurrence information by determining the fire detection information received in response.

[0112] Additionally, the fire detection device (200) having a thermal imaging-based fire detection function may perform additional fire detection when a fire occurs in the second parking area and the fire detection device (100, Q) having an image-based camera function corresponding to the second parking area does not operate properly.

[0113] FIG. 12 is a flowchart illustrating a fire detection method according to one embodiment of the present invention.

[0114] As illustrated in FIG. 12, the on-device artificial intelligence-based fire detection method proceeds by driving a camera in the fire detection device (100) to capture images of the surroundings (S100).

[0115] When a frame of the video changes as a specific object (e.g., an electric vehicle) appears in the video captured in step S100 or moves after appearing (S102), the video can be stored for a certain period of time or transmitted through a network (S104).

[0116] In the video analysis of S102, if an object enters a stationary state, the system switches to a fire detection mode and performs an AI-based precise analysis on the video (S106). If the result of the precise video analysis of S106 indicates a fire occurrence precursor state or a fire is occurring, one or more pre-set fire alarm output means may be executed (S108 and S110).

[0117] The description of the detailed steps for each of the aforementioned steps and the description of the additional steps shall be based on the previously mentioned FIGS. 1 to 9 and the detailed description thereof.

[0118] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without changing its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.

[0119] In addition, the present invention is intended to provide an on-device artificial intelligence-based fire detection device and a method applied thereto that is easy to build and can substantially provide continuous utility value, and is an invention that has industrial applicability as it is not only sufficient for commercialization or business but also practically and clearly implementable.

Claims

1. Camera; Memory; and Includes at least one processor; and An on-device artificial intelligence-based fire detection device wherein the processor drives the camera to capture images of the surroundings, generates a result of comparing and analyzing the captured images with dedicated fire detection data that has been pre-learned and stored in the memory, and executes one or more pre-set fire alarm output means if the generated comparison and analysis result indicates a fire precursor state or a fire is occurring.

2. In Paragraph 1, An on-device artificial intelligence-based fire detection device that further includes the thermal imaging camera function of the thermal imaging camera adapter or is connected to the thermal imaging camera adapter through an external interface.

3. In Paragraph 1 or 2, An on-device artificial intelligence-based fire detection device equipped with a smartphone or tablet PC.

4. In Paragraph 1, An on-device artificial intelligence-based fire detection device in which the above-mentioned dedicated fire detection data is updated via a network or updated with dedicated fire detection data received via an external interface.

5. In Paragraph 1, An on-device artificial intelligence-based fire detection device that operates the recording of the above video in CCTV mode and switches the CCTV mode to a fire detection mode when an object appearing in the above video moves and then stops.

6. In Paragraph 5, An on-device artificial intelligence-based fire detection device that extracts fire-related key point data from the image in the above fire detection mode and performs comparative analysis with the above dedicated fire detection data.

7. In Paragraph 5, An on-device artificial intelligence-based fire detection device that stores the captured video in the memory or processes the transmission through a connected network when an object appears in the video or moves in the above CCTV mode.

8. In Paragraph 7, An on-device artificial intelligence-based fire detection device that automatically deletes the captured video stored in the memory after a predetermined period has elapsed.

9. In Paragraph 1, An on-device AI-based fire detection device that links with a management server via a network and integrates the management server's notification and action management in the event of a fire.

10. In Paragraph 9, An on-device artificial intelligence-based fire detection device that generates corresponding fire occurrence information when video analysis indicates a fire occurrence precursor state or a fire is in progress, and authenticates the authenticity of the generated fire occurrence information through the management server.

11. In a fire detection device, a step of driving a camera to capture images of the surroundings; A step of generating a comparison analysis result by comparing and analyzing the captured video with dedicated fire detection data that has been pre-trained and stored; and An on-device artificial intelligence-based fire detection method comprising the step of executing one or more preset fire alarm output means when the above comparative analysis result indicates a fire precursor state or a fire is occurring.

12. In Paragraph 11, An on-device artificial intelligence-based fire detection method comprising the step of operating the above video in CCTV mode and switching the CCTV mode to a fire detection mode when an object appearing in the above video moves and then stops.

Images