KR20230168242A - Method for identifying shape of target object and artificial intelligence based gas leak detection device performing the same - Google Patents

Abstract

The present disclosure relates to an electronic device for detecting gas leaks based on artificial intelligence and a method for the electronic device to identify the shape of a target object. According to one embodiment, a method of identifying the shape of a target object by an electronic device for detecting an artificial intelligence-based gas leak includes receiving first type images of a space to be monitored from at least one first type of camera connected to the electronic device. acquiring; Obtaining second type images of the space to be monitored from at least one second type camera connected to the electronic device; And when the first type images and the second type images are input, an object recognition model that identifies whether a target object in the first type images and the second type images is detected, the first type image identifying whether the target object is detected in the first type images and the second type images by inputting the first type images and the second type images; Includes, and the target object may include at least one of a smoke object, a gas object, or a flame object.

Description

Method for identifying the shape of a target object and an artificial intelligence-based gas leak detection device that performs the same {METHOD FOR IDENTIFYING SHAPE OF TARGET OBJECT AND ARTIFICIAL INTELLIGENCE BASED GAS LEAK DETECTION DEVICE PERFORMING THE SAME}

The present disclosure relates to an electronic device for artificial intelligence-based gas leak detection and a method for the electronic device to identify the shape of a target object such as smoke or gas. More specifically, it is about how an electronic device for artificial intelligence-based gas leak detection shapes smoke and gas objects to image and detect artificial intelligence-based harmful gases in real time.

Among the various types of gases used in industrial sites and everyday life, detection of gas leaks that are lethal or explosive to the human body is very important, and real-time monitoring is necessary to prevent them from escalating into secondary safety accidents. In particular, gases such as methane generated from coal storage and ammonia generated from ammonia storage tanks or gas pipes are difficult to visually confirm, so it is necessary to prevent safety accidents by detecting leaks of these gases in advance.

However, in the case of conventional gas alarm and detection technology, some technologies utilizing gas sensors and video cameras are disclosed, but there is a limit to the accuracy of gas detection and tracking, and gas leaks within the site are monitored by linking with an alarm device. In many cases, manpower is used to detect the origin of the leak, which has limitations in responding to accidents. Therefore, there is a need to develop technology to effectively detect and manage preset target objects such as gas, smoke, and flame.

Korean Patent No. 10-2355884

According to one embodiment, an artificial intelligence-based harmful gas real-time imaging detection system and a method of operating the same may be provided.

More specifically, an artificial intelligence-based gas leak detecting electronic device that can be used in the system and a method for the artificial intelligence-based gas leak detecting electronic device to identify the shape of a target object may be provided.

As a technical means for achieving the above-described technical problem, a method of identifying the shape of a target object by an artificial intelligence-based gas leak detection electronic device according to an embodiment includes at least one first type of camera connected to the electronic device. Obtaining first type images of a space to be monitored from; Obtaining second type images of the space to be monitored from at least one second type camera connected to the electronic device; And when the first type images and the second type images are input, an object recognition model that identifies whether a target object in the first type images and the second type images is detected, the first type image identifying whether the target object is detected in the first type images and the second type images by inputting the first type images and the second type images; may include, and the target object may include at least one of a smoke object, a gas object, or a flame object.

As a technical means for achieving the above-mentioned technical problem, according to another embodiment, an electronic device for identifying the shape of a target object for detecting an artificial intelligence-based gas leak, comprising: a network interface; A memory that stores one or more instructions; and at least one processor executing the one or more instructions; Includes, wherein the at least one processor executes the one or more instructions to obtain first type images of the space to be monitored from at least one first type camera connected to the electronic device, and the at least one processor connected to the electronic device Obtain second type images for the monitoring target space from at least one second type camera, and when the first type images and the second type images are input, the first type images and the second type images The target object in the first type images and the second type images by inputting the first type images and the second type images into an object recognition model that identifies whether the target object in the images is detected. An electronic device may be provided, wherein the target object includes at least one of a smoke object, a gas object, and a flame object.

In addition, as a technical ability to achieve the above-described technical problem, according to another embodiment, in a method for an artificial intelligence-based gas leak detection electronic device to identify the shape of a target object, at least one connected to the electronic device Obtaining first type images of a space to be monitored from a first type of camera; Obtaining second type images of the space to be monitored from at least one second type camera connected to the electronic device; And when the first type images and the second type images are input, an object recognition model that identifies whether a target object in the first type images and the second type images is detected, the first type image identifying whether the target object is detected in the first type images and the second type images by inputting the first type images and the second type images; A computer-readable recording medium storing a program for performing the method may be provided, wherein the target object includes at least one of a smoke object, a gas object, and a flame object.

According to one embodiment, the shape of a gas object can be accurately identified.

According to one embodiment, the shape information of the gas object can be identified and the leak origin of the gas object can be effectively detected based on the identified shape information.

1 is a diagram schematically illustrating the operation process of an artificial intelligence-based gas leak detection system including an electronic device for detecting an artificial intelligence-based gas leak and an electronic device for detecting the artificial intelligence-based gas leak according to an embodiment. am.
Figure 2 is a diagram showing the operation process of an artificial intelligence-based gas leak detection system according to another embodiment.
Figure 3 is a flowchart showing a method of operating an artificial intelligence-based gas leak detection system according to an embodiment.
Figure 4 is a flowchart of a method for identifying whether an artificial intelligence-based gas leak detection electronic device detects a target object according to an embodiment.
FIG. 5 is a diagram illustrating a process in which an artificial intelligence-based gas leak detecting electronic device extracts shape information of a target object identified according to a frequency pattern, according to an embodiment.
FIG. 6 is a diagram illustrating a process in which an artificial intelligence-based gas leak detecting electronic device extracts shape information of a target object identified according to a frequency pattern according to an embodiment.
Figure 7 is a flowchart of a specific method in which an artificial intelligence-based gas leak detection electronic device extracts shape information of a target object according to an embodiment.
FIG. 8 is a diagram illustrating a process in which an artificial intelligence-based gas leak detection electronic device identifies the shape of a target object and tracks it, according to an embodiment.
Figure 9 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device extracts shape information of a target object according to another embodiment.
Figure 10 is a flowchart of a specific method in which an artificial intelligence-based gas leak detection electronic device determines the leak origin and flow vector of a target object according to an embodiment.
Figure 11 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device generates a composite image for gas leak detection according to an embodiment.
FIG. 12 is a diagram illustrating a leak image generated by an electronic device for detecting an artificial intelligence-based gas leak and a process of combining the leak image into a panoramic image, according to an embodiment.
Figure 13 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device generates a leak image according to an embodiment.
Figure 14 is a flowchart of a specific method in which an artificial intelligence-based gas leak detecting electronic device generates a leak image according to an embodiment.
Figure 15 is a flowchart of a method in which an electronic device for detecting gas leaks based on artificial intelligence generates a composite image according to an embodiment.
Figure 16 is a flowchart of a method by which an artificial intelligence-based gas leak detection electronic device determines the coordinates of a leak image for a panoramic image, according to an embodiment.
FIG. 17 is a diagram illustrating a process in which an electronic device for detecting an artificial intelligence-based gas leak according to an embodiment converts the coordinates of the origin of the leak based on laser measurement values and camera pan-tilt information.
Figure 18 is a diagram showing the operation process of an artificial intelligence-based gas leak detection system according to another embodiment.
Figure 19 is a flowchart of a method in which an artificial intelligence-based gas leak detecting electronic device detects a gas leak using an artificial intelligence model according to an embodiment.
FIG. 20 is a diagram illustrating a process in which an electronic device for detecting an artificial intelligence-based gas leak learns an object recognition model and utilizes the learned object recognition model, according to an embodiment.
Figure 21 is a flowchart showing a process of learning a likelihood ratio test model and a support vector model by an electronic device for detecting gas leaks based on artificial intelligence according to an embodiment.
FIG. 22 is a diagram illustrating a process in which an electronic device for detecting gas leaks based on artificial intelligence learns a likelihood ratio test model and a support vector model according to an embodiment.
FIG. 23 is a diagram illustrating a process in which an artificial intelligence-based gas leak detection electronic device detects a gas leak using a plurality of software modules according to another embodiment.
FIG. 24 is a diagram illustrating a process for detecting a gas leak by linking a measuring device, an artificial intelligence-based gas leak detection server, an alarm/control server, and a control device with each other, according to another embodiment.
FIG. 25 is a diagram illustrating a process for detecting a gas leak by linking an electronic device, an alarm device, and a server with each other, according to another embodiment.
Figure 26 is a block diagram of an electronic device according to an embodiment.
Figure 27 is a block diagram of an electronic device according to another embodiment.
Figure 28 is a diagram for explaining the device configuration of a gas leak detection system according to another embodiment.
Figure 29 is a block diagram of a server according to one embodiment.
Figure 30 is a diagram illustrating a user interface screen provided by an electronic device for detecting an artificial intelligence-based gas leak according to an embodiment.
Figure 31 is a diagram illustrating a user interface screen provided by an electronic device for detecting an artificial intelligence-based gas leak according to another embodiment.

Terms used in this specification will be briefly described, and the present disclosure will be described in detail.

The terms used in the present disclosure have selected 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 precedents of those skilled in the art, the emergence of new technologies, 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 description of the relevant invention. 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 may be implemented as hardware or software, or as a combination of hardware and software. .

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 order to clearly explain the present disclosure in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

1 is a diagram schematically illustrating the operation process of an artificial intelligence-based gas leak detection system including an electronic device for detecting an artificial intelligence-based gas leak and an electronic device for detecting the artificial intelligence-based gas leak according to an embodiment. am.

According to one embodiment, the electronic device 1000 (hereinafter referred to as 'electronic device') for detecting artificial intelligence-based gas leaks is a monitoring target from a camera installation set 102 including at least one type of camera and a laser measuring device. By acquiring image information 104 about the space and inputting the acquired images into an artificial intelligence model, it can be determined whether there is a target object including at least one of smoke, gas, or flame in the acquired images. According to one embodiment, the electronic device 1000 inputs a plurality of images according to at least one type from at least one type of camera to an artificial intelligence model, and based on the output value of the artificial intelligence model, displays a gas object, smoke, etc. When a target object including at least one of an object or a flame object is detected, shape information of the target object may be extracted, and the target object may be modeled and tracked based on the extracted shape information.

In this specification, the target object detected by the artificial intelligence-based gas leak detection system 10 or the electronic device 1000 is a gas object, or smoke in which gas particles and air particles are integrated as the gas flows in a gaseous state in the air. It may include at least one of an object and a flame object generated as gas burns. A gas object, a smoke object, or a flame object according to an embodiment may be detected through visible light or electromagnetic waves (eg, infrared, etc.) in a wavelength range different from the visible light.

According to one embodiment, the image information that the electronic device 1000 acquires from the camera installation set 102 is visible light obtained by photographing a coal storage site and a work site where ammonia pipes are installed using at least one of an EO camera or an OGI camera. These may be video or infrared images. Additionally, the visible light images and infrared images may include a plurality of visible light frame images and a plurality of infrared frame images having preset frame intervals.

According to one embodiment, when the electronic device 1000 analyzes the image information 104 and identifies that a target object is detected, the electronic device 1000 sends a control signal that causes the warning device 132 to output visual or auditory content. By transmitting, the fact that a dangerous situation such as a gas leak has occurred can be effectively communicated to workers. According to one embodiment, the electronic device 1000 may be connected to a control terminal 134 installed adjacent to the monitoring target space. According to one embodiment, the electronic device 1000 transmits image information 104 and laser measurement values 108 to the control terminal, or the electronic device 10000 analyzes the image information 104 using an artificial intelligence model. Results (e.g., whether the target object is detected, shape information about the target object, origin of leak, gas leak, information about the type of gas) may be transmitted to the control terminal 134.

According to one embodiment, the control terminal 134 may be directly connected to the camera installation set 102, and directly stores image information obtained from the electronic device 1000 or the camera installation set 102, as well as the acquired image. Information can be printed. In addition, the control terminal 134 includes components corresponding to the memory, processor, and network interface included in the electronic device 1000, and inputs the acquired image information into an artificial intelligence model for gas leak detection, so that the target object It is possible to determine whether an event such as a gas leak is detected, shape information of the target object, and when an event such as target object detection occurs, a control signal for output of warning content is transmitted directly to the alarm device 132. Alternatively, information about the target object may be directly output through a display device (eg, a monitor for a control terminal). According to one embodiment, the control terminal 134 may include an artificial intelligence-based gas imaging server, an alarm/control server, and a cooling system, but is not limited thereto, and is used for efficient processing of assets requiring calculation within the work site. It may further include MEC (Multi Access Edge Computing) servers. Additionally, the control terminal 134 may be connected in real time to various external devices through a multi-Ethernet device.

According to one embodiment, the control server 140 is connected to at least one of the camera installation set 102, the electronic device 1000, the control terminal 134, or the alarm device 132 within the gas leak detection system 10. You can control the behavior of components. For example, the control server 140 receives image information 104 about the space to be monitored from the electronic device 1000, information about the target object according to target object detection (e.g., type, leak origin, and actual camera installation set). physical location information) and a composite image can be acquired, and the acquired image information, information on the target object, and composite image can be output on the display. The control server 140 can facilitate management of the space to be monitored by integrated management of the collected data.

In addition, according to one embodiment, the control server 140 outputs image information about the space to be monitored, synthetic images, and information about the target object, and displays a user interface for detecting and managing gas leaks on the screen. Can be printed. The control server 140 controls the operation of the devices in the gas leak detection system 10 based on the controllers' user input on the output user interface, so that monitoring of the origin of the gas leak can be continued in the event of a gas leak. there is.

In addition, according to one embodiment, the electronic device 1000 determines the coordinates of the panoramic image of the leakage image generated for the tracked target object based on the laser measurement value and the pan-tilt information of the camera installation set 102. And, based on the determined coordinates, a composite image is generated by combining the leakage image of the target object with the panoramic image, and information on the original image information, leak origin, coordinate information, leakage image, and composite image is sent to the control terminal 134 or By transmitting to at least one of the control servers 140, the controller can easily monitor the leaked target gas.

In addition, according to one embodiment, as a result of analyzing the image information, the electronic device 1000 provides not only the location on the image of the panoramic image of the leak image of the target object, but also location information about the physical space of the actual leak origin. may be further transmitted to the control terminal 134 and the control server 140. For example, the electronic device 1000 acquires the laser measurement value 108 from the laser measuring device of the camera installation set 102, and sets the camera installation set based on the obtained laser measurement value and predetermined position information of the camera installation set. Identify the physical distance where the actual target object occurs from (102) and the location in physical space where the actual target object occurs, and provide information on the identified physical distance and location to the control server 140 to enable the controller to detect the target object. You can accurately recognize the location of the leak.

In addition, according to one embodiment, the electronic device 1000 for detecting an artificial intelligence-based gas leak analyzes image information obtained from the camera installation set 102 by inputting it into an artificial intelligence model, and identifies that a target object is detected. If so, it is determined that a predetermined event has occurred, and a control signal is transmitted to the warning device 132 to output visual or auditory warning contents.

According to one embodiment, the electronic device 1000 may store software modules necessary for gas leak detection (e.g., one or more instruction sets for gas leak detection), and the processor 120 may store the software modules required for gas leak detection in the memory 122. The gas leak detection process can be performed by executing one or more stored instructions. Additionally, the network interface 126 may perform a function of obtaining or transmitting data from other electronic devices connected to the electronic device 1000.

According to one embodiment, the gas leak detection system 10 shown in FIG. 1 includes a camera installation set 102, an electronic device 1000, an alarm device 132, a control terminal 134, and a control server 140. It may include, and may further include other devices for detecting gas leaks. Additionally, devices in the gas leak detection system 10 according to one embodiment may be connected to each other through at least one network. When transmitting data between devices in the gas leak detection system 10 according to an embodiment, data may be transmitted according to a preset data loss rate (e.g., less than 5%) and a preset data transmission period (e.g., less than 1 second). , but is not limited to this.

According to one embodiment, the at least one network is a local area network (LAN), a wide area network (WAN), a value added network (VAN), a mobile radio communication network, and a satellite. It may include a communication network, a comprehensive data communication network that allows each network constituent to communicate, and their mutual combination.

The gas leak detection system 10 and the electronic device 1000 according to the present disclosure may be used in a wide range of coal storage monitoring systems, or may be used in a monitoring system in a space where a large amount of ammonia storage tank/gas pipe is installed, but are not limited thereto. , can be used in work sites that require tracking and monitoring of target objects including at least one of gas objects, smoke objects, or flame objects.

Figure 2 is a diagram showing the operation process of an artificial intelligence-based gas leak detection system according to another embodiment.

The operation process of the artificial intelligence-based gas leak detection system 20 according to another embodiment will be described with reference to FIG. 2. According to one embodiment, the artificial intelligence-based gas leak detection system 20 includes a camera installation set 210, an AI gas image detection server 222, an alarm/control server 224, a gas leak alarm and warning device 230, and It may include a control terminal 228 and a situation room video system 242. According to one embodiment, the camera installation set 210 may include a gas imaging camera 212 (eg, Optical Gas Imaging, OGI camera), an EO camera (eg, Electro-Optical, EO), and a laser measuring device.

According to one embodiment, the gas imaging camera 212 electronically senses infrared radiant energy emitted from an object, thereby generating an infrared image including images of invisible gas objects or a plurality of images at preset frame intervals. It may be a camera that creates. Additionally, according to one embodiment, the EO camera may be a camera that generates a visible light image or a visible light image including a plurality of visible light images at preset frame intervals by sensing visible light reflected from an object through an image sensor. there is.

According to one embodiment, the laser measurer 214 emits a laser and measures the time when the emitted laser is reflected by the target object to determine the distance value from the laser measurer 214 or the camera installation set 210 to the target object. It is possible to obtain laser measurement values that represent The camera installation set 210 may transmit infrared images, visible light images, and laser measurement values to at least one device in the gas leak detection system 10. According to one embodiment, the shooting angle of view and shooting cycle of the gas imaging camera 212 and the EO camera 216 may be set differently, and the gas imaging camera 212 and the EO camera 216 may capture infrared images and visible images. Operating conditions for transmitting light images may be set in advance according to user input conditions.

According to one embodiment, the AI gas image detection server 222 detects a target object appearing in images acquired from the camera installation set 210, identifies the origin of leakage of the detected target object, and detects leakage of the target object. The visualization function can be performed by combining images into a panoramic image. According to one embodiment, the alarm/control server 224 may transmit a gas leak alarm and an alarm signal to the alarm device when an event occurring due to detection of a target object is detected. Additionally, the alarm/control server 224 may store image information acquired from the camera installation set 210, leaked images and synthesized images regarding the target object.

In addition, the alarm/control server 224 has a search function for stored images based on user input, an environment setting function for setting the device operation environment within the gas leak detection system 20, and a function for setting the device operation environment within the gas leak detection system 20. An operation control function may be provided to control the operation of devices. According to one embodiment, the alarm/control server 224 may transmit and receive event occurrence signal data to the alarm device or situation room video system 242 through TCP/IP.

Since the operation of the control terminal 228 may correspond to the operation of the control terminal 134 shown in FIG. 1, detailed description will be omitted. As shown in FIG. 2 , when a target object is detected, the situation room imaging system 242 may output a composite image generated by combining a leaked image of the target object with a panoramic image. Since the operation of the situation room video system 242 may correspond to the operation of the control server 140 shown in FIG. 1, detailed description will be omitted.

Figure 3 is a flowchart showing a method of operating an artificial intelligence-based gas leak detection system according to an embodiment.

Referring to FIG. 3, a method of operating an artificial intelligence-based gas leak detection system according to an embodiment will be described. Of course, the operation of the artificial intelligence-based gas leak detection system shown in FIG. 3 may also be performed by the electronic device 1000 shown in FIG. 1.

At S302, the gas leak detection system 10 may acquire an EO image and an OGI image. For example, the gas leak detection system 10 may acquire a plurality of EO images and OGI images from at least one EO camera and at least one OGI camera. At S304, the gas leak detection system 10 may input the images acquired at S304 into the object recognition model 304. For example, the gas leak detection system 10 may input EO images acquired from at least one EO camera and OGI images acquired from at least one OGI camera into an object recognition model.

Although not shown in FIG. 3, according to one embodiment, the gas leak detection system 10 acquires an EO image and an OGI image, and performs binarization, resizing, normalization, and data augmentation on the acquired EO image and OGI image. Preprocessing processes such as (Data Augmentation) can be performed. The gas leak detection system 10 can improve object recognition accuracy by inputting preprocessed EO images and OGI images into an object recognition model.

In S306, the gas leak detection system 10 may identify whether a target object is detected from the images based on the output value of the object recognition model in S302. According to one embodiment, when camera images are input, the object recognition model 304 is configured in advance to output a target object related to at least one of a gas object, a smoke object, or a flame object, or an object candidate area including at least a portion of the target object. It may be an artificial intelligence model that is being learned. According to one embodiment, the object recognition model may be a neural network model in which a plurality of weights are modified and updated so that loss or cost values are reduced or minimized.

For example, the object recognition model is learned based on images or image learning data for gas objects, flame objects, or smoke objects, such that object candidate regions at least partially include gas objects, flame objects, or smoke objects in the images or images. In addition to outputting, pixel information about the target object in the object candidate area and coordinate information on the image of the object candidate area may be arranged to be output together. For example, the object recognition model 304 may include a Convolutional Neural Network (CNN), a Deep Neural Network (DNN), a Recurrent Neural Network (RNN), a Restricted Boltzmann Machine (RBM), a Deep Belief Network (DBN), and a Bidirectional Recurrent Neural Network (BRDNN). It may be, but is not limited to, Deep Neural Network) or Deep Q-Networks. Additionally, the object recognition model 304 determines a probability value that a predetermined object candidate area corresponds to a target object based on pixel values in EO images or OGI images, and determines a predetermined priority value based on the determined probability value. It may be an object computer vision model such as YOLO, R-CNN, or SSD that is learned to output an object candidate area according to .

At S308, when the gas leak detection system 10 identifies that the target object is not detected from the EO images and OGI images using an object recognition model, the OGI image may be frequency converted to generate converted image information. . For example, when the gas leak detection system 10 primarily identifies that the target object is not detected using an object recognition model, the transformed image in the frequency domain is obtained by applying DFT discrete Fourier transform to the OGI image, which is an infrared image. Information can be generated. According to one embodiment, the converted image information may include a frequency pattern including magnitude and phase information for each frequency.

In S310, the gas leak detection system 10 inputs the converted image information into the likelihood ratio test model 310, which identifies whether a frequency pattern according to the target object exists when the converted image information is input, thereby forming a frequency pattern according to the target object. can be analyzed. In S312, the gas leak detection system 10 applies the likelihood ratio test model 310 to the converted image information and determines whether a frequency pattern according to the target object is identified based on the output value of the likelihood ratio test model 310. can be decided.

In S314, if the frequency pattern according to the target object is not identified from the converted image information as a result of converting the likelihood ratio test model, the gas leak detection system 10 uses the SVM model 314 (e.g., Support Vector Machine model) ) is applied to the converted image information to reanalyze the frequency pattern according to the target object in the converted image information. In S316, the gas leak detection system 10 may identify whether there is a frequency pattern according to the target object in the converted image information based on the output value of the SVM model.

In S322, if the gas leak detection system 10 determines that the target object is detected as a result of inputting the EO images and OGI images to the object recognition model in S306, EO images in which the target object is detected or the target object is detected Target objects in OGI images can be separated from the background. For example, the gas leak detection system 10 may generate masking data based on detected target object information and separate target object areas by applying the generated masking data to EO images and OGI images.

At S324, the gas leak detection system 10 may detect the shape of the object from the separated target object area. For example, the gas leak detection system 10 may output outline information, center point information, and direction vector within the target object area as object shape information of the target. In S326, the gas leak detection system 10 may determine a leak origin regarding the location where the target object was created and a flow vector regarding the flow direction and speed of the target object based on shape information appearing in the target object areas.

In S328, the gas leak detection system 10 may identify distance coordinates and spherical coordinates based on the leak origin and flow vector, and convert location information determined according to the identified coordinates. Specifically, the gas leak detection system 10 acquires laser measurement values by transmitting and receiving a laser at a location where a target object is determined to be identified in the image based on the leak origin and flow vector, and based on the obtained laser measurement values, the camera The distance from the installation set to the origin of the leak can be obtained. Additionally, the gas leak detection system 10 can acquire pan-tilt information of a camera installation set that monitors the origin of the leak.

The gas leak detection system 10 according to the present disclosure can determine the first coordinate value in a spherical coordinate system based on the distance from the camera installation set to the leak origin and camera pan/tilt information. The gas leak detection system 10 generates a third coordinate value by converting the first coordinate value in the determined spherical coordinate system into a second coordinate value in the Cartesian coordinate system and reflecting the camera distortion value in the second coordinate value converted to the Cartesian coordinate system. can do. The gas leak detection system 10 may identify the reference position of the panoramic image of the leak image where the target object occurs based on the third coordinate value, and synthesize the leak image into the panoramic image according to the identified reference position.

In S330, the gas leak detection system 10 may generate a composite image by combining the leak image with the panoramic image based on the third coordinate value. In S332, the gas leak detection system 10 may store the composite image and output the stored composite image based on a user input to the gas leak detection system 10. According to one embodiment, the gas leak detection system 10 may search for at least a portion of the stored composite image based on a user input accessing the stored composite image and reproduce a portion of the retrieved composite image.

Figure 4 is a flowchart of a method for identifying whether an artificial intelligence-based gas leak detection electronic device detects a target object according to an embodiment.

In S410, the electronic device 1000 may acquire first type images of the space to be monitored from at least one first type camera connected to the electronic device. For example, the first type of camera includes an EO camera, and the first type images may be visible light images with a preset frame interval. In S420, the electronic device 1000 may acquire second type images of the space to be monitored from at least one second type camera connected to the electronic device. For example, the second type of camera may be an OGI camera, and the second type images may be infrared images including a plurality of infrared images at preset frame intervals. According to one embodiment, the shooting parameters such as the angle of view and the image acquisition period for the monitoring target space of the first type camera and the second type camera to which the electronic device 1000 is connected may be prepared differently.

In S430, when first type images and the second type images are input, the electronic device 1000 uses an object recognition model to identify whether a target object is detected in the first type images and the second type images. By inputting first type images and second type images, it is possible to identify whether a target object is detected in the first type images and the second type images. According to one embodiment, the object recognition model is pre-trained to output an object candidate area that includes at least a target object related to at least one of a gas object, a smoke object, or a flame object from each of the visible light images and the infrared images. It could be an artificial intelligence model. The object recognition model is a neural network model in which a plurality of weights are modified and updated to reduce or minimize the loss or cost value using a known loss function. The pixels that make up the object candidate area containing the target object are Coordinate information on the image, outline information, boundary information indicating the boundary of the object candidate area, and type information of the recognized target object can be output.

According to one embodiment, the object recognition accuracy for image information of the electronic device 1000 may be 70% or more, but is not limited thereto, and the recognition accuracy of a gas leak using at least one artificial intelligence model, which will be described later, may be 99% or more. It may be more than %, but it is not limited to this. In addition, although not shown in FIG. 4, the electronic device 1000 identifies whether a target object is detected in the first type images and the second type images through at least one artificial intelligence model, and detects the identified target object. It can be tracked and monitored at preset resolutions and video speeds.

Although not shown in FIG. 4, when it is identified that a target object is detected from at least one of the first type images and the second type images, the electronic device 1000 according to an embodiment selects a type for which the target object is detected. Shape information of the target object can be extracted from images. A specific method for the electronic device 1000 to extract shape information of a target object will be described in detail with reference to FIGS. 7 to 9 described later.

FIG. 5 is a diagram illustrating a process in which an artificial intelligence-based gas leak detecting electronic device extracts shape information of a target object identified according to a frequency pattern, according to an embodiment.

In S510, when the electronic device 1000 identifies that the target object is not detected in the first type images and the second type images based on the output value of the object recognition model, the electronic device 1000 generates second type images. Transformed image information in the frequency domain can be generated by Fourier transform. For example, when the electronic device 1000 identifies that the target object is not primarily detected from the first type images and the second type images using an object recognition model, it uses the object recognition model to generate second type images that are infrared images. Transformed image information in the frequency domain can be generated by using discrete Fourier transform (DFT). Although not shown in FIG. 5, according to another embodiment, the electronic device 1000 may generate transformed image information in the frequency domain by preprocessing second type images and performing Fourier transform on the preprocessed second type images.

In S520, the electronic device 1000 may extract spectrum information for each of the target object and background interference materials around the target object from the converted image information. For example, converted image information may include spectral information representing a target object based on a specific wavelength range in the frequency domain. According to one embodiment, the spectrum information may include magnitude and phase information for each frequency. According to another embodiment, the electronic device 1000 may extract spectrum information for each of the target object and background interference materials around the target object within the converted image information. According to another embodiment, the electronic device 1000 may extract spectral information about size and phase for each frequency from converted image information without distinguishing between target objects and background interference materials.

In S530, the electronic device 1000 inputs the extracted spectrum information into a likelihood ratio test model that outputs whether there is a frequency pattern according to the target object when the extracted spectrum information is input. It is possible to identify whether there is a frequency pattern according to the target object. For example, the electronic device 1000 trains in advance a likelihood ratio test model that classifies the frequency pattern of the target object, and applies the learned likelihood ratio test model to the converted image information under predetermined conditions to target the target object together with the object recognition model. Objects can be detected effectively.

If the target object is not identified as being detected from the object recognition model, the electronic device 1000 according to the present disclosure uses the frequency pattern of the spectral information appearing from the infrared image to identify the first type images and the second type images using a likelihood ratio test model. Target object detection accuracy can be improved by re-identifying whether the target object in the type images is detected. Although not shown in FIG. 5, when the electronic device 1000 identifies that the target object is detected as a result of re-identification using the likelihood ratio test model, the electronic device 1000 selects the target from the images in which the target object was detected according to the output value of the likelihood ratio test model. A masking data creation process is performed to separate the object area.

In S540, when a frequency pattern according to the target object is identified, the electronic device 1000 may extract shape information of the target object from the second type images based on the output value of the likelihood ratio test model. For example, the electronic device 1000 determines that the target object has been detected when a frequency pattern according to the target object is identified in the converted image information based on the output value of the likelihood ratio test model, and selects the second type images. Shape information about the shape of the target object is extracted from. A method by which the electronic device 1000 extracts shape information will be described with reference to FIGS. 7 to 9 described later.

FIG. 6 is a diagram illustrating a process in which an artificial intelligence-based gas leak detecting electronic device extracts shape information of a target object identified according to a frequency pattern according to an embodiment.

In S610, when the frequency pattern according to the target object is not identified based on the output value of the likelihood ratio test model, the electronic device 1000 generates the extracted spectrum information, and when the extracted spectrum information is input, the target object By inputting it into a support vector machine model that outputs whether there is a frequency pattern according to , it is possible to identify whether there is a frequency pattern according to the target object in the extracted spectrum information.

That is, when the target object is not detected in the first type images and the second type images as a result of using the object recognition model, the electronic device 1000 according to the present disclosure uses the likelihood ratio test model and the support vector machine model to detect the target object. The detection accuracy of the target object can be improved by re-identifying whether the target object is detected. When it is determined that a frequency pattern according to the target object is identified as a result of using the support vector machine model, the electronic device 1000 separates the target object area from the images in which the target object is detected according to the output value of the support vector machine model. The masking data generation process is performed for

When spectral information extracted from converted image information is input, the electronic device 1000 according to the present disclosure trains a support vector machine model that classifies the frequency pattern of the target object in advance, and uses the learned support vector machine model under predetermined conditions. By applying the transformation to image information, the target object can be accurately detected along with the object recognition model and likelihood ratio test model.

In S620, when a frequency pattern according to the target object is identified, the electronic device 1000 may extract shape information of the target object from the second type images based on the output value of the support vector machine model. . A method for the electronic device 1000 to extract shape information will be described in detail with reference to FIGS. 7 to 9 described later. As described above with reference to FIGS. 4 to 6 , the electronic device 1000 analyzes at least one type of images using at least one of an object recognition model, a likelihood ratio test model, or a support vector machine model to determine the target object. Not only can it accurately identify whether it is detected, but it can also generate an infrared image as converted image information in the frequency domain, and based on the frequency pattern for each predetermined frequency that appears in the converted image information, the type of target object detected (for example, is it methane? , ammonia, general smoke, or flame) can be identified together.

Figure 7 is a flowchart of a specific method in which an artificial intelligence-based gas leak detection electronic device extracts shape information of a target object according to an embodiment.

In S710, the electronic device 1000 may preprocess images of the type in which the target object was detected. For example, as described above in FIGS. 3 to 6, the electronic device 1000 uses an object recognition model and a likelihood ratio test model or support vector machine model for transformed image information according to preset conditions to create a first type image. and identify whether the target object is detected from at least one of the second type images, and if it is identified that the target object is detected, preprocess the images of the type in which the target object is detected (e.g., binarization, normalization, augmentation, etc. )do.

In S720, the electronic device 1000 may generate masking data based on preset target object information. For example, the electronic device 1000 may store information about the target object in advance. According to one embodiment, the information about the target object may store in advance moving average values regarding the degree to which gas objects, smoke objects, or flame objects move between frame images in the images, based on the information about the target object. Thus, masking data can be generated.

Although not specifically shown in FIG. 7, the electronic device 1000 determines a moving average value for frame images preceding the current frame image of images of the type in which the target object is detected, and masks data based on the determined moving average value. can be created. For example, the electronic device 1000 determines a moving average value for a set of frame images preceding the current frame of images of the type in which the target object is detected, generates a background model based on the determined moving average value, and creates a background model based on the generated background. Generate masking data to remove (e.g. subtract) the model from the current frame image. The electronic device 1000 according to an embodiment of the present disclosure may generate masking data using a running average background subtraction technique.

In S730, the electronic device 1000 may separate target object areas generated by applying masking data to images of the type in which the target object was detected from the preprocessed images. The electronic device 1000 continuously modifies and updates the moving average value of all frame images whenever new frame images are acquired, and applies masking data generated based on the modified and updated moving average value to the current frame image to identify the target object. Regions can be separated from preprocessed images. In S740, the electronic device 1000 extracts shape information of the target object from the separated target object areas. The process of extracting shape information by the electronic device 1000 will be further described with reference to FIGS. 8 and 9 described later.

FIG. 8 is a diagram illustrating a process in which an artificial intelligence-based gas leak detection electronic device identifies the shape of a target object and tracks it, according to an embodiment.

A method by which the electronic device 1000 extracts shape information will be described in more detail with reference to FIG. 8 . Referring to Figure 810, the electronic device 1000 moves from all frame images constituting images of the type in which the target object was detected or from all frame images preceding the current frame images according to the method described above in FIG. 7. A process of separating the target object area 816 from the current frame image 814 by determining an average value 812 and applying masking data determined based on the determined moving average value 812 to the current frame image 814 is shown.

The electronic device 1000 may identify outline information about a target object within a separated target object area, identify center point information of the target object based on the outline information, and identify center point information of the target object based on the identified center point information and the outline information. The direction vector of the target object area can be identified. The electronic device 1000 according to the present disclosure may extract the outline information, center point information, and direction vector as shape information of the target object for each frame image constituting the image of the type in which the target object is detected.

The electronic device 1000 according to the present disclosure may determine a motion vector based on shape information of the target object separated from adjacent frame images within images of the type in which the target object was detected. As shown in Figure 820, the electronic device 1000 uses a motion vector determined from the images within the type in which the target object was detected to determine target object areas 822 within the images of the type in which the target object was detected. ) can be modeled. As shown in Figure 830, the electronic device 1000 can identify the sum vector of motion vectors used to model the target object, and the location where the target object was created based on the size and direction of the sum vector. The flow vector regarding the flow direction and speed of the leak origin and target object can be determined.

Figure 9 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device extracts shape information of a target object according to another embodiment.

In S910, the electronic device 1000 may identify outline information about the target object within the separated target object area. For example, the electronic device 1000 may identify the coordinates of pixels constituting the separated target object area by applying masking data, and determine the coordinate set of the outermost pixels based on the coordinate values of the identified pixels. You can. The electronic device 1000 may identify outline information about the target object based on the coordinate set of the outermost pixels.

In S920, the electronic device 1000 may identify center point information of the target object based on outline information. For example, the electronic device 1000 targets the target by applying a Mean Shift algorithm to pixel values inside the boundary of the target object determined based on outline information, or by applying an object tracking algorithm to find the center point of the target object that is moving. The center point information of an object can be identified. In S930, the electronic device 1000 may identify the direction vector of the target object area based on center point information and outline information. For example, the electronic device 1000 separates the target object area for each frame image in the images of the type in which the target object is detected, and selects adjacent frame images (e.g. A motion vector can be determined based on shape information extracted from (images having two consecutive frame numbers).

The electronic device 1000 may model the target object using a motion vector determined between frame images with adjacent frame numbers among images within the type in which the target object is detected, and the motion used to model the target object. A direction vector in the same direction as the direction of the vector sum vector can be identified. In S940, the electronic device 1000 may extract the outline information, center point information, and direction vector determined in the above-described process as shape information of the target object.

Figure 10 is a flowchart of a specific method in which an artificial intelligence-based gas leak detection electronic device determines the leak origin and flow vector of a target object according to an embodiment.

In S1010, the electronic device 1000 may determine a motion vector between adjacent target object areas in images of the type in which the target object is detected, based on shape information of the separated target object areas. For example, as described above in FIG. 9, the electronic device 1000 selects each frame image between adjacent frame images (e.g., two frame images with consecutive frame numbers) within images of the type in which the target object is detected. A motion vector may be determined based on shape information of the separated target object in the images.

In S1020, the electronic device 1000 may model separated target object areas based on the motion vector. In S1030, the electronic device 1000 may determine a sum vector of motion vectors representing modeled target object areas. For example, the electronic device 1000 may determine a sum vector by performing geometric vector sum on motion vectors appearing about the target object from images.

In S1040, based on the size and direction of the determined sum vector, the electronic device 1000 determines the leak origin regarding the location where the target object was created and the flow direction and speed of the target object in the images in which the target object was detected. The flow vector can be determined. For example, the electronic device 1000 may determine the starting point of the sum vector as the leak source, and determine the direction indicated by the sum vector from the start point to the end point as the direction of the flow vector. Additionally, the electronic device 1000 may determine the distance from the start point to the end point, the speed of the target object, and the size of the flow vector. The electronic device 1000 may determine a flow vector having the size and direction determined by the method described above.

Figure 11 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device generates a composite image for gas leak detection according to an embodiment.

In S1110, the electronic device 1000 receives first type images of the space to be monitored from at least one first type of camera connected to the electronic device and the monitoring target from at least one second type of camera connected to the electronic device. A second type of images of space can be obtained. According to one embodiment, the first type of camera and the second type of camera include an Electro-Optical (EO) camera and an Optical Gas Imaging (OGI) camera, respectively, and the first type of images and the second type of camera include The type images may include visible light images and infrared images, respectively. Since S1110 may correspond to S410 to S420 of FIG. 4, detailed description will be omitted.

In S1120, when the electronic device 1000 identifies that a target object is detected from at least one of the first type images or the second type images, the first type images related to the detected target object and the detection A leakage image can be generated by combining second type images corresponding to the first type images regarding the target object. According to another embodiment, the electronic device 1000 may generate one type of images in which a target object is detected among the first type images or the second type images as a leak image.

In S1130, the electronic device 1000 may generate a composite image by combining the leakage image with the panoramic image based on the coordinates of the previously acquired panoramic image of the leakage image. For example, the electronic device 1000 determines the leak origin based on the sum vector of the motion vectors of the modeled target object, and determines the leak origin based on the laser measurement value for the determined leak origin and pan-tilt information of the camera installation set. The coordinates of the panoramic image of the origin can be determined. The electronic device 1000 may generate a composite image by combining the leakage image with a panoramic image based on the coordinate values.

FIG. 12 is a diagram illustrating a leak image generated by an electronic device for detecting an artificial intelligence-based gas leak and a process of combining the leak image into a panoramic image, according to an embodiment.

The electronic device 1000 may be connected to at least one EO camera and at least one OGI camera by photographing the space to be monitored. As shown in Figure 910, the electronic device 1000 may generate a leakage image by combining the acquired first type images and second type images. For example, the leakage image 910 may include a target object (eg, a gas object, a smoke object, or a flame object) area 914 and a background area 912 detected as an image of the space to be monitored.

As shown in Figure 930, the electronic device 1000 determines the coordinates 932 for the panoramic image of the leak image and synthesizes the leak image 936 into the panoramic image 934 based on the determined coordinate values. By doing this, a composite image can be created. The electronic device 1000 can directly output the generated composite image or transmit it to a control terminal or control server connected to the electronic device 1000 so that the controller managing the monitoring target space can easily recognize the target object.

Figure 13 is a flowchart of a method in which an artificial intelligence-based gas leak detection electronic device generates a leak image according to an embodiment.

In S1310, the electronic device 1000 may identify first type images related to the detected target object and second type images corresponding to the first type images. However, according to another embodiment, the electronic device 1000 may generate one type of image in which the target object is detected among the first type images and the second type images as a leak image. For example, the electronic device 1000 detects a target object by using at least one of an object recognition model, a likelihood ratio test model, or a support vector machine model according to predetermined conditions, and determines the type of image in which the target object is detected. You can. Additionally, the electronic device 1000 may identify different types of images obtained by photographing the same physical space corresponding to the type of image in which the target object was detected.

In S1320, the electronic device 1000 may geometrically match first type images and second type images corresponding to the first type images. According to one embodiment, the electronic device 1000 uses a homography matching technique or a homography matching technique to match first type images (e.g., general camera images) and second type images (OGI images) corresponding to the first type images. Geometric matching can be performed based on at least one of the H matrix correspondence method techniques. However, according to another embodiment, the electronic device 1000 may geometrically match the first type images and the second type images using a general template matching method.

In S1330, the electronic device 1000 may generate a leakage image by correcting matching errors in geometrically matched first type images and second type images corresponding to the first type images. For example, the electronic device 1000 geometrically matches the first type images and the second type images using a perspective transformation matrix to perform perspective transformation between two corresponding types of images. Afterwards, the incorrectly matched matching error can be determined in the homography matching process using the homography function. The electronic device 1000 may generate a leakage image by extracting feature points from each type of image and correcting matching errors based on the extracted feature points, as shown in FIG. 14 described later.

Figure 14 is a flowchart of a specific method in which an artificial intelligence-based gas leak detecting electronic device generates a leak image according to an embodiment.

In S1410, when a target object is detected based on at least one of an object recognition model, a likelihood ratio test model, or a support vector machine model, the electronic device 1000 displays first type images and second type images related to the detected target object. Feature points can be extracted from images. According to one embodiment, the electronic device 1000 may extract feature points from each of the first and second type images related to the detected target object using a feature point extraction technique such as SIFT, SURF, or ORB. .

In S1420, the electronic device 1000 may identify feature points of second type images that match feature points extracted from first type images. According to one embodiment, the electronic device 1000 is configured to match a second feature point extracted from the first type images based on various feature point matching techniques such as a brute force matching technique, a hash matching technique, and a template matching technique. Characteristic points of type images can be identified. In S1430, the electronic device 1000 generates first type images and the first type images based on the average displacement value between the feature points extracted from the first type images and the feature points extracted from the second type images. Matching errors can be corrected by changing the matching positions of the second type images corresponding to the images.

For example, the electronic device 1000 may determine the displacement value based on the difference between the coordinate values of the feature points extracted from the first type images and the coordinate values of the feature points extracted from the matching second type images, The average displacement value of feature points extracted from each type of image can be identified. The electronic device 1000 corrects the matching error by changing the matching positions of the geometrically matched first-type images and second-type images based on the average displacement value. In S1440, the electronic device 1000 may generate first type images and second type images that are geometrically matched according to the changed matching position as a leak image.

Figure 15 is a flowchart of a method in which an electronic device for detecting gas leaks based on artificial intelligence generates a composite image according to an embodiment.

In S1510, when it is identified that the target object is detected from at least one of the first type images or the second type images, the electronic device 1000 detects the target object from the images of the type in which the target object was detected. Shape information can be extracted. The process by which the electronic device 1000 extracts shape information of the target object may correspond to the process described above with reference to FIGS. 7 to 9 .

In S1520, the electronic device 1000 may determine the coordinates of the previously acquired panoramic image of the leaked image generated based on shape information. For example, the electronic device 1000 may previously store a panoramic image of the space to be monitored. According to one embodiment, the electronic device 1000 may convert visible light images obtained from at least one EO camera into one panoramic image and store the converted panoramic image in advance.

The electronic device 1000 determines a motion vector based on shape information extracted from adjacent frame images in images of the type in which the target object is detected, determines a sum vector based on the determined motion vector, and determines the sum vector of the sum vector. The leak origin and flow vector are determined based on the size and direction, and the coordinates for the panoramic image of the leak image are determined based on the determined leak origin and flow vector. In S1530, the electronic device 1000 may generate a composite image by combining the leaked image with the panoramic image based on the determined coordinates. According to one embodiment, the electronic device 1000 may display the leak image by overlapping it with the panoramic image, and may further overlap and display the type of target object and location information about the origin of the leak where the target object occurred.

Figure 16 is a flowchart of a method by which an artificial intelligence-based gas leak detection electronic device determines the coordinates of a leak image for a panoramic image, according to an embodiment.

In S1610, the electronic device 1000 may determine a leak origin regarding the location where the target object was created and a flow vector regarding the flow direction and speed of the target object, based on shape information of the separated target object areas. For example, the electronic device 1000 determines a motion vector based on shape information extracted from adjacent frame images in images of the type in which the target object is detected, models the target object based on the motion vectors, and , the leak origin and the flow vector can be determined based on the magnitude and direction of the sum vector of motion vectors used to model the target object.

In S1620, the electronic device 1000 may determine the coordinates of the panoramic image of the leak image based on the leak origin and flow vector. According to one embodiment, the electronic device 1000 may determine the starting point of the sum vector as the leak source, and determine the direction indicated by the sum vector from the start point to the end point as the direction of the flow vector. Additionally, the electronic device 1000 may determine the distance from the start point to the end point, the speed of the target object, and the size of the flow vector.

According to one embodiment, the electronic device 1000 receives a laser measurement value reflected from a physical space corresponding to the leak origin from a laser measuring device based on the direction indicated by the leak origin and the flow vector determined according to the sum vector. The distance to physical space can be identified based on the acquired laser measurement values. Additionally, the electronic device 1000 may acquire pan-tilt information of the camera installation set with respect to the origin of the leak. The electronic device 1000 may determine the first coordinate value in a spherical coordinate system based on the leak origin, flow vector, distance to the physical space where the target object occurs, and pan-tilt information. The electronic device 1000 converts the first coordinate value into a second coordinate value in the Cartesian coordinate system and reflects the lens distortion value preset for the camera type that transmitted the images in which the target object was detected in the second coordinate value. The third coordinate value can be obtained. The electronic device 1000 may determine the third coordinate value determined according to the above-described method as the coordinate for the panoramic image of the leaked image.

FIG. 17 is a diagram illustrating a process in which an electronic device for detecting an artificial intelligence-based gas leak according to an embodiment converts the coordinates of the origin of the leak based on laser measurement values and camera pan-tilt information.

In S1710, the electronic device 1000 may obtain the leak origin and flow vector. In S1710, the process of determining the leak origin and flow vector based on the motion vector used by the electronic device 1000 to model the target object may correspond to the process described above with reference to FIG. 16 and will therefore be omitted. In S1720, the electronic device 1000 may transmit a control signal according to the direction of the leak origin and flow vector to the laser range finder. In S1730, the electronic device 1000 may obtain a laser measurement value reflected from a point in physical space corresponding to the leak origin from a laser measuring device.

In S1740, the electronic device 1000 may acquire camera pan-tilt information of the camera installation set with respect to the leak origin. In S1750, the electronic device 1000 may determine a first coordinate value in a spherical coordinate system based on the leak origin, the flow vector, the distance to the physical space where the target object occurs, and pan-tilt information. In S1760, the electronic device 1000 may convert the determined first coordinate value into a second coordinate value in the Cartesian coordinate system. In S1770, the electronic device 1000 reflects the lens distortion value preset for the type of camera that transmitted the images in which the target object was detected among the first type camera or the second type camera to the second coordinate value, thereby The 2 coordinate value can be changed to the 3rd coordinate value. The electronic device 1000 may generate a composite image by combining the leakage image with the panoramic image based on the determined third coordinate value.

Figure 18 is a diagram showing the operation process of an artificial intelligence-based gas leak detection system according to another embodiment.

Referring to FIG. 18, a process in which the gas leak detection system 10 uses a composite image generated by an electronic device according to the method described in FIGS. 11 to 16 is shown. However, according to another embodiment, it goes without saying that each operation described in FIG. 18 can be performed not only by the gas leak detection system 10 but also by the electronic device 1000. For example, the electronic device 1000 may generate a composite image by combining the leakage image with the panoramic image according to the method described above with reference to FIGS. 11 to 16 in S330.

In S332, the electronic device 1000 may store and transmit the composite image. According to one embodiment, the electronic device 1000 may manage image data according to the RDBMS 331 (Relative Database Management System) structure for efficient design within limited data. According to one embodiment, the electronic device 1000 may output certain images based on a user input to the electronic device 1000 to search stored image data. According to one embodiment, the electronic device 1000 may output certain stored image data based on a portion of a keyword and a word combination using DB SELECT. In addition, according to one embodiment, the electronic device 1000 configures a system based on SQL for efficient search of RDBMS, and has a SELECT structure for arranging strings that limit conditions within multiple conditions to improve performance and optimize user experience. It can be designed as a basic door.

In S334, when the electronic device 1000 identifies that a target object is detected by using at least one of an object recognition model, a likelihood ratio test model, or a support vector machine model, the electronic device 1000 determines the leak origin based on the shape information of the detected target object. and determine the type of target object, and generate alert and notification content including information on the leak origin and type of target object. Additionally, the electronic device 1000 may also generate a predetermined control signal for operating the alarm devices.

In S336, the electronic device 1000 transmits a control signal for controlling the alarm device to control the operation of the warning light to emit light, or controls the speaker device to output a predetermined sound signal. According to one embodiment, of course, the electronic device 1000 may control a certain alarm device through a control terminal installed at the work site.

At S338, the electronic device 1000 may output the content generated at S334. For example, the electronic device 1000 may output the content generated in S334 on the display of the electronic device, but the display of the control terminal connected to the electronic device 1000 or the display device of the situation room system connected through the control server You can also transmit content. According to one embodiment, contents transmitted by the electronic device 1000 may be stored in the RDBMS 331 in real time, and are called based on the controller's user input to the electronic device 1000, the control terminal, or the situation room system. By doing so, it may be output on a certain display.

At S340, the electronic device 1000 may generate control signals for controlling the camera installation set 342, the laser range finder 344, and the EO and OGI cameras 346. According to one embodiment, the electronic device 1000 installs a camera installation set 342, a laser range finder 344, and EO and OGI cameras 346 based on the controller's user input to the electronic device, control terminal, or situation room system. A control signal to control can be generated and a specific point within the monitoring target space can be specifically monitored.

Figure 19 is a flowchart of a method in which an artificial intelligence-based gas leak detecting electronic device detects a gas leak using an artificial intelligence model according to an embodiment.

In S1910, the electronic device 1000 receives first type learning images of a space to be monitored from at least one first type of camera connected to the electronic device, and the monitoring image from at least one second type of camera connected to the electronic device. Second type learning images for the target space can be obtained. According to another embodiment, the electronic device 1000 may acquire visible light images and infrared images including a target object as first type learning images and second type learning images from an external device connected to the electronic device. there is. According to one embodiment, the first type learning images and the second type learning images may be visible light images and infrared images.

In S1920, the electronic device 1000 may generate learning data and verification data based on the first type training images and the second type learning images. For example, the electronic device 1000 may generate training data and verification data according to a preset ratio among the first type training images. Additionally, the electronic device 1000 may generate training data and verification data according to a preset ratio among the second type training images.

In S1930, when first type images obtained from the first type camera and second type images obtained from the second type camera are input based on the generated learning data, the electronic device 1000 performs the An object recognition model that identifies whether a target object is detected in the first type images and the second type images can be trained. According to one embodiment, the electronic device 1000 creates a first object recognition model that identifies whether a target object in the first type images is detected when the first type images are input, based on the first type learning images. After training, based on the second type training images, a second object recognition model that identifies whether the target object in the second type images is detected when the second type images are input, a first object recognition model And one object recognition model may be created using the second object recognition model.

In S1940, the electronic device 1000 detects the target object in the first type images and the second type images by inputting the first type images and the second type images into a learned object recognition model. can be identified. Although not shown in FIG. 19, when the electronic device 1000 identifies that a target object is detected, it extracts shape information of the target object from images of the type in which the target object was detected.

FIG. 20 is a diagram illustrating a process in which an electronic device for detecting an artificial intelligence-based gas leak learns an object recognition model and utilizes the learned object recognition model, according to an embodiment.

In S2002, the electronic device 1000 may acquire image data for learning an artificial intelligence model and analyze the acquired image data. In S2004, the electronic device 1000 may preprocess the acquired image data. In S2006, the electronic device 1000 may separate some of the acquired image data into learning data and the remaining part into verification data according to a preset ratio.

At S2008, the electronic device 1000 may determine the artificial intelligence algorithm and model to be used for target object recognition. In S2010, the electronic device 1000 may learn an artificial intelligence model based on training data, which is some of the image data, according to the determined algorithm and model details. In S2012, the electronic device 1000 may verify the performance of an artificial intelligence model learned based on verification data. In S2014, the electronic device 1000 may fine-tune the parameters of the artificial intelligence model if, as a result of verifying the performance of the artificial intelligence model, it is determined that the performance of the artificial intelligence model does not satisfy a predetermined condition. However, in S2016, if the electronic device 1000 verifies the performance of the artificial intelligence model and determines that the performance of the artificial intelligence model satisfies a predetermined condition, the electronic device 1000 may extract the parameters of the corresponding artificial intelligence model.

In S2018, the electronic device 1000 may determine whether the artificial intelligence algorithm and model need to be changed based on the results of fine tuning the parameters of the artificial intelligence model. If there is a need to change the artificial intelligence algorithm and model, the electronic device 1000 selects a new artificial intelligence algorithm and model. If there is no need to change the artificial intelligence algorithm and model, the electronic device 1000 re-trains the algorithm and model. You can evaluate whether the performance of the learned model is satisfactory.

In S2020, the electronic device 1000 may acquire an input image for target object recognition and preprocess the obtained input image. In S2022, the electronic device 1000 may learn the artificial intelligence model by modifying and updating the parameters of the actually used artificial intelligence model based on the parameters of the artificial intelligence model determined in S2016. In S2024, the electronic device 1000 identifies whether the target object is detected based on the result of inputting the preprocessed input image to the learned artificial intelligence model, and if it is identified that the target object is detected, the target object Information about can be stored and returned.

Figure 21 is a flowchart showing a process of learning a likelihood ratio test model and a support vector model by an electronic device for detecting gas leaks based on artificial intelligence according to an embodiment.

In S2110, the electronic device 1000 generates the second type images or the spectrum extracted from the second type images based on the second type learning images or spectrum learning information extracted from the second type learning images. When information is input, a likelihood ratio test model that identifies whether there is a frequency pattern according to the target object in the extracted spectrum information can be trained.

In S2120, the electronic device 1000 generates the second type images or the spectrum extracted from the second type images based on the second type learning images or spectrum learning information extracted from the second type learning images. When information is input, a support vector machine model that identifies whether there is a frequency pattern according to the target object in the extracted spectrum information can be trained. The specific process by which the electronic device 1000 learns the likelihood ratio test model and the support vector model will be described in detail with reference to FIG. 22, which will be described later.

FIG. 22 is a diagram illustrating a process in which an electronic device for detecting gas leaks based on artificial intelligence learns a likelihood ratio test model and a support vector model according to an embodiment.

In S2202, the electronic device 1000 acquires frequency data for learning a likelihood ratio test model and analyzes the acquired frequency data. According to one embodiment, the frequency data may be image data generated by converting second type training images into the frequency domain. In S2204, the electronic device 1000 may preprocess frequency data. In S2206, the electronic device 1000 may set some of the preprocessed frequency data as learning data and the remaining part as verification data.

In S2208, the electronic device 1000 may select an algorithm for classifying the frequency pattern according to the target object. According to one embodiment, the electronic device 1000 may use a likelihood ratio test algorithm as an algorithm for classifying spectral information obtained from an infrared image. In S2210, the electronic device 1000 may learn an algorithm selected based on learning data and a model according to the algorithm.

In S2212, the electronic device 1000 may verify whether the performance of the learned model satisfies a predetermined target value. In S2214, when the electronic device 1000 determines that the performance of the learned model does not meet a predetermined target value, the electronic device 1000 may perform fine tuning on the model. In S2216, if the performance of the model learned in S2212 is identified as satisfying a predetermined target value, the electronic device 1000 may extract parameters of the learned model.

In S2218, the electronic device 1000 may acquire an infrared image that is a target object detection target to input it into the learned model, and perform preprocessing on the frequency data obtained by Fourier transforming the acquired infrared image. In S2220, the electronic device 1000 determines whether a frequency pattern according to the target object is identified by inputting preprocessed frequency data or spectrum information obtained from the frequency data into a learned artificial intelligence learning model (e.g., likelihood ratio test model). You can check it.

Additionally, according to one embodiment, the electronic device 1000 may further use another pattern classification algorithm and a model learned based on it to detect a frequency pattern according to the target object. For example, the electronic device 1000 can accurately extract a frequency pattern according to a target object by training a support vector machine model and using the learned support vector machine model together with a likelihood ratio test model under predetermined conditions. Since the operations shown in S2230 to S2250 may correspond to S2202 to S2220, detailed description will be omitted.

FIG. 23 is a diagram illustrating a process in which an artificial intelligence-based gas leak detecting electronic device detects a gas leak using a plurality of software modules according to another embodiment.

Each module shown in FIG. 12 may be a set of one or more instructions stored in memory that can be executed by the processor of the electronic device 1000 for gas leak detection. According to one embodiment, the electronic device 1000 includes one or more instructions divided into an AI gas image and smoke detection module 2320, an origin detection module 2330, a visualization module 2340, and an alarm and control module 2350. By storing in memory and executing the stored instructions, target objects for the space to be monitored can be detected. According to one embodiment, S/W used by the electronic device in S2322 to S2328 may be included in the AI gas image and smoke detection module 2320.

For example, at S2322, the electronic device 1000 receives visible light images (e.g., EO images) from a camera setup 2310 that includes an EO camera 2314, an OGI camera 2312, and a laser meter 2316. and infrared images (eg, OGI images) can be acquired. In S2323, the electronic device 1000 executes the object recognition technology S/W, and when the target object is identified as not primarily detected from the visible light images and the infrared images, the electronic device 1000 performs the infrared image frequency conversion S/W. By executing, infrared image information according to the space-time domain can be converted into converted image information in the frequency domain. The electronic device 1000 may extract a spectrum from converted image information in the frequency domain.

In S2324, the electronic device 1000 may learn a likelihood ratio test model and an SVM model for identifying a frequency pattern according to a target object by executing S/W for learning a likelihood ratio test model and an SVM model. In S2325, the electronic device 1000 may identify whether there is a frequency pattern according to the target object in the infrared image using at least one of a likelihood ratio test model or an SVM model by executing frequency-specific pattern analysis S/W.

In S2326, when the target object is recognized by executing the object recognition technology S/W or a frequency pattern according to the target object is identified from the infrared image, the electronic device 1000 executes the smoke/background separation S/W to determine the target object area. can be separated. In S2327, the electronic device 1000 uses masking data by executing the smoke shape detection S/W to separate the target object area from the images of the type in which the target object was detected, and to provide shape information of the separated target object areas. It can be extracted. In S2328, the electronic device 1000 executes the smoke origin/flow detection S/W to separate target object areas based on shape information extracted from each adjacent frame image in the images of the type in which the target object was detected. The motion vector for can be determined, and the leak origin and flow vector can be determined based on the sum vector of the determined motion vectors.

According to one embodiment, the origin detection module 2330 used by the electronic device 1000 may include distance spherical coordinate recognition S/W and location information conversion S/W. For example, in S2329, by executing the distance spherical coordinate recognition S/W, the distance to the physical space where the target object corresponding to the leak origin occurs is obtained, and the pan-tilt information of the camera about the acquired leak origin is acquired; , Coordinates in a spherical coordinate system can be determined based on the leak origin, flow vector, distance to the physical space where the target object occurs, and pan-tilt information. In S2330, the electronic device 1000 may convert the coordinates in the determined spherical coordinate system into coordinate values according to the rectangular coordinate system or distortion information correction of the camera by executing the location information conversion S/W.

In S2340, the electronic device 1000 may generate a composite image by compositing the leaked image with the panoramic image based on the coordinate values determined in S2330 by executing the image synthesis S/W. According to one embodiment, the visualization module 2340 may include image synthesis S/W. Additionally, according to one embodiment, the alarm/control module 2350 may include environment setting S/W, control function S/W, storage function S/W, search function S/W, and alarm function S/W.

In S2352, the electronic device 1000 can adjust the operating environment of the camera installation set 2310 for the monitoring target space by executing the environment setting S/W. In S2354, the electronic device 1000 can control the operation of the camera device and the laser measuring device in the camera installation set 2310 in detail by executing the control function S/W. In S2356, the electronic device 1000 may store the composite image acquired in S2340 in memory by executing the storage function S/W. In S2357, the electronic device 1000 may perform a function of searching and outputting a predetermined image based on a user input to access a stored composite image by executing the search function S/W. According to one embodiment, the electronic device 1000 stores EO images and OGI images by time, camera, and location, and can not only search predetermined images according to search conditions by time, but also detect target objects. Event records can be stored, and records for each event can be output upon user request.

In S2358, the electronic device 1000 generates predetermined alarm content and a control signal when it is identified that a target object in the image is detected by executing the alarm function S/W, and sends the generated alarm content to the output unit 2360. The light emitting operation can be controlled by transmitting it to the image display or transmitting the generated control signal to the warning light of the output unit 2360.

FIG. 24 is a diagram illustrating a process for detecting a gas leak by linking a measuring device, an artificial intelligence-based gas leak detection server, an alarm/control server, and a control device with each other, according to another embodiment.

According to one embodiment, the gas leak detection processes described above in FIGS. 1 to 23 may be performed by the gas leak detection system 10 or the electronic device 1000. According to one embodiment, the gas leak detection system 10 may include a measurement device 2410, an artificial intelligence-based gas leak detection server 2420, an alarm and control server 2340, and a control device 2440. With reference to FIG. 24, the process of detecting a gas leak by linking the measuring device 2410, the artificial intelligence-based gas leak detection server 2420, the alarm and control server 2340, and the control device 2440 with each other will be explained. .

In S2402, the measurement device 2410 may acquire image information and laser measurement values. According to another embodiment, the measurement device 2410 acquires image information first, and when the leak origin is detected under the control of the alarm and control server 2340, the laser measurement value for the leak origin is later acquired. You may. In S2404, the measurement device 2410 may transmit image information and laser measurement values to the artificial intelligence-based gas leak detection server 2420.

In S2406, the artificial intelligence-based gas leak detection server 2420 may transmit image information and laser measurement values to the alarm and control server 2430. However, the above-described process may not be performed, and the artificial intelligence-based gas leak detection server 2420 may analyze image information and laser measurement values only within its own server device. In S2408, the artificial intelligence-based gas leak detection server 2420 may identify whether a target object in the acquired image information is detected using at least one artificial intelligence model.

In S2410, the artificial intelligence-based gas leak detection server 2420 may determine the leak origin and flow vector based on shape information of the target object when it is identified that the target object is detected. In S2412, the artificial intelligence-based gas leak detection server 2420 may determine the distance spherical coordinates based on the leak origin and the flow vector, and determine the coordinates for the panoramic image of the leak image by converting the determined distance spherical coordinates. In S2414, the artificial intelligence-based gas leak detection server 2420 may generate a composite image by combining the leak image with the panoramic image.

In S2416, the artificial intelligence-based gas leak detection server 2420 may transmit event detection results, image information, laser measurements, synthetic images, and analysis results of the artificial intelligence model to the alarm and control server 2430. In S2418, the alarm and control server 2431 may transmit the event detection result, image information, laser measurement value, synthetic image, and analysis result of the artificial intelligence model to the control device 2440. In S2420, the control device 2420 may display a control image including a composite image through the control display, generate predetermined warning content, and display the generated warning content by overlapping it with the control image.

In S2422, the control device 2440 may transmit a camera control signal and a laser measurer control signal to the measurement device 2410. According to one embodiment, the control device 2440 may perform a function of switching between multiple types of images by transmitting control signals to multiple types of cameras. According to another embodiment, the control device 2440 may transmit a control signal to the measurement device 2410 through the alarm and control server 2430. For example, in S2424, the control device 2440 may generate a control signal based on the controller's user input to the control device 2440 and transmit the generated control signal to the alarm and control server 2430. At S2426, the measurement device 2410 may operate and obtain new information according to the control signal.

FIG. 25 is a diagram illustrating a process for detecting a gas leak by linking an electronic device, an alarm device, and a server with each other, according to another embodiment.

According to another embodiment of the present disclosure, the gas leak detection system 10 may include an electronic device 1000, an alarm device 2520, and a server 2000. The processes shown in FIG. 25 may be performed by the gas leak detection system 10, but of course, they may also be performed by the electronic device 1000.

In S2502, the electronic device 10000 may acquire image information and laser measurement values. In S2504, the electronic device 1000 may transmit the acquired image information and laser measurement value to the server 2000. In S2506, the electronic device 1000 may use an artificial intelligence model to determine whether a target object is detected in the image information and analyze the type of the target object. In S2508, the electronic device 1000 may transmit the object detection result using the artificial intelligence model to the server 2000. Although not shown in FIG. 25, the server 2000 may output image information, laser measurement values, and object detection results transmitted from the electronic device 1000 to the display of the control device in real time.

In S2510, the electronic device 1000 may determine whether a predetermined warning event has occurred based on whether a target object in image information is detected using an artificial intelligence model. In S2512, the electronic device 1000 may transmit alarm information to the alarm device 2520 when an event is detected. Additionally, the electronic device 1000 may transmit alarm information and detection results to the server 2000 in S2514. According to one embodiment, in S2516, the server 2000 further transmits alarm information separate from that of the electronic device 1000 to the alarm device 2520, thereby securing operation reliability of the alarm device through multiple routes.

In S2518, the electronic device 1000 may determine the leak origin and coordinate information when detecting an object. In S2520, the electronic device 1000 may generate a leakage image related to the target object and generate a composite image by combining the generated leakage image with a panoramic image. At S2524, the server 2000 may output synthetic image information and detection results. In S2526, the server 2000 may transmit a control signal for controlling various components within the alarm device to the alarm device 2520. In S2528, the server 2000 may transmit a control signal for controlling components in the camera installation set to the electronic device 1000.

Figure 26 is a block diagram of an electronic device according to an embodiment.

Figure 27 is a block diagram of an electronic device according to another embodiment.

As shown in FIG. 26, the electronic device 1000 according to an embodiment may include a processor 1300, a network interface 1500, and a memory 1700. However, not all of the illustrated components are essential components. The electronic device 1000 may be implemented with more components than the illustrated components, or may be implemented with fewer components.

For example, as shown in FIG. 27, the electronic device 1000 includes a processor 1300, a network interface 1500, and a memory 1700, as well as a user input interface 1100, an output unit 1200, and a sensing unit. It may further include 1400, a network interface 1500, an A/V input unit 1600, and a memory 1700.

The user input interface 1100 refers to a means for a user to input data to control the electronic device 1000. For example, the user input interface 1100 includes a key pad, a dome switch, and a touch pad (contact capacitive type, pressure resistive type, infrared detection type, surface ultrasonic conduction type, There may be a mathematical tension measurement method, a piezo effect method, etc.), a jog wheel, a jog switch, etc., but it is not limited to these.

The user input interface 1100 may receive user input for controlling a camera installation set or enlarging or reducing a specific area in an image. According to another embodiment, the user input interface 1100 may, of course, obtain a user input to access an image for a specific time and place among stored composite images. The output unit 1200 may output an audio signal, a video signal, or a vibration signal, and the output unit 1200 may include a display unit 1210, an audio output unit 1220, and a vibration motor 1230. there is.

The display unit 1210 includes a screen for displaying and outputting information processed by the electronic device 1000. Additionally, the screen can display multiple types of images acquired from a camera installation set, leakage images, panoramic images, and composite images. For example, at least a portion of the screen may output a composite image of a space to be monitored, a user interface for controlling a camera installation set, etc. to closely monitor a target object appearing in the composite image.

The audio output unit 1220 outputs audio data received from the network interface 1500 or stored in the memory 1700. Additionally, the sound output unit 1220 outputs sound signals related to functions performed in the electronic device 1000 (eg, call signal reception sound, message reception sound, notification sound).

The processor 1300 typically controls the overall operation of the electronic device 1000. For example, the processor 1300 executes programs stored in the memory 1700, such as the user input interface 1100, the output unit 1200, the sensing unit 1400, the network interface 1500, and the A/V input unit. (1600) etc. can be controlled overall. Additionally, the processor 1300 may perform the functions of the electronic device 1000 shown in FIGS. 1 to 25 by executing programs stored in the memory 1700.

According to one embodiment, the processor 1300 acquires first type images of the space to be monitored from at least one first type camera connected to the electronic device by executing the one or more instructions. and acquire second type images of the monitoring target space from at least one second type camera connected to the electronic device, and when the first type images and the second type images are input, the first type images By inputting the first type images and the second type images into an object recognition model that identifies whether a target object in the images and the second type images is detected, the first type images and the second type images It is possible to identify whether the target object is detected within type images.

According to one embodiment, by executing the one or more instructions, the processor 1300 collects first type images of the space to be monitored from at least one first type camera connected to the electronic device and at least one camera connected to the electronic device. When second type images of the space to be monitored are acquired from a second type camera, and a target object is identified as being detected from at least one of the first type images or the second type images, Generating a leakage image by combining first type images about the detected target object and second type images corresponding to the first type images about the detected target object, and obtaining the generated leakage image in advance Based on the coordinates of the panoramic image, a composite image can be generated by combining the leaked image with the panoramic image.

According to one embodiment, the processor 1300 receives first type learning images of a space to be monitored from at least one first type of camera connected to the electronic device, and at least one second type of camera connected to the electronic device. Obtain second type learning images for the monitored space from, generate learning data and verification data based on the first type learning images and the second type learning images, and generate learning data and verification data based on the generated learning data. Thus, when first type images obtained from the first type of camera and second type images obtained from the second type of camera are input, the target object in the first type images and the second type images By training an object recognition model that identifies whether is detected, and inputting the first type images and the second type images into the learned object recognition model, the first type images and the second type images It is possible to identify whether or not the target object is detected.

In addition, according to one embodiment, when a target object is detected by executing one or more instructions, the processor 1300 determines the origin of the leak where the target object occurred, location information in physical space about the origin of the leak, and the type of the target object. Information about at least one item may be output on the screen along with the composite image.

The sensing unit 1400 may detect the state of the electronic device 1000 or the state surrounding the electronic device 1000 and transmit the sensed information to the processor 1300. The sensing unit 1400 can sense specification information of the electronic device 1000, temperature, humidity, and atmospheric pressure information for the space to be monitored.

For example, the sensing unit 1400 includes a geomagnetic sensor 1410, an acceleration sensor 1420, a temperature/humidity sensor 1430, an infrared sensor 1440, and a gyroscope sensor 1450. ), a location sensor (eg, GPS) 1460, an air pressure sensor 1470, a proximity sensor 1480, and an RGB sensor (illuminance sensor) 1490, but is not limited thereto. Since the function of each sensor can be intuitively deduced by a person skilled in the art from its name, detailed description will be omitted.

The network interface 1500 may include one or more components that allow the electronic device 1000 to communicate with other devices (not shown) and the server 2000. The other device (not shown) may be a computing device such as the electronic device 1000 or a sensing device, but is not limited thereto. For example, the network interface 1500 may include a wireless communication interface 1510, a wired communication interface 1520, and a mobile communication unit 1530. The wireless communication interface 1510 includes a short-range wireless communication unit, a Bluetooth communication unit, a Bluetooth Low Energy (BLE) communication unit, a Near Field Communication unit, a WLAN (Wi-Fi) communication unit, and a Zigbee communication unit. , may include, but is not limited to, an infrared data association (IrDA) communication unit, a Wi-Fi Direct (WFD) communication unit, and an ultra wideband (UWB) communication unit.

The wired communication interface 1520 may include at least one wired interface for exchanging data with an external device connected to the electronic device through wired communication. The mobile communication unit 1520 transmits and receives wireless signals with at least one of a base station, an external terminal, and a server on a mobile communication network. Here, the wireless signal may include various types of data according to voice call signals, video call signals, or text/multimedia message transmission and reception.

The A/V (Audio/Video) input unit 1600 is for inputting audio or video signals, and may include a camera 1610, a microphone 1620, etc. The camera 1610 can obtain image frames such as still images or videos through an image sensor in video call mode or shooting mode. Images captured through the image sensor may be processed through the processor 1300 or a separate image processing unit (not shown). According to one embodiment, the camera module 1610 may include an OGI camera that visualizes gas or smoke through infrared images and an EO camera that acquires visible light images by attaching a predetermined filter to the front part.

The microphone 1620 receives external sound signals and processes them into electrical voice data. For example, the microphone 1620 may receive an acoustic signal from an external device or a user. The microphone 1620 can receive a user's voice input. The microphone 1620 can use various noise removal algorithms to remove noise generated in the process of receiving an external acoustic signal.

The memory 1700 may store programs for processing and control of the processor 1300, and may also store data input to or output from the electronic device 1000. In addition, the memory 1700 includes at least one artificial intelligence model used by the electronic device 1000, image information about the monitoring target space acquired by the electronic device, laser measurement values, synthetic images, leakage images, and analysis results of the artificial intelligence model. Information about can be stored.

Additionally, the memory 1700 may store information about at least one neural network model used by the electronic device 1000. For example, the memory 1700 may store layers, nodes, and weight values related to connection strengths of the layers within at least one neural network model. Additionally, the electronic device 1000 may further store learning data generated by the electronic device 1000 in order to learn a neural network model. Additionally, the memory 1700 may further store information about the operating environment of cameras or servers connected to the electronic device.

The memory 1700 may be a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (for example, SD or XD memory, etc.), or RAM. (RAM, Random Access Memory) SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic disk , and may include at least one type of storage medium among optical disks.

Programs stored in the memory 1700 can be classified into a plurality of modules according to their functions, for example, a UI module 1710, a touch screen module 1720, a notification module 1730, etc. .

The UI module 1710 may provide specialized UI, GUI, etc. that are linked with the electronic device 1000 for each application for gas leak detection. The touch screen module 1720 may detect a user's touch gesture on the touch screen and transmit information about the touch gesture to the processor 1300. The touch screen module 1720 according to some embodiments can recognize and analyze touch codes. The touch screen module 1720 may be composed of separate hardware including a controller.

The notification module 1730 may generate a signal to notify the occurrence of an event in the electronic device 1000. For example, the electronic device 1000 may identify that a target object due to a gas leak has been detected and output warning content accordingly. Examples of events according to an embodiment include receiving a call signal, receiving a message, inputting a key signal, and schedule notification. The notification module 1730 may output a notification signal in the form of a video signal through the display unit 1210, may output a notification signal in the form of an audio signal through the sound output unit 1220, and the vibration motor 1230 A notification signal can also be output in the form of a vibration signal.

Figure 28 is a diagram for explaining the device configuration of a gas leak detection system according to another embodiment.

According to one embodiment, the gas leak detection system 10 may include a camera installation set 2810, a camera set 2820, and a control terminal 2830. According to one embodiment, the camera installation set 2810 may include a camera housing 2812, a pan/tilt 2814, and an elevation pole 2816. The gas leak detection system 10 transmits a control signal to the camera installation set 2810, remotely changes the pan/tilt value of the camera, or changes the elevation value of the pole, thereby providing image information and information about the desired monitoring target space. Laser measurement values can be obtained.

According to one embodiment, the elevation pole 2816 may be extended beyond a preset height value (e.g., 20 m), and the pan and tilt 2814 may be controlled to Pan 360 degrees and tilt to -30 degrees to 30 degrees. It is not limited to this. In addition, according to one embodiment, the camera housing 2812 may be provided to support a weight of 30KG or more, and the alarm sound of the alarm device connected to the control terminal according to the present disclosure may be provided to satisfy performance conditions of 90Db/1m or more. You can. According to one embodiment, a camera set 2820 may be provided inside the camera housing 2812. According to one embodiment, the camera set 2820 may include an OGI gas imaging camera 2822, a laser measuring device 2824, and an EO camera 2826. According to one embodiment, of course, all components of the camera 2820 may be included in the camera installation set 2810. The control terminal 2830 may include a plurality of computer assembly server 2832 devices and a computer assembly-specific monitor 2834 device.

Figure 29 is a block diagram of a server according to one embodiment.

According to one embodiment, the server 2000 may include a network interface 2100, a database 2200, and a processor 2300. The configuration of the server 2000 shown in FIG. 29 may correspond to the artificial intelligence-based gas leak detection server 2420 and the alarm and control server 2430 shown in FIG. 24, and the configuration of the control server 140 in FIG. 1 It can correspond to the configuration.

The network interface 2100 may correspond to the network interface (not shown) of the electronic device 1000 described above. For example, the network interface 2100 may obtain information about multiple types of image information, analysis results of an artificial intelligence model, leak origin, flow vector, leak image, and composite image from the electronic device 1000. According to another embodiment, the network interface 2100 allows an electronic device to receive information about a learned artificial intelligence model or information about a neural network model (e.g., weight values regarding layers and connection strengths between layers). You can. According to another embodiment, the network interface 2100 is information about the artificial intelligence model learned by the server, information about the layers of the artificial neural network and nodes included in the layers, or weights about the connection strength of the layers within the neural network. Values may also be transmitted to the electronic device 1000.

Additionally, according to one embodiment, the database 2200 may correspond to the memory described above in FIGS. 26 and 27. For example, the database 2200 may store image information acquired from the electronic device 1000, laser measurement values, analysis results of the artificial intelligence model, synthetic images, information about the artificial intelligence model itself, etc. According to one embodiment, the database 2200 is designed so that database information can be searched using only a combination of some keywords and words, and predetermined image data by date, time, and camera location can be searched and searched.

In addition, according to one embodiment, the database 2200 can provide information on images about the target object when an event occurs due to detection of a general image and a target object, and may use a separate video storage format for efficient capacity management. You can use it to store data. In addition, the database 2200 can manage existing stored video data and event occurrence data as a new table using join, and in order to maintain database design principles such as data integrity, the table is accessed in view format and the original Table access can be restricted. In addition, according to one embodiment, the database 2200 can manage code in a procedure format to facilitate maintenance/repair, process data based on the explain command for optimization when querying the database, and record data deletion details. Triggers for management can be applied.

According to one embodiment, the processor 2300 may control the overall operation of the server 2000. For example, the processor 2300 controls the network interface 2100 and the database 2200, thereby controlling all or all of the operations performed by the electronic device 1000 or the gas leak detection system 10 shown in FIGS. 1 to 28. At least some of them can be done together.

Figure 30 is a diagram illustrating a user interface screen provided by an electronic device for detecting an artificial intelligence-based gas leak according to an embodiment.

Referring to FIG. 30, an example of a user interface screen provided by an electronic device is shown. The user interface screen shown in FIG. 30 may be displayed on the display of a situation room system controlled not only by the electronic device 1000 but also by a control terminal or a control server connected to the electronic device. According to one embodiment, the electronic device 1000 includes a control interface 3004 for controlling the operation of multiple types of cameras and laser measuring devices in a camera installation set, and storing, managing, and storing image information processed by the electronic device 1000. A video control interface 3002 for calling can be output. The electronic device 1000 may access a specific image based on the controller's user input obtained through the image control interface 3002, or may output the accessed specific image on the screen.

According to one embodiment, the electronic device 1000 may additionally provide a magnification interface 3006, a focus interface 3008, and a sensor interface 3010 on the screen based on a user input for selecting the control interface 3004. It may be possible. According to one embodiment, the electronic device 1000 may acquire an enlarged image of a specific area within the space to be monitored based on a user input to the enlargement interface 3006. Additionally, the electronic device 1000 may perform a focusing function that automatically matches the focus to a pre-designated area based on a user input to the focus interface 3008. According to one embodiment, the electronic device 1000 may set an autofocusing area in advance for a specific area within the monitoring target space. According to one embodiment, the electronic device 1000 may change the operating environment (shutter cycle, information acquisition cycle, etc.) of a sensor located in the monitoring target space based on a user input to the center interface 3010.

Figure 31 is a diagram illustrating a user interface screen provided by an electronic device for detecting an artificial intelligence-based gas leak according to another embodiment.

Referring to FIG. 31, an example of a user interface screen provided by an electronic device is shown. According to one embodiment, the electronic device 10000 may provide an image control interface 3002 for storing, managing, and recalling image information, and the provided image control interface 3002 includes a video interface 3102. can do. For example, the electronic device 1000 uses an image interface 3104, an enhancement interface 3106, an automatic adjustment interface 3108, and a device for storing set conditions based on a user input for selecting the video interface 3102. The settings storage interface 3110 can be output.

According to one embodiment, the image interface 3104 may include a user interface for enlarging the image through digital zoom, adjusting the brightness value of the image, and adjusting the contrast value within the currently displayed image. According to one embodiment, the enhancement interface 3106 may include editing interfaces for improving the quality of video data, such as removing noise in an image and removing noise. According to one embodiment, the automatic adjustment interface 3108 may include an interface for editing video clips, and the settings storage interface 3110 may provide a function for saving changed values through the video interface 3102. there is.

The gas leak detection method, the synthetic image generation method for gas leak detection, and the method for identifying the shape of the target object according to the present disclosure are implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. You can. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and constructed for the present invention or may be known and usable by those skilled in the art of computer software.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements made by those skilled in the art using the basic concept of the present invention defined in the following claims are also possible. falls within the scope of rights.

Claims (15)

In a method for an artificial intelligence-based gas leak detection electronic device to identify the shape of a target object,
Obtaining first type images of a space to be monitored from at least one first type camera connected to the electronic device;
Obtaining second type images of the space to be monitored from at least one second type camera connected to the electronic device; and
When the first type images and the second type images are input, the first type images are input to an object recognition model that identifies whether a target object is detected in the first type images and the second type images. and identifying whether the target object is detected in the first type images and the second type images by inputting the second type images. Including,
The method of claim 1, wherein the target object includes at least one of a smoke object, a gas object, or a flame object.
The method of claim 1, wherein the method
When it is identified that the target object is detected from at least one of the first type images or the second type images, extracting shape information of the target object from images of the type in which the target object is detected; A method further comprising:
According to paragraph 2,
The first type of camera and the second type of camera are an EO camera and an OGI camera, respectively, and the first type images and the second type images are visible light images and infrared images, respectively. .
The method of claim 3, wherein the method
Based on the output value of the object recognition model, when the target object is identified as not being detected in the first type images and the second type images, transform in the frequency domain by Fourier transforming the second type images Generating image information;
Extracting spectral information for each of the target object and background interference materials around the target object from the converted image information;
When the extracted spectrum information is input, by inputting the extracted spectrum information into a likelihood ratio test model that outputs whether there is a frequency pattern according to the target object, the frequency pattern according to the target object in the extracted spectrum information is determined. identifying whether there is; and
When a frequency pattern according to the target object is identified, extracting shape information of the target object from the second type images based on an output value of the likelihood ratio test model; A method further comprising:
The method of claim 4, wherein the method
If the frequency pattern according to the target object is not identified based on the output value of the likelihood ratio test model, the extracted spectrum information is input to determine whether there is a frequency pattern according to the target object. Identifying whether there is a frequency pattern according to the target object in the extracted spectrum information by inputting it into an output support vector machine model; and
When a frequency pattern according to the target object is identified, extracting shape information of the target object from the second type images based on an output value of the support vector machine model; A method further comprising:
The method of claim 3, wherein the step of extracting the shape information
Preprocessing images of the type in which the target object is detected;
Generating masking data based on preset target object information;
separating target object areas generated by applying the masking data to images of a type in which the target object was detected from the preprocessed images; and
extracting shape information of the target object from the separated target object areas; Method, including.
The method of claim 6, wherein extracting shape information of the target object from the separated target object areas comprises:
identifying outline information about the target object within the separated target object area;
identifying center point information of a target object based on the identified outline information;
identifying a direction vector of the target object area based on the identified center point information and the outline information; and
extracting the identified outline information, center point information, and direction vector as shape information of the target object; Method, including.
The method of claim 6, wherein generating the masking data includes
determining a moving average value for frame images preceding a current frame of images of the type in which the target object is detected or for all frame images including the current frame images; and
generating the masking data based on the determined moving average value; Method, including.
The method of claim 8, wherein the method
Based on shape information of the separated target object areas, determining a motion vector between adjacent target object areas in images of the type in which the target object is detected;
modeling the separated target object regions based on the determined motion vector;
determining a sum vector of motion vectors representing the modeled target object areas; and
Based on the size and direction of the determined sum vector, determining a leak origin related to the location where the target object was created and a flow vector related to the flow direction and speed of the target object in the images in which the target object was detected. ; A method further comprising:
In an electronic device that identifies the shape of a target object for detecting artificial intelligence-based gas leaks,
network interface;
A memory that stores one or more instructions; and
At least one processor executing the one or more instructions; Including,
The at least one processor executes the one or more instructions,
Obtaining first type images of a space to be monitored from at least one first type camera connected to the electronic device,
Obtaining second type images of the monitored space from at least one second type camera connected to the electronic device,
When the first type images and the second type images are input, the first type images are input to an object recognition model that identifies whether a target object is detected in the first type images and the second type images. And by inputting the second type images, identify whether the target object is detected in the first type images and the second type images,
The electronic device, wherein the target object includes at least one of a smoke object, a gas object, or a flame object.
The method of claim 10, wherein the at least one processor
When it is identified that the target object is detected from at least one of the first type images or the second type images, extracting shape information of the target object from images of the type in which the target object was detected, electronic Device.
According to clause 11,
The first type camera and the second type camera are an EO camera and an OGI camera, respectively, and the first type images and the second type images are visible light images and infrared images, respectively. Device.
13. The method of claim 12, wherein the at least one processor
Based on the output value of the object recognition model, when the target object is identified as not being detected in the first type images and the second type images, transform in the frequency domain by Fourier transforming the second type images Generate video information,
Extracting spectral information for each of the target object and background interference materials around the target object from the converted image information,
When the extracted spectrum information is input, by inputting the extracted spectrum information into a likelihood ratio test model that outputs whether there is a frequency pattern according to the target object, the frequency pattern according to the target object in the extracted spectrum information is determined. identify whether there is a
When a frequency pattern according to the target object is identified, the electronic device extracts shape information of the target object from the second type images based on the output value of the likelihood ratio test model.
14. The method of claim 13, wherein the at least one processor
If the frequency pattern according to the target object is not identified based on the output value of the likelihood ratio test model, the extracted spectrum information is input to determine whether there is a frequency pattern according to the target object. Identify whether there is a frequency pattern according to the target object in the extracted spectrum information by inputting it into the output support vector machine model,
When a frequency pattern according to the target object is identified, the electronic device extracts shape information of the target object from the second type images based on the output value of the support vector machine model.
In a method for an artificial intelligence-based gas leak detection electronic device to identify the shape of a target object,
Obtaining first type images of a space to be monitored from at least one first type camera connected to the electronic device;
Obtaining second type images of the space to be monitored from at least one second type camera connected to the electronic device; and
When the first type images and the second type images are input, the first type images are input to an object recognition model that identifies whether a target object is detected in the first type images and the second type images. and identifying whether the target object is detected in the first type images and the second type images by inputting the second type images. Including,
A computer-readable recording medium storing a program for performing a method, wherein the target object includes at least one of a smoke object, a gas object, and a flame object.