KR102653485B1 - Electronic apparatus for building fire detecting model and method thereof - Google Patents

Abstract

A tram device and method for constructing a fire detection system are disclosed. The electronic device according to the present disclosure divides, preprocesses, and labels fire image data on a frame basis, learns a fire detection model, obtains detection results, and further obtains detection results in the next frame through a target tracking model. can do.

Description

Electronic device and method for building a fire detection model for fire detection {ELECTRONIC APPARATUS FOR BUILDING FIRE DETECTING MODEL AND METHOD THEREOF}

This disclosure relates to an electronic device and method for constructing a fire detection model for fire detection by performing learning and testing of the model based on fire image data.

According to the 2019 Korea National Fire Data System's annual fire statistics report, the total number of deaths from 426,521 fires from 2010 to 2019 was 3,026 and the number of injuries was 18,792. In addition, property damage has reached KRW 4.4313 trillion, causing enormous damage to life and property.

Accordingly, efforts by many researchers are continuing to minimize environmental problems and financial losses caused by fire. Existing fire detection methods are divided into sensor-based fire detection methods and image processing-based fire detection methods. The sensor-based method has the disadvantage that system performance can be greatly limited depending on various factors in the surrounding environment. In addition, image processing-based methods are also difficult to apply to real situations due to empirical and experimental threshold settings, and may generate false alarms for objects similar to flames, so there are still many shortcomings.

The disclosed embodiments divide, preprocess, and label fire image data on a frame basis, and then identify a fire area represented by a flame through a fire detection model and target tracking module based on a convolutional neural network (CNN). It is for efficient detection and prediction.

A method of building a fire detection model for fire detection according to an embodiment disclosed includes the steps of collecting fire image data; Splitting the fire image data into frames; Preprocessing the divided data; Performing labeling by displaying a bounding box indicating a fire zone on the preprocessed data; and learning a fire detection model based on a convolutional neural network (CNN) based on the labeled data.

According to one embodiment, the preprocessing step includes deleting at least a portion of overlapping data among the divided data, deleting at least a portion of data whose resolution is less than a preset threshold value among the divided data, It may be characterized by including at least one of classifying the divided data into learning data, verification data, and test data.

According to one embodiment, the fire detection model includes a Focus network structure, a C3 structure, a Cross Stage Partial (CSP) network structure, a Feature Pyramid Network (FPN) structure, and a Path Aggregation Network (PAN); It may be characterized as including a Path Aggregation Network (Path Aggregation Network) structure.

According to one embodiment, the fire detection model may be characterized as including a bottleneck structure within the C3 structure.

According to one embodiment, the method of building a fire detection model for fire detection includes inputting test data into the fire detection model to obtain a detection result for the test data; And it may further include inputting the detection result into a target tracking module to obtain a prediction result in a frame next to the frame corresponding to the detection result and assigning an identifier.

According to one embodiment, the target tracking module may be characterized by tracking a plurality of targets based on the DeepSORT (Deep Simple Online and Real-time Tracking) algorithm.

According to one embodiment, the target tracking module includes a filtering module that outputs a prediction result in a frame next to the frame corresponding to the detection result based on the detection result; and a matching module that correlates the detection result and the prediction result and assigns an identifier to the detection result-prediction result pair.

According to one embodiment, the filtering module outputs the position at which a bounding box indicating a fire zone is most likely to appear in the next frame of the frame corresponding to the detection result based on a Kalman Filter. can do.

According to one embodiment, the matching module correlates the detection result and the prediction result based on a Hungarian algorithm, and determines the identifier based on the amount of position change between the detection result and the prediction result. It may be characterized by selecting a granting method.

According to one embodiment, when the amount of position change is less than a preset reference value, the matching module assigns an identifier based on a Mahalanobis distance value corresponding to the amount of position change, and the amount of position change is preset. If it is greater than the reference value, an identifier corresponding to the amount of position change may be assigned using a Deep Appearance Descriptor.

An electronic device for building a fire detection system according to an embodiment of the disclosure includes a transceiver, a memory for storing commands, and a processor, wherein the processor is connected to the transceiver and the memory to collect fire image data, Splitting the fire image data into frames, preprocessing the divided data, performing labeling by displaying a bounding box indicating a fire zone on the preprocessed data, based on the labeled data. Thus, a fire detection model based on a convolutional neural network (CNN) can be learned.

Specific details of other embodiments are included in the detailed description and drawings.

According to the present disclosure, an electronic device for building a fire detection system detects flames by automatically extracting the characteristics of the flame in the image from fire image data through preprocessing, labeling, and learning and testing through a deep learning model for the image. can do.

The effect of the invention is not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

1 is a schematic configuration diagram for explaining a fire detection system according to an embodiment.
Figure 2 is a flowchart illustrating a method of building a fire detection model for fire detection according to an embodiment.
Figure 3 is an exemplary diagram of the architecture of a fire detection model according to one embodiment.
Figure 4 is a flowchart illustrating a method of building a fire detection model for fire detection according to an additional embodiment.
Figure 5 is an exemplary diagram showing a testing process according to one embodiment.
FIG. 6 is a block diagram illustrating an electronic device for building a fire detection system according to an embodiment.

The terms used in the embodiments are general terms that are currently widely used as much as possible while considering the functions in the present disclosure, but this may vary depending on the intention or precedent of a person working in the art, the emergence of new technology, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meaning will be described in detail in the relevant description. Therefore, the terms used in this disclosure should be defined based on the meaning of the term and the overall content of this disclosure, rather than simply the name of the term.

When it is said that a part "includes" a certain element throughout the specification, this means that, unless specifically stated to the contrary, it does not exclude other elements but may further include other elements. In addition, terms such as "... unit" and "... module" used in the specification refer to a unit that processes at least one function or operation, which is implemented as hardware or software, or as a combination of hardware and software. It may be possible, and unlike the example shown, specific operations may not be clearly distinguished.

The expression “at least one of a, b, and c” used throughout the specification means ‘a alone’, ‘b alone’, ‘c alone’, ‘a and b’, ‘a and c’, ‘b and c ', or 'all of a, b, and c'.

The “terminal” mentioned below may be implemented as a computer or portable terminal that can connect to a server or other terminal through a network. Here, the computer includes, for example, a laptop, desktop, laptop, etc. equipped with a web browser, and the portable terminal is, for example, a wireless communication device that guarantees portability and mobility. , all types of communication-based terminals such as IMT (International Mobile Telecommunication), CDMA (Code Division Multiple Access), W-CDMA (W-Code Division Multiple Access), and LTE (Long Term Evolution), smartphones, tablet PCs, etc. It may include a handheld-based wireless communication device.

In the following description, “transmission,” “communication,” “transmission,” “reception,” “transmission,” “transmission,” “reception,” and similar terms refer to the direct transmission of signals, messages, or information from one component to another. It includes not only what is done, but also what is transmitted through other components.

In particular, “transmitting” or “transmitting” a signal, message or information as a component indicates the final destination of the signal, message or information and does not mean the direct destination. The same applies to “receiving” signals, messages or information. Additionally, in the present disclosure, “related” to two or more data or information means that if one data (or information) is acquired, at least part of other data (or information) can be obtained based on it.

Additionally, terms such as first and second may be used to describe various components, but the components should not be limited by the terms. The above terms may be used for the purpose of distinguishing one component from another component.

For example, a first component may be referred to as a second component, and similarly, the second component may also be referred to as the first component without departing from the scope of the present disclosure.

Below, with reference to the attached drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily practice them. However, the present disclosure may be implemented in many different forms and is not limited to the embodiments described herein.

In describing the embodiments, description of technical content that is well known in the technical field to which the present invention belongs and that is not directly related to the present invention will be omitted. This is to convey the gist of the present invention more clearly without obscuring it by omitting unnecessary explanation.

For the same reason, some components are exaggerated, omitted, or schematically shown in the accompanying drawings. Additionally, the size of each component does not entirely reflect its actual size. In each drawing, identical or corresponding components are assigned the same reference numbers.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are merely provided to ensure that the disclosure of the present invention is complete and to provide common knowledge in the technical field to which the present invention pertains. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

It will be understood that each block of the processing flow diagrams and combinations of the flow diagram diagrams may be performed by computer program instructions. These computer program instructions can be mounted on a processor of a general-purpose computer, special-purpose computer, or other programmable data processing equipment, so that the instructions performed through the processor of the computer or other programmable data processing equipment are described in the flow chart block(s). It creates the means to perform functions. These computer program instructions may also be stored in computer-usable or computer-readable memory that can be directed to a computer or other programmable data processing equipment to implement a function in a particular manner, so that the computer-usable or computer-readable memory It is also possible to produce manufactured items containing instruction means that perform the functions described in the flowchart block(s). Computer program instructions can also be mounted on a computer or other programmable data processing equipment, so that a series of operational steps are performed on the computer or other programmable data processing equipment to create a process that is executed by the computer, thereby generating a process that is executed by the computer or other programmable data processing equipment. Instructions that perform processing equipment may also provide steps for executing the functions described in the flow diagram block(s).

Additionally, each block may represent a module, segment, or portion of code that includes one or more executable instructions for executing specified logical function(s). Additionally, it should be noted that in some alternative execution examples it is possible for the functions mentioned in the blocks to occur out of order. For example, it is possible for two blocks shown in succession to be performed substantially at the same time, or it is possible for the blocks to be performed in reverse order depending on the corresponding function.

Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings.

1 is a schematic configuration diagram for explaining a fire detection system according to an embodiment.

As shown, the fire detection model building system 100 according to one embodiment includes an electronic device 110 and a network 120. The fire detection model building system 100 shown in FIG. 1 shows only configurations related to this embodiment. Accordingly, those skilled in the art can understand that other general-purpose components may be included in addition to the components shown in FIG. 1.

The electronic device 110 receives a still image such as a photo, a picture, or a video composed of a sum of still images in frame units as input, and learns a fire detection model, obtains a detection result through a fire detection model, or , refers to a device that obtains prediction results through a target tracking module.

According to one embodiment, the electronic device 110 may acquire fire image data to be processed from the outside through the wired or wireless network 120. More specifically, the fire image data received by the electronic device 110 may be an image captured from a photographing device represented by a camera, may be an image pre-stored in a cloud server, or may be an image pre-stored in a local database (DB). It may be possible. However, this is an example, and the image capturing means or storage location may be set in various ways depending on the embodiment. Additionally, the creation location and storage location of the image may be the same, but may differ depending on the embodiment.

According to one embodiment, network 120 includes a local area network (LAN), a wide area network (WAN), a value added network (VAN), a mobile radio communication network, It is a comprehensive data communication network that includes satellite communication networks and their mutual combinations, and allows each network constituent shown in Figure 1 to communicate smoothly with each other, and may include wired Internet, wireless Internet, and mobile wireless communication networks. You can. Wireless communications include, for example, wireless LAN (Wi-Fi), Bluetooth, Bluetooth low energy, ZigBee, WFD (Wi-Fi Direct), UWB (ultra wideband), and infrared communication (IrDA, infrared Data Association). ), NFC (Near Field Communication), etc., but are not limited thereto.

In relation to the above, it will be described in more detail through the drawings below.

Figure 2 is a flowchart illustrating a method of building a fire detection model for fire detection according to an embodiment.

Below, in FIG. 2, the method is described divided into a plurality of steps, but at least some of the steps are performed in a different order, performed in combination with other steps, omitted, performed divided into detailed steps, or not shown. One or more steps may be added and performed.

In step 210, the electronic device 110 may collect fire image data.

According to one embodiment, the electronic device 110 may collect a video including a plurality of frames or a single frame of fire photos. In addition, as described above with reference to FIG. 1, the electronic device 110 may receive fire image data directly from a photographing device, or may receive fire image data from an external cloud server or local DB. In the embodiment, Accordingly, fire image data may be collected by loading the fire image data stored in the storage space within the electronic device 110 into a memory within the electronic device 110 to perform operations to be described later. In this case, the imaging device, cloud server, local DB, etc. shown in FIG. 1 may mean a place where test data or execution data is generated or stored.

For example, fire video data may be classified based on the scale of occurrence, such as large-scale flames or small-scale flames, flames in buildings, flames in grasslands, flames in forests, and vehicles (cars, trucks). , motorcycle, etc.), it may be data classified based on the location of occurrence, such as a flame. Additionally, the data may be classified based on the time of occurrence, such as fires that occur during the day and fires that occur at night.

In step 220, the electronic device 110 may divide fire image data into frames.

In step 230, the electronic device 110 may preprocess the divided data.

According to one embodiment, the electronic device 110 may perform at least one of the following (1) to (3) as pre-processing on the divided data.

(1) Delete at least part of the duplicate data among the divided data

(2) Delete at least part of the data whose resolution is below a preset threshold among the divided data.

(3) Classify the divided data into learning data, validation data, and test data.

Specifically, for the preprocessing of (1), the electronic device 110 determines the similarity between the divided data based on the pixel value, brightness value, etc. between the data divided in frame units, and selects data whose similarity is greater than or equal to a preset standard value. You can also delete all but one.

Meanwhile, specifically, for preprocessing in (3), the electronic device 110 may classify more than half of the divided data as learning data, half of the remaining data as verification data, and half as test data. For example, the electronic device 110 may classify 70% of the divided data as training data, 15% as validation data, and 15% as test data.

According to one embodiment, the electronic device 110 may perform preprocessing to adjust the size of the divided data to a preset standard. For example, the electronic device 110 can collectively adjust the size of the divided data to 640*640 (640 pixels horizontally and 640 pixels vertically).

In step 240, the electronic device 110 may perform labeling by displaying a bounding box indicating a fire zone on the preprocessed data.

For example, the electronic device 110 may perform labeling by displaying a bounding box using CVAT (Computer Vision Annotation Tool).

In step 250, the electronic device 110 may learn a fire detection model based on a convolutional neural network (CNN) based on the labeled data.

According to one embodiment, the fire detection model to be learned by the electronic device 110 is a focus network structure, a C3 structure, a cross stage partial (CSP) network structure, and a feature pyramid network (FPN). ) structure and may include a Path Aggregation Network (PAN) structure.

Specifically, the C3 structure included in the fire detection model may include a bottleneck structure within the C3 structure.

According to one embodiment, the fire detection model to be learned by the electronic device 110 may include a network structure based on YOLOv5.

In this regard, this will be described in more detail later with reference to FIG. 3 below.

Figure 3 is an exemplary diagram of the architecture of a fire detection model according to one embodiment.

Referring to FIG. 3, a fire detection model according to an embodiment may be composed of a backbone, neck, and head parts.

First, the backbone part is a part that extracts a feature map from input data and may include a focus network structure, C3 structure, and CSP structure.

According to one embodiment, the focus network structure of the backbone portion includes a slice layer (slice), a concatenation layer (Concat), and a convolution module (Conv), and each input data is input to a plurality of slice layers, and each slice The output from the layer is input to the concatenated layer, the output from the concatenated layer is input to the convolution module, and the output from the convolution module is the output of the focus network structure.

Specifically, the convolution module may include a 2-dimensional convolution layer (Conv2d), a 2-dimensional batch normalization layer (BatchNorm2d), and a sigmoid activation function (SiLU).

According to one embodiment, the C3 structure of the backbone portion includes a convolution module, a concatenation layer, and a bottleneck structure, and the output that has gone through the convolution module and the bottleneck structure and the output that has gone through only the convolution module are concatenated through a concatenation layer, and the output that has gone through only the convolution module is concatenated. It has a structure in which the output is input to the convolution module. In particular, the bottleneck structure has a structure in which the output that has passed through two convolution modules and the input that has not passed through the convolution module are summed.

According to one embodiment, the CSP structure of the backbone portion includes a max pooling layer, a concatenation layer, and a convolution module, where input is input to a plurality of max pooling layers, and the output from each max pooling layer is It has a structure that is input to the convolution module through a concatenation layer.

Meanwhile, the neck part connects the backbone and the head and refines and reconstructs the feature map extracted by the backbone, and may include a convolution module, up-sampling layer (UP-sample), concatenation layer, and C3 structure.

As shown, the concatenation layer of the neck part performs the function of concatenating the output from the previous layer and the output from the previous structure, and the output of each convolution module placed at the end of the neck part constitutes the head part, which will be described later. It has a structure that becomes the input of a 2D convolution layer.

Meanwhile, the head part is a part that performs localization on the feature map extracted by the backbone. Specifically, it can perform class-specific probability prediction and bounding box creation for objects on the feature map.

As shown, the head part may include a plurality of two-dimensional convolutional layers, and the output of each two-dimensional convolutional layer is the coordinates at which the bounding box will be displayed, a matrix containing class-specific probabilities for objects on the feature map, etc. It may contain information about

Figure 4 is a flowchart illustrating a method of building a fire detection model for fire detection according to an additional embodiment.

Below, in FIG. 4, the method is divided into a plurality of steps, but at least some of the steps are performed in a different order, combined with other steps, omitted, divided into detailed steps, or not shown. One or more steps may be added and performed.

In step 210, the electronic device 110 may collect fire image data.

In step 220, the electronic device 110 may divide fire image data into frames.

In step 230, the electronic device 110 may preprocess the divided data.

In step 240, the electronic device 110 may perform labeling by displaying a bounding box indicating a fire zone on the preprocessed data.

In step 250, the electronic device 110 may learn a fire detection model based on a convolutional neural network (CNN) based on the labeled data.

In step 460, the electronic device 110 may input test data into a fire detection model to obtain a detection result for the test data.

In step 470, the electronic device 110 may input the detection result for the test data into the target tracking module, obtain a prediction result in the next frame of the frame corresponding to the detection result, and assign an identifier.

According to one embodiment, the target tracking module used by the electronic device 110 may provide a multi-target tracking function capable of tracking a plurality of targets based on the Deep Simple Online and Real-time Tracking (DeepSORT) algorithm.

According to one embodiment, the target tracking module includes a filtering module that outputs a prediction result in the next frame of the frame corresponding to the detection result based on the detection result of the test data; and a matching module that correlates the detection result and the prediction result of the target tracking module and assigns an identifier (descriptor) to the pair of detection result and prediction result.

Specifically, the filtering module within the target tracking module can output the location where the bounding box representing the fire zone is most likely to appear in the next frame of the frame corresponding to the detection result for the test data based on the Kalman Filter. there is.

Meanwhile, the matching module within the target tracking module correlates the detection results for the test data with the prediction results of the target tracking module based on the Hungarian algorithm, and based on the amount of position change between the detection results and the prediction results, You can choose how to assign an identifier.

More specifically, the matching module within the target tracking module assigns an identifier based on the Mahalanobis distance value corresponding to the position change amount when the amount of position change between the detection result and the prediction result is less than a preset reference value, If the amount of location change between the detection result and the prediction result is greater than a preset standard value, an identifier corresponding to the amount of location change can be assigned using a deep appearance identifier (Deep Appearance Descriptor).

According to one embodiment, the deep appearance identifier is a module that extracts the features of a detected object using a CNN structure and expresses them as a low-dimensional vector, and extracts the features of the object for each input image data in each frame unit. You can.

Specifically, the deep appearance identifier consists of 2 convolution layers, 1 Max pooling layer, 6 residual blocks, 1 dense layer, and 1 A batch normalization layer may be included.

Through this, the target tracking module applies the Re-identification (ReID) model to solve the ID switching problem (the model fails to track the correct target and assigns a different ID to the original target) while also implementing Matching Cascade logic. By applying it, the accuracy of target tracking can be improved.

Figure 5 is an exemplary diagram showing a testing process according to one embodiment.

According to one embodiment, the electronic device 110 may obtain a detection result for the image sequence by inputting an image sequence composed of a plurality of frame-unit images as test data to a YOLOv5-based fire detection module.

According to the arrow flow shown in FIG. 5, it may appear that test data is also used in the training process, but this is an expression of using part of the data constituting the existing data set as learning data and part as test data. Anyone skilled in the art can tell that this is the case.

Meanwhile, according to one embodiment, the electronic device 110 inputs the detection result for the video sequence to a target tracking module that provides a multi-object tracking function, obtains a prediction result in the next frame, and detects Identifiers can be assigned to results and prediction results.

FIG. 6 is a block diagram illustrating an electronic device for building a fire detection model building system according to an embodiment.

The electronic device 110 may include a transceiver 111, a processor 113, and a memory 115, according to one embodiment. The electronic device 110 can be connected to a photographing device, a cloud server, a local DB, and other external devices through the transceiver 111 and exchange data.

The processor 113 may perform at least one method described above with reference to FIGS. 1 to 5 . The memory 115 may store information for performing at least one method described above with reference to FIGS. 1 to 5 . Memory 115 may be volatile memory or non-volatile memory.

The processor 113 can control the electronic device 110 to execute programs and provide information. The code of the program executed by the processor 113 may be stored in the memory 115.

The processor 113 is connected to the transceiver 111 and the memory 115, collects fire image data, divides the fire image data into frames, preprocesses the divided data, and targets the preprocessed data to the fire area. Labeling is performed by displaying a bounding box representing , and a fire detection model based on a convolutional neural network (CNN) can be learned based on the labeled data.

Additionally, the electronic device 110 according to one embodiment may further include an interface that can provide information to the user.

The electronic device 110 shown in FIG. 6 shows only components related to this embodiment. Accordingly, those skilled in the art can understand that other general-purpose components may be included in addition to the components shown in FIG. 6.

Devices according to the above-described embodiments include a processor, memory for storing and executing program data, permanent storage such as a disk drive, a communication port for communicating with an external device, a touch panel, keys, buttons, etc. It may include the same user interface device, etc. Methods implemented as software modules or algorithms may be stored on a computer-readable recording medium as computer-readable codes or program instructions executable on the processor. Here, computer-readable recording media include magnetic storage media (e.g., ROM (read-only memory), RAM (random-access memory), floppy disk, hard disk, etc.) and optical read media (e.g., CD-ROM). ), DVD (Digital Versatile Disc), etc. The computer-readable recording medium is distributed among computer systems connected to a network, so that computer-readable code can be stored and executed in a distributed manner. The media may be readable by a computer, stored in memory, and executed by a processor.

This embodiment can be represented by functional block configurations and various processing steps. These functional blocks may be implemented in various numbers of hardware or/and software configurations that execute specific functions. For example, embodiments include integrated circuit configurations such as memory, processing, logic, look-up tables, etc. that can execute various functions under the control of one or more microprocessors or other control devices. can be hired. Similar to how the components can be implemented as software programming or software elements, the present embodiments include various algorithms implemented as combinations of data structures, processes, routines or other programming constructs, such as C, C++, Java ( It can be implemented in a programming or scripting language such as Java), assembler, etc. Functional aspects may be implemented as algorithms running on one or more processors. Additionally, this embodiment may employ conventional technologies for electronic environment setting, signal processing, message processing, and/or data processing. Terms such as “mechanism,” “element,” “means,” and “configuration” may be used broadly and are not limited to mechanical and physical configurations. The term may include the meaning of a series of software routines in connection with a processor, etc.

The above-described embodiments are merely examples and other embodiments may be implemented within the scope of the claims described below.

Claims (12)

An electronic device collecting fire image data;
The electronic device dividing the fire image data into frames;
Preprocessing, by the electronic device, the divided data;
performing labeling, by the electronic device, on the preprocessed data by displaying a bounding box indicating a fire zone; and
Comprising the step of the electronic device learning a fire detection model based on a convolutional neural network (CNN) based on the labeled data,
In the preprocessing step,
The electronic device determines the similarity between the divided data based on the pixel value and brightness value between the divided data, deletes all but one data whose similarity is higher than a preset standard value, and resolves the resolution of the divided data. A method of building a fire detection model for fire detection, characterized in that deleting at least a portion of data that is below a preset threshold.
delete In claim 1,
The fire detection model is,
Characterized by including a Focus network structure, a C3 structure, a Cross Stage Partial (CSP) network structure, a Feature Pyramid Network (FPN) structure, and a Path Aggregation Network (PAN) structure. ,How to build a fire detection model for fire detection. In claim 3,
The fire detection model is,
A method of building a fire detection model for fire detection, characterized in that it includes a bottleneck structure in the C3 structure. In claim 1,
The electronic device inputting test data into the learned fire detection model to obtain a detection result for the test data; and
A method of building a fire detection model for fire detection, further comprising the step of the electronic device inputting the detection result to a target tracking module to obtain a prediction result in the next frame of the frame corresponding to the detection result and assigning an identifier. . In claim 5,
The target tracking module,
A method of building a fire detection model for fire detection, characterized by tracking a plurality of targets based on the DeepSORT (Deep Simple Online and Real-time Tracking) algorithm. In claim 6,
The target tracking module,
a filtering module that outputs a prediction result in a next frame of the frame corresponding to the detection result based on the detection result; and
A method of building a fire detection model for fire detection, comprising a matching module that correlates the detection result and the prediction result and assigns an identifier to the detection result-prediction result pair. In claim 7,
The filtering module is,
A method of building a fire detection model for fire detection, characterized in that outputting the position at which a bounding box representing a fire zone is most likely to appear in the next frame of the frame corresponding to the detection result based on Kalman Filter. . In claim 7,
The matching module is,
Characterized in selecting a method of correlating the detection result and the prediction result based on a Hungarian algorithm and assigning the identifier based on the amount of position change between the detection result and the prediction result. How to build a fire detection model for fire detection. In claim 9,
The matching module is,
If the amount of position change is less than a preset reference value, an identifier is assigned based on the Mahalanobis distance value corresponding to the amount of position change,
A method of building a fire detection model for fire detection, characterized in that, when the amount of position change is greater than a preset reference value, an identifier corresponding to the amount of position change is assigned using a Deep Appearance Descriptor. A computer-readable non-transitory recording medium recording a program for executing the method of any one of claims 1, 3 to 10 on a computer.
delete

Images