Fire point detection method, system, apparatus, device, and storage medium

Abstract

This application provides a fire detection method, system, apparatus, device, and storage medium. The fire detection method includes: acquiring raw image data of a region to be detected, wherein the raw image data is acquired by a first acquisition device through dual narrowband light transmission elements; extracting the response values ​​of the R channel and B channel of each pixel in the raw image data; identifying pixels whose R channel response values ​​and B channel response values ​​satisfy a set recognition condition as valid fire point pixels; and determining that a fire point exists in the region to be detected if the number of valid fire point pixels in the raw image data exceeds a preset first threshold. This application improves the accuracy of fire detection.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a fire detection method, system, device, equipment, and storage medium. Background Technology

[0002] Fire is one of the major disasters causing casualties and property damage. Early detection and warning of fires through fire point detection can help control fires and reduce losses.

[0003] In related technologies, fire detection is performed by training a neural network to identify flames in video streams captured by a visible light camera. However, this method has insufficient generalization ability in complex lighting environments, making it difficult to distinguish between real flames and highly similar interfering light sources, resulting in insufficient accuracy. Summary of the Invention

[0004] This application provides a fire detection method, system, apparatus, device, and storage medium that can improve the accuracy of fire detection.

[0005] The technical solution of this application embodiment is implemented as follows: This application provides a method for detecting fire points, the method comprising: Acquire raw image data of the area to be detected, wherein the raw image data is acquired by a first acquisition device through a dual narrowband light transmission element, the dual narrowband light transmission element being used to filter out energy of bands other than the first specific band and the second specific band; Extract the response value of the R channel and the response value of the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. Pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions are identified as valid fire point pixels. The number of valid fire point pixels in the original image data is counted. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the detection area.

[0006] This application provides a fire detection system, including a first acquisition device and an image processing device; The first acquisition device is used to acquire raw image data of the area to be detected while transmitting light through dual narrowband light transmission elements, wherein the dual narrowband light transmission elements are used to filter out energy of bands other than the first specific band and the second specific band. The image processing device is used to extract the response value of the R channel and the response value of the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. Pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions are identified as valid fire point pixels. The number of valid fire point pixels in the original image data is counted. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the detection area.

[0007] This application provides a fire detection device, including: The original image acquisition module is used to acquire the original image data of the area to be detected. The original image data is acquired by the first acquisition device through a dual narrowband light transmission element. The dual narrowband light transmission element is used to filter out the energy of bands other than the first specific band and the second specific band. The response value extraction module is used to extract the response value of the R channel and the response value of the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. The effective fire point pixel determination module is used to identify pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions as effective fire point pixels. The fire point confirmation module is used to count the number of valid fire point pixels in the original image data. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the area to be detected.

[0008] This application provides an electronic device, the electronic device comprising: Memory is used to store executable instructions or computer programs. The processor, when executing computer-executable instructions or computer programs stored in the memory, implements the fire detection method provided in the embodiments of this application.

[0009] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions for implementing the fire detection method provided in this application when executed by a processor.

[0010] This application provides a computer program product, including a computer program or computer executable instructions. When the computer program or computer executable instructions are executed by a processor, they implement the fire detection method provided in this application.

[0011] The embodiments of this application have the following beneficial effects: By extracting the R-channel and B-channel response values ​​of each pixel from the raw image data obtained through dual narrowband light transmission elements, and utilizing the different effects of different types of light sources on the R-channel and B-channel to set the recognition conditions, the independent energies of the first specific band and the second specific band are decoupled from the raw image according to the recognition conditions. The effective fire point pixels are determined, and the presence of a fire point is judged based on the number of effective fire point pixels. By utilizing the different spectral characteristics of real flames in different bands, the flames are distinguished from other interfering light sources from a physical property perspective, thus improving the accuracy of fire point detection. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the fire detection system architecture provided in the embodiments of this application; Figure 2 This is a schematic diagram of the fire detection device provided in the embodiments of this application; Figure 3 This is a schematic flowchart of the fire detection method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the Bayer array on the photosensitive surface of the first acquisition device provided in this application embodiment; Figure 5 This is a schematic diagram illustrating the quantum efficiency characteristics of the first acquisition device provided in this application embodiment; Figure 6 This is a schematic diagram of the wavelength distribution of different interfering light emission provided in the embodiments of this application; Figure 7 This is a comparative schematic diagram of solar spectrum and flame spectrum provided in an embodiment of this application; Figure 8 This is a flowchart of a fire detection method provided in a specific application scenario according to an embodiment of this application.

[0013] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0016] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0017] Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit this application.

[0018] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0019] 1) Raw image format (RAW) refers to a collection of digital signals directly generated by an image sensor (such as CMOS or CCD) in an acquisition device through a photoelectric conversion process, without any nonlinear image signal processing (ISP) or with only minimal preprocessing. Its essential property is a high-bit-depth digital matrix (pixels), where each value (DN value) in the matrix has a linear mapping relationship with the number of photons received by the corresponding sub-pixel of the sensor.

[0020] 2) A dual narrow-band optical element refers to a precision optical device with specific spectral transmission characteristics. Its function is to selectively allow light from two discrete and narrow wavelength ranges to pass through, while efficiently blocking or absorbing light from wavelength ranges outside these two discrete and narrow wavelength ranges. In the embodiments of this application, this element is typically placed at the front end of the photosensitive path of the acquisition device to physically filter out energy from ambient light other than a first specific wavelength band and a second specific wavelength band. For example, in some embodiments of this disclosure, the dual narrow-band optical element can be specifically manifested as a dual-peak narrow-band filter, with its transmission peak center wavelengths set to 940nm (near-infrared band) and 450nm (visible blue light band), respectively, thereby achieving targeted extraction of the flame characteristic spectrum and physical-level suppression of background noise.

[0021] 3) A fixed camera (or bullet camera) refers to a video camera (acquisition device) with a fixed field of view (FOV) and a fixed detection direction. It is typically equipped with a fixed-focus or zoom lens, but under normal operating conditions, the shooting angle and focal length remain constant. In this embodiment, the fixed camera is used for wide-area detection, undertaking the task of all-weather real-time scanning and preliminary detection of a large field of view. Its main technical function is to utilize its wide field of view coverage to quickly identify suspected abnormal heat sources or light spots in the scene and generate preliminary detection results containing suspected target location information, providing guiding coordinates for subsequent high-precision verification.

[0022] 4) A PTZ camera, also known as a pan-tilt-zoom camera, is a video camera (acquisition device) that integrates a fully rotating pan-tilt head and an optical zoom lens. Its core feature is its ability to rotate horizontally (Pan), tilt vertically (Tilt), and zoom, allowing it to quickly change its field of view and adjust its focal length according to control commands to obtain detailed images of distant targets. In this embodiment, the PTZ camera is used for high-precision verification. Responding to the guiding spatial coordinates generated by the camera module, it automatically rotates and aligns with the suspected fire point area for optical magnification. Within this magnified field of view, dual narrowband light-transmitting elements are applied for spectral analysis, thereby achieving accurate authentication of minute fire points and eliminating false alarms.

[0023] 5) A Bayer array, also known as a Bayer filter array, refers to a color filter arrangement structure disposed on the surface of the photosensitive element of a single-chip digital image sensor (image sensor in an acquisition device). Its typical arrangement is a 2x2 sub-matrix cyclically covering RGGB (red-green-green-blue), ensuring that each sub-pixel in a pixel can only sense one of the three primary colors: red (R), green (G), or blue (B). In this embodiment, the Bayer array is the physical carrier structure of the raw image data. By analyzing the Bayer array, the system can separate the R-channel and B-channel components from a single RAW image, and then utilize the differences in spectral response of the sensor in the acquisition device at different wavelengths (such as infrared leakage characteristics) to analyze the spectral energy data of the first and second specific wavelength bands.

[0024] 6) The spectral response characteristic curve is a functional relationship describing the quantum efficiency (QE) or response sensitivity of the photosensitive element surface corresponding to each sub-pixel in an image sensor under illumination of different wavelengths of light. This curve reflects the photoelectric conversion capability of each color channel (R, G, B) of the sensor in the acquisition device for photons of different wavelengths. In the embodiments of this application, the spectral response characteristic curve is the theoretical basis for constructing the band decoupling algorithm. Specifically, by utilizing the characteristic that both the R and B channels have high response in the near-infrared band (e.g., 940nm) (i.e., infrared leakage), while only the B channel has a response in the short-wave visible light band (e.g., 450nm), a mathematical model can be constructed to separate the energy of different bands through channel subtraction operations (e.g., BR).

[0025] 7) Confidence Level: In target detection algorithms based on statistics or deep learning, confidence level is a quantitative indicator of the probability or certainty that a detected target belongs to a specific category (such as "flame"). Its value is typically normalized to between 0 and 1. In the detection process of this application, confidence level is used as a threshold parameter for multi-level discrimination. For example, in the initial detection stage of the bullet camera, the system sets a lower confidence level threshold to improve recall and ensure that no suspected targets are missed; while in the verification stage of the PTZ camera, combined with spectral analysis results, the system requires targets to meet a higher confidence level standard to improve precision and effectively filter out false alarms.

[0026] In related technologies, fire detection methods mainly include pure vision methods based on deep learning, sensor-based methods, and thermal imaging-based methods. Pure vision methods based on deep learning use ordinary visible light cameras and trained neural network models to identify flames in real-time video streams, relying on visual features such as texture, shape, and color to distinguish flames from the environmental background. Sensor-based methods utilize infrared, ultraviolet, or thermal sensors to detect the unique radiation or temperature characteristics of flames, achieving fire detection based on these detected characteristics. Thermal imaging-based methods use infrared thermal imaging equipment to generate thermal images, detecting flames by identifying the infrared thermal radiation distribution of the target.

[0027] However, purely vision-based methods lack generalization ability in complex lighting environments, struggling to distinguish real flames from highly similar interference sources (such as lights, reflected light, red clothing, etc.) based on physical properties, especially in nighttime or backlit scenes where false alarms are frequent. Sensor-based methods are greatly affected by the environment, prone to false alarms, and have insufficient effective detection distance, making it impossible to accurately locate the fire point. Thermal imaging-based methods rely on ambient temperature differences, making detection difficult when the temperature difference between the flame and the environment is less than a certain value, and the equipment cost is high. These issues make it difficult for related technologies to simultaneously meet the application requirements of high accuracy, low false alarm rate, and low cost when facing complex scenarios (such as electric vehicle charging sheds, factory workshops, and underground parking lots).

[0028] This application provides a fire detection method, system, apparatus, device, and storage medium, which can improve the accuracy of fire detection. The exemplary application of the fire detection device provided in this application is described below. The device provided in this application can be implemented as various types of terminals such as laptops, tablets, desktop computers, smartphones, vehicle terminals, and image acquisition devices, or it can be implemented as a server. The exemplary application when the device is implemented as a server will be described below.

[0029] See Figure 1 , Figure 1This is a schematic diagram of the architecture of the fire detection system 100 provided in this application embodiment. To perform fire detection operations, a fire detection application can be provided, such as an application specifically for fire detection. The fire detection system 100 in this application embodiment includes at least: a first acquisition device 400, a network 300, and an image processing device 200. The first acquisition device 400 is connected to the image processing device 200 via the network 300, which can be a wide area network (WAN), a local area network (LAN), or a combination of both. The image processing device 200 can be implemented as a server or a terminal. In some embodiments, the fire detection system 100 may further include at least one second acquisition device 500, which is connected to the image processing device 200 via the network 300. In the case of multiple second acquisition devices 500, the different second acquisition devices 500 face different directions. The first acquisition device 400 and the second acquisition device 500 can be connected via a controller (…). Figure 1 (Not shown in the image) is used for control. In some embodiments, the first acquisition device 400 and the second acquisition device 500 can be an integrated structure, that is, the first acquisition device 400 and the second acquisition device 500 are integrated into the same camera housing.

[0030] See Figure 2 , Figure 2 This is a schematic diagram of the structure of the image processing device provided in the embodiments of this application. Figure 2 The image processing device shown includes at least one processor 210, a memory 250, at least one network interface 220, and a user interface 230. The various components in the image processing device 200 are coupled together via a bus system 240. It is understood that the bus system 240 is used to implement communication between these components. In addition to a data bus, the bus system 240 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 240.

[0031] User interface 230 includes one or more output devices 231 and one or more input devices 232 that enable the presentation of media content.

[0032] The memory 250 can be a solid-state storage device, a hard disk drive, an optical disk drive, etc.

[0033] In some embodiments, memory 250 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0034] Operating system 251 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks; The network communication module 252 is used to reach other electronic devices such as Bluetooth, WiFi, and Universal Serial Bus (USB) via one or more (wired or wireless) network interfaces 220. Presentation module 253 is configured to enable the presentation of information (e.g., a user interface for operating peripheral devices and displaying content and information) via one or more output devices 231 associated with user interface 230 (e.g., a display screen, a speaker, etc.). The input processing module 254 is used to detect and translate one or more user inputs or interactions from one or more input devices 232.

[0035] In some embodiments, the apparatus provided in this application can be implemented in software. Figure 2 A fire detection device 255 stored in memory 250 is shown. This device can be software in the form of programs and plug-ins, and includes the following software modules: a raw image acquisition module 2551, a response value extraction module 2552, a valid fire pixel determination module 2553, and a fire confirmation module 2554. These modules are logically linked and can therefore be arbitrarily combined or further separated according to their implemented functions. The functions of each module will be described below.

[0036] See Figure 3 , Figure 3 This is a flowchart illustrating the fire detection method provided in the embodiments of this application, which will be combined with... Figure 3 The steps shown are explained as follows: Figure 3 As shown, the fire detection method is illustrated using an image processing device as the executing entity. The method includes the following steps 101 to 104.

[0037] In step 101, the original image data of the region to be detected is obtained.

[0038] The original image data is acquired by the first acquisition device through a dual narrowband light transmission element, which is used to filter out the energy of bands other than the first specific band and the second specific band.

[0039] Here, a dual narrowband light-transmitting element refers to an optical filter with specific spectral transmission characteristics, which allows light to pass through only two discrete, narrow wavelength ranges while blocking or absorbing light of other wavelengths. In the embodiments of this application, the first specific band and the second specific band can correspond to the near-infrared and blue light bands, which are characterized by significant flame features, respectively.

[0040] In some embodiments, raw image data can be acquired by mounting a custom dual-bandpass filter in front of the lens of a visible light camera or in front of a photosensitive element (Sensor) of a first device. The first acquisition device can be a CMOS or CCD image sensor, outputting RAW format data without image signal processing (ISP) to preserve the original spectral energy information.

[0041] It's important to note that raw image data (RAW format data) is the original, unprocessed image data captured by the first acquisition device. RAW format data directly records the intensity of the photoelectric signal (energy) received by each photosensitive unit (pixel) of the image sensor during exposure (the response value and energy intensity are linearly related). A visualized image is obtained by performing specific image signal processing (ISP, typically including depixelation, white balance, color correction, sharpening, and noise reduction) on the RAW format data. Furthermore, to accommodate the human eye's sensitivity to dark areas, gamma correction is also performed, so the pixel response value of the visualized image is no longer linearly related to the energy of the photoelectric signal. See also... Figure 4 The first acquisition device can be, for example, a PTZ camera. The photosensitive element surface of the sensor in this first acquisition device is covered with red (R), green (G), and blue (B) filters. These filters are arranged in a Bayer array (typically 2×2). Through each filter, the sensor can acquire the response value of a sub-pixel (channel) in RAW format data. That is, each pixel in RAW format data is typically composed of 2×2 sub-pixels (channels). Each sub-pixel responds according to the energy of the photoelectric signal of its corresponding channel. For example, the R channel sub-pixel (the sub-pixel corresponding to the sensor photosensitive element surface covered by the red filter) typically only responds to photoelectric signals within a specific range of the red band (including the near-infrared band and the red visible light band), while the B channel sub-pixel (the sub-pixel corresponding to the sensor photosensitive element surface covered by the blue filter) responds to photoelectric signals (each pixel in the visualized image has complete values ​​for the R, G, and B channels).

[0042] In some embodiments, acquiring raw image data of the area to be detected can be achieved by setting the center wavelength of the first specific band of the dual narrowband light-transmitting element to 940 nm (near-infrared band) and the center wavelength of the second specific band to 450 nm (blue light band). When light passes through the dual narrowband light-transmitting element, only the light energy near 940 nm (e.g., ±20 nm) and near 450 nm (e.g., ±20 nm) can reach the photosensitive surface of the sensor in the first acquisition device; other visible light (such as green and red light) and other infrared light are filtered out. The sensor converts the received light signal into a digital signal to generate a RAW image containing scene information of the area to be detected.

[0043] In some embodiments, prior to step 101, preliminary fire detection may be performed by executing steps 105 to 107.

[0044] In step 105, a wide-angle image including the area to be detected is acquired, wherein the wide-angle image is acquired by at least one second acquisition device, and the field of view coverage of the second acquisition device is greater than that of the first acquisition device.

[0045] Here, a wide-angle image refers to a panoramic detection image with a large field of view (FOV), typically covering a horizontal field of view of 90 degrees or greater, and can be an RGB image. The second acquisition device can be a fixed-focus camera (bullet camera) for global detection, whose main responsibility is to identify scene images over a large area. Fire detection can be performed in collaboration with the first acquisition device using one or more second acquisition devices. When multiple second acquisition devices are used, they can be positioned at multiple locations within the area of ​​interest to form blind-spot-free coverage. Conversely, the first acquisition device can be a zoom camera (PTZ camera) for local detail observation. Field of view coverage refers to the physical spatial angle that the sensor can observe at the current focal length. In some embodiments, the first and second acquisition devices can be integrated into a single structure. For example, the upper part of the integrated structure can be configured as a second acquisition device containing a fixed-focus lens for preliminary detection and initial screening of fire points in the wide-angle field of view. The lower part of the integrated structure can be configured as a first acquisition device with a rotatable pan-tilt head and an optical zoom lens for local magnification of suspected fire points and spectral detection based on dual narrowband light transmission elements. In this integrated structure, the second acquisition device and the first acquisition device are integrated into the same housing, forming a one-to-one hardware linkage. Since their spatial relative positions are fixed, the physical coordinate system can be accurately calibrated during the device's manufacture or initial installation, thereby enabling extremely low latency and high-precision spatial coordinate guidance and alignment.

[0046] In some embodiments, wide-angle images can be acquired by continuously capturing a full-color video stream of the region of interest at a preset frame rate (e.g., 25fps) using a wide-angle fixed-focus camera mounted high above the detection point. As an example, a wide-angle fixed-focus camera is deployed within the region of interest. There are two secondary acquisition devices, each responsible for detecting a specific sector. When N≥2, the fields of view of all secondary acquisition devices can be stitched together or overlapped to cover the entire area of ​​interest. The primary acquisition device serves as a shared high-precision confirmation resource, either in standby mode or performing a patrol mission. Each secondary acquisition device independently and in parallel acquires the real-time video stream of its respective responsible area.

[0047] In step 106, the first suspected fire point area in the wide-angle image is identified by the image recognition model according to the set first confidence level, and the spatial coordinates of the first suspected fire point area are determined.

[0048] Here, the image recognition model refers to a deep learning neural network trained on a large number of flame samples, such as YOLOv5, SSD, or Faster R-CNN object detection models. The first confidence score is the probability threshold at which the model determines the presence of a flame within a certain detection box in a wide-angle image. The first suspected fire region refers to the rectangular region (bounding box) in the wide-angle image selected by the model that possesses visual characteristics of a flame (such as color, texture, and shape). Spatial coordinates typically refer to the pixel center position of this region in the image coordinate system. Or, after calibration, the corresponding horizontal azimuth (Pan) and vertical pitch (Tilt) of the gimbal.

[0049] In some embodiments, the first confidence level is set to a value that is lower than the subsequent second confidence level (e.g., 0.15). The purpose of this setting is to achieve a high recall strategy and avoid missing the first suspected fire spot area, thereby ensuring that all possible fire spot targets are identified during the large-scale search phase, including those distant small fire sources or concealed fire sources that are obscured.

[0050] In some embodiments, the spatial coordinates of the first suspected fire point region can be determined by inputting a wide-angle image into a trained YOLO model, which then outputs several detection boxes. The corresponding confidence levels were then used to identify all bounding boxes with a confidence level greater than 0.15 as the first suspected fire point regions. Subsequently, the center point of each bounding box was calculated. The physical direction vector of the target is calculated by taking into account the offset relative to the image center and the installation parameters of the second acquisition device.

[0051] As an example, local image coordinates can be mapped using a homography matrix. Convert to azimuth in a unified world coordinate system.

[0052] Set one homography matrix , is used to describe the mapping relationship from the pixel coordinate system of the second acquisition device to the angular coordinate system of the first acquisition device, as shown in formula (1): Formula (1); in, : Indicates the first suspected fire point area is in the The x and y coordinates of pixels in the image from the second acquisition device. : Represents the homography matrix obtained through pre-calibration. The elements are in the matrix. This matrix is ​​usually calculated by selecting at least four corresponding point pairs during the installation phase (i.e., selecting a point in the camera's view, manually rotating the PTZ camera to align with the same point, and recording the pixel coordinates and angle coordinates). : Indicates the horizontal azimuth angle (Pan) that the first acquisition device (PTZ camera) needs to rotate to. : Indicates the vertical pitch angle (Tilt) that the first acquisition device (PTZ camera) needs to rotate to. This represents the scaling factor (a non-zero constant) in the homogeneous coordinate system.

[0053] In step 107, the orientation of the first acquisition device is controlled to align with the position corresponding to the spatial coordinates, and the first suspected fire point area is taken as the area to be detected.

[0054] Here, controlling the orientation of the first acquisition device refers to scheduling the first acquisition device (PTZ camera) to the corresponding orientation (position corresponding to the spatial coordinates) based on the received fire point coordinates (spatial coordinates) request. That is, multiple low-cost wide-angle bullet cameras are responsible for large-area initial screening, and once a suspected target is found, high-precision PTC camera resources are immediately called up for verification, realizing a collaborative working mode between bullet cameras and PTC cameras.

[0055] In some embodiments, controlling the first acquisition device can be achieved by establishing a task scheduling queue. When multiple second acquisition devices simultaneously or sequentially report the spatial coordinates of the first suspected fire point area, the requests are sorted according to a preset priority strategy (such as first-come, first-served, distance-priority, or confidence-priority). The first acquisition device responds to the requests in the queue sequentially, based on the spatial coordinates of the target. Calculate its own rotational parameters and zoom magnification It quickly rotates and zooms in on the target area to complete the alignment and locking of the area to be detected.

[0056] By substituting the center coordinates of the detection frame of the first suspected fire point area into formula (1), the coordinate values ​​of the angle coordinate system of the first acquisition device can be obtained. Based on the calculated coordinate values ​​of the angle coordinate system of the first acquisition device, a control command such as PTZ_Control(Pan=88,Tilt=48) is generated. This allows the first acquisition device to rotate the pan-tilt unit horizontally to the corresponding angle after receiving the control command, thereby accurately aligning it with the first suspected fire point area identified in step 107.

[0057] In this embodiment, the second acquisition device can be mounted on a two-dimensional electric pan-tilt unit, or a motor drive mechanism can be integrated inside the second acquisition device, and the orientation of the second acquisition device can be controlled by a controller through a bus or network interface.

[0058] This application embodiment utilizes a second acquisition device with a large field of view (such as a wide-angle bullet camera) for all-time, wide-area initial screening, enabling rapid identification of suspected abnormal targets within the field of view. Meanwhile, a first acquisition device with a variable field of view and high positioning accuracy (such as a zoom PTZ camera) is used for targeted, detailed confirmation. This collaborative approach effectively resolves the contradiction between a single sensor's inability to simultaneously achieve "wide field of view coverage" and "recognition of small, distant fire points." Compared to relying solely on a single large field-of-view sensor, it achieves rapid detection and location of weak, distant fire sources with lower computational costs. Furthermore, the use of an image recognition model with a low confidence threshold in the wide-area detection stage (the stage detected by the second acquisition device) reflects a "high recall" design strategy. This strategy allows for the retention of more suspected targets (including partially obscured or extremely weak fire sources) in the initial screening stage, which are then eliminated through high-magnification zoom and dual-band physical attribute verification by the first acquisition device. This ensures a reduced false alarm rate while controlling the false alarm rate through high-precision verification in subsequent steps, achieving overall optimization of detection performance.

[0059] In some embodiments, the first acquisition device typically includes a light-transmitting element. After controlling the orientation of the first acquisition device in step 107 to align with the position corresponding to the spatial coordinates, steps 108 to 1010 can also be executed.

[0060] In step 108, the first suspected fire point image is acquired.

[0061] The first suspected fire point image is obtained by the first acquisition device by magnifying the first suspected fire point area to a preset image size in the field of view of the first acquisition device after aligning it with the position corresponding to the spatial coordinates, and then using a full-transmission light element. The first suspected fire point image is an image including the first suspected fire point area.

[0062] Here, the preset image size refers to the pixel ratio or absolute pixel (box) size of the target (the complete image of the first suspected fire point) in the frame, aiming to acquire the fine texture features of the target through optical zoom. The full-light-passing element here typically refers to a conventional infrared cut-off filter (IR-Cut), and the first acquisition device acquires a conventional full-color image.

[0063] In some embodiments, the process of acquiring the first suspected fire point image is as follows: the controller drives the zoom motor inside the first acquisition device to zoom in according to a set zoom level. For example, the system sets that the width of the target area in the image must be more than 1 / 4 of the total image width. Through optical magnification, a tiny fire source that originally occupied only a few pixels in a wide-angle image now becomes a detailed image containing tens of thousands of pixels, with a clear outline of flame movement and color gradient.

[0064] In step 109, the second suspected fire point region in the first suspected fire point image is identified by the image recognition model according to the set second confidence level, wherein the second confidence level is higher than the first confidence level.

[0065] Here, the second confidence level is a probability threshold used for fine-tuning. Since the image has been optically magnified, the target features are more prominent than in a wide-angle image (such as the sharp corners of a flame, dynamic movement, core brightness distribution, etc.), so a higher threshold can be set to exclude obvious visual interference.

[0066] In some embodiments, the first confidence level (initial screening) is set to 0.15, and the second confidence level (verification) is set to 0.5. This gradient setting maintains high sensitivity during the "wide-area search phase" and transitions to high rigor during the "zoom confirmation phase." If the model's output confidence level still cannot reach 0.5 in the magnified high-definition image, the system determines the target as a false alarm (such as a streetlight, a distant reflective strip, etc.), and the process terminates, thus avoiding unnecessary filter switching.

[0067] In step 1010, if a second suspected fire spot area is identified, the second suspected fire spot area is taken as the area to be detected, and the step of acquiring the original image data of the area to be detected is triggered.

[0068] Here, "triggered" refers to the formal initiation of the verification process based on spectral physical properties. The second suspected fire spot area has now been identified as a new area to be detected.

[0069] In some embodiments, the triggering mechanism includes: the controller issuing a command to drive the filter switch of the first acquisition device to move out the full-transmission element, switch to the dual narrow-band transmission element, and enter RAW data acquisition mode. Since the precise coordinates of the second suspected fire point region have been locked at this time, the spectral ratio calculations in subsequent steps 101 to 104 will only be performed on the ROI region or its neighborhood, which reduces the computational load and significantly improves the real-time performance of the detection.

[0070] This application's embodiment employs a tiered discrimination logic of "low-threshold initial screening (wide-angle) + high-threshold verification (zoom)," ensuring a high capture rate (recall rate) for fire hazards over a wide area while rigorous secondary recognition using high-definition images eliminates most visual false alarms. This optimizes the system's recognition accuracy without sacrificing detection speed. Furthermore, by adding a "high-confidence full-color verification" step before switching the dual narrow-band light-transmitting elements, the filter switcher is only triggered when visual features are highly suggestive of a fire. This avoids ineffective mechanical switcher operation due to frequent false alarms during the wide-angle phase, reduces mechanical wear on precision optical components, and improves the stability and durability of the equipment under long-term detection conditions.

[0071] In some embodiments, the first acquisition device further includes a switcher, and step 1011 may be performed before step 101.

[0072] In step 1011, if a second suspected fire point area is identified, the second suspected fire point area is taken as the area to be detected, and the full-transmission light element set in the first acquisition device is switched to a dual narrow-band light element by a switcher.

[0073] Here, the switcher can be a precision mechanical drive device mounted at the rear of the lens of the first acquisition device and driven by a controller. It may contain a slide rail or turntable structure. The switcher carries two optical elements: a full-transmission element for normal imaging and a dual narrowband transmission element for spectral analysis.

[0074] As an example, in step 1010, if a second suspected fire point area is identified, the controller can send a pulse signal to the switch's drive circuit. The switch's electromagnetic coil or micro stepper motor is controlled to move away from the full-transmitting element in the optical path within a very short time (e.g., within 100ms to 300ms), and precisely push the dual narrow-band transmitting element into the center position of the optical axis. Simultaneously with the filter switching, the controller can synchronously adjust the gain and exposure time of the first acquisition device and switch to RAW data output mode to compensate for the decrease in light transmittance caused by the dual narrow-band filters. Using the second suspected fire point area as the region of interest (ROI) has a spatial guiding effect. After the switching is completed, the first acquisition device no longer scans the entire image, but instead concentrates resources on high-frequency sampling of the photosensitive unit where the region of interest is located, acquiring the raw image data required in subsequent step 101.

[0075] This embodiment of the application alters the physical characteristics of the optical sensing system through physical means (switching the light-transmitting element via a switcher). This mode switching allows the first acquisition device to switch from acquiring "full-color visual features" to acquiring "dual-band energy features." This ensures that the system can not only identify the "shape and color" of the flame but also further detect its specific "spectral radiation ratio," thereby distinguishing the real fire point (effective fire point pixels) from visual interference and sunlight interference. By removing the full-transmitting element through the physical switcher, interference from non-target band light to the sensor is physically blocked. This ensures that the RAW data acquired in subsequent step 101 consists entirely of the two characteristic bands of 940nm and 450nm, eliminating visible light background noise from an optical physics perspective and providing an interference-free data source for subsequent high-precision energy ratio calculations such as R / (BR).

[0076] In step 102, the response values ​​of the R channel and the B channel of each pixel in the original image data are extracted.

[0077] The response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band.

[0078] Here, the channel response value refers to the digital brightness value (DN value) output by the pixel of the corresponding color channel in the image sensor after receiving light, and its magnitude is linearly related to the energy intensity of the incident light.

[0079] In some embodiments, the response values ​​of the R channel and B channel can be extracted by parsing the Bayer array of the RAW image data, directly reading the R channel (sub-pixels covered by the red filter) value as the R channel response value, and reading the B channel (sub-pixels covered by the blue filter) value as the B channel response value. However, see also... Figure 5 Due to the characteristics of quantum efficiency of sensors, in the 940nm band, the filters of the R, G and B channels usually have infrared leakage (i.e., transparent to infrared light), while in the 450nm band, only the B channel has a response (the response values ​​of the R and G channels are both below 10%). Therefore, even if the response value of the B channel is analyzed, only the superposition energy of the energy of the first specific band and the energy of the second specific band can be obtained.

[0080] In some embodiments, the extraction of the R channel response value and B channel response value of each pixel in the original image data in step 102 can be achieved through the following step 1021.

[0081] In step 1021, for each pixel, the original image data is analyzed based on the spectral response characteristic curve of the first acquisition device to obtain the response values ​​of the R channel and the B channel.

[0082] Here, the spectral response curve (Quantum Efficiency Curve, QECurve) of the first acquisition device refers to the graph provided by the sensor manufacturer, which describes the photosensitivity of each RGB color channel of the sensor to different wavelengths of light, and is used to represent, for example, Figure 5 The visualization of the sensor's quantum efficiency characteristics shows the quantum efficiency (QE) at different wavelengths in the spectral response curve. For common silicon-based image sensors (CMOS or CCD), their physical characteristics are typically as follows: R channel (red pixel): High response in the visible red band (approximately 600-700nm), and usually exhibits "leakage" in the near-infrared band (e.g., 850-1000nm), meaning it also has high transmittance and responsivity for 940nm infrared light. B channel (blue pixel): High response in the visible blue band (approximately 400-500nm, including 450nm), but usually also exhibits some crosstalk in the near-infrared band (e.g., 940nm). G channel (green pixel): Typically has the strongest response at 500-600nm, but in the specific bands (450nm / 940nm) of this application, its response characteristics can be used for auxiliary verification or ignored.

[0083] In some embodiments, extracting the R-channel response value and B-channel response value of each pixel in the original image data can be achieved by iterating through a two-dimensional matrix of pixels in the original image data, for coordinates... For each pixel, read its R channel component. and B channel components Based on the pre-calibrated spectral response characteristics, determine It is mainly contributed by the light energy in the 940nm band, while It is contributed by both the light energy in the 450nm band and the infrared leakage energy in the 940nm band. For example, for a single pixel, the extracted... , .

[0084] In step 103, pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions are identified as valid fire point pixels.

[0085] Here, an effective fire point pixel refers to a pixel that simultaneously satisfies the conditions of high infrared radiation intensity and a specific infrared / visible light energy ratio in terms of spectral characteristics, and is used to distinguish real flames from interference sources. In the embodiments of this application, effective fire point pixels can be marked as 1, and non-fire point pixels can be marked as 0, resulting in a binarized image.

[0086] In some embodiments, pixels that meet set recognition criteria are identified as valid fire points, which can be achieved by setting multi-dimensional feature boundaries or region segmentation algorithms. For example, the system can construct a two-dimensional feature vector from the extracted R-channel and B-channel response values ​​of a specific pixel. The two-dimensional feature vectors are then mapped to a pre-established feature coordinate system; subsequently, a pre-calibrated discriminant function is used. Alternatively, the decision boundary can be used to classify the vector. If the two-dimensional feature vector falls within a preset numerical region representing "flame features" (i.e., the discrimination result matches the target category), the pixel position is marked as 1 (valid fire pixel) in the binarized image matrix; otherwise, it is marked as 0 (background or interference pixel). This method achieves flexible identification of the physical attributes of target pixels through the distribution law of the feature space.

[0087] In some embodiments, the pixel whose response value of the R channel and the response value of the B channel meet the set recognition conditions in step 103 is identified as a valid fire point pixel, which can be achieved by performing the following steps 1031 to 1032.

[0088] In step 1031, the difference between the response value of channel B and the response value of channel R is calculated.

[0089] Here, the difference between pixels can be used to represent the photoelectric signal energy that is separated by mathematical operations and is only related to a second specific band (such as the 450nm blue light band), which can be used for subsequent spectral comparison and analysis.

[0090] In some embodiments, the difference can be calculated using pixel-level subtraction. When calculating the difference, it is assumed that the response efficiency of the R channel to the first specific band is close to the leakage response efficiency of the B channel to the first specific band (e.g., Figure 5 The efficiency curves for the R, G, and B channels shown overlap in the 850nm to 1000nm band, or have been corrected by coefficients.

[0091] In some embodiments, the difference between the response values ​​of channel B and channel R can be calculated using the following method: [using the formula...] Perform the calculations. Among them, Indicates the difference. This is the response value for channel B. This is the response value for the R channel. This calculation aims to remove the superimposed energy from the first specific band (940nm) mixed in with the B channel, thereby restoring the energy value mainly excited by the second specific band (450nm). As an example: suppose a pixel... (Characterizing 940nm energy). (Characterizing the energy at 450nm + 940nm). The difference is then calculated as: The difference This approximates the light intensity at the 450nm wavelength at that location.

[0092] In step 1032, pixels that satisfy the condition that the response value of the R channel is greater than a preset second threshold and the ratio between the response value of the R channel and the difference is greater than a preset third threshold are identified as valid fire points. The second threshold and the third threshold are determined according to the set identification conditions.

[0093] In some embodiments, the preset second and third thresholds can be determined through statistical analysis of a large number of flame samples (such as candles, oil fires) and interfering samples (such as sunlight, car headlights, flashlights). See also Figure 6 , Figure 6The diagram shows the spectral positions of other interfering light luminescence. By comparing the spectral distribution patterns of various luminescent bodies in the ultraviolet (UV), visible (VIS), and infrared (IR) bands, the physical benchmark for fire detection can be seen: the energy of other interfering light luminescence in environments such as sunlight, LEDs, fluorescent lamps, incandescent lamps, gas-emitting lamps, and panel lights is mainly concentrated in the visible light range of 400nm to 850nm. The band above 850nm is a bandpass filtering range that can eliminate most interference. Therefore, by selecting a bandpass filtering range above 850nm (such as 940nm used in this application), most spectral interference from artificial and natural light sources can be filtered out at the physical source, providing core spectral theoretical support for achieving high accuracy and low false alarm fire detection. Furthermore, fire detection based solely on energy analysis in the 940nm band cannot distinguish between sunlight and flame emission; see [link to relevant documentation]. Figure 7 , Figure 7 This diagram compares the solar and flame spectra, with normalized energy density (NED) plotted on the ordinate to represent the relative intensity of sunlight and flame energy across different frequency bands. It shows that the energy peaks of the solar spectrum are concentrated in the 450 nm region (visible light region), where the NED can reach above 2.0. However, in the near-infrared region (940 nm-980 nm), the NED drops below 1.0. Similarly, the energy peaks of the flame spectrum are concentrated in the near-infrared region (940 nm-980 nm), where the NED can reach above 1.5. In the 450 nm region, the NED approaches 0. By utilizing the physical property that the radiant energy of a flame is significantly higher at 940 nm than at 450 nm, it is possible to effectively distinguish between sunlight and flame light in the original image data.

[0094] In some embodiments, marking any pixel as a valid fire point pixel can be achieved by setting a second threshold as follows: The third threshold is For any pixel Extract arbitrary pixels The response value of the R channel and any pixel The response value of channel B Determine whether the following conditions are met simultaneously: , If both conditions are met, the pixel is marked as 1 (fire point); otherwise, it is marked as 0 (background). As an example: Set... , Case A (Flame): Pixel of , Determine the intensity: The response value of the R channel is higher than the preset second threshold. Judgment ratio: ,because The ratio between the response value of the R channel and the difference between any pixel and the corresponding value is higher than a preset third threshold. Result: Marker Valid fire point pixels. Case B (sunlight reflection): pixels. of , (Sunlight also has high energy in the visible light spectrum). Judging intensity: The condition is met. Determine the ratio: ,because The condition is not met. Result: Pixels that are not marked as fire points are thus excluded from false alarms.

[0095] In step 104, the number of valid fire point pixels in the original image data is counted. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the detection area.

[0096] Here, the first threshold refers to the number of pixels set according to the required size and detection distance of the fire point. If the number of fire point pixels is less than or equal to the first threshold, it is considered that there may be interference. If the number of fire point pixels is higher than the first threshold, it is considered that the set fire point that needs to be detected has appeared.

[0097] In some embodiments, the number of valid fire point pixels in the original image data can be obtained by a connected component labeling algorithm or by direct traversal counting.

[0098] In some embodiments, determining the presence of a fire point within the detection area can be achieved by: calculating the detection area. The total number of all pixels marked as valid fire points Set the first threshold. (For example If the formula is satisfied... If a fire alarm signal is detected, it will be output. For example, in the binarized image processed in step 103, a clustered area containing... 0 valid fire point pixels. Because... The image processing equipment determines that a real fire point exists in the area and triggers subsequent alarm or location procedures. This step combines spectral feature analysis and spatial distribution feature analysis, further improving the robustness of the detection results.

[0099] In some embodiments, the number of valid fire pixels within the detection area can also be counted by performing channel separation processing on the RAW data. Assume the resolution of the raw image data acquired by the first acquisition device is... (For example First, the original The RAW image data is parsed into four independent sub-channel matrices. Because in the RGGB array, each... A pixel unit contains 1 R pixel, 1 B pixel, and 2 G pixels. The image processing device can sample at a step size of 2: extracting coordinates as... The pixels are used to form an R-channel response map (R-Map), with a resolution of [resolution missing]. ; Extract coordinates as The pixels are used to form a B-channel response map (B-Map), whose resolution is also [resolution missing]. At this point, each point in the R-channel response graph... Points with the same coordinates as in the B channel response graph Although there is a tiny spatial displacement of one pixel in the original RAW image, in the optical imaging of long-range fire detection, it is treated as a spectral response from the same physical spatial location. The system then processes the separated... The R-channel and B-channel response maps are matrix-processed to generate a binary mask map of the same size. For coordinates... A pixel at a certain location must meet both of the following conditions to be marked as a "valid fire pixel" (marked as 1, otherwise 0): Condition A (Intensity Filtering): (That is, the R channel response value at this point must be greater than the preset second threshold, for example, 200, to eliminate dark background noise); Condition B (spectral ratio screening): (That is, the ratio of infrared to blue light (including crosstalk correction) at this point must be greater than a preset third threshold, for example, 1.5.) To prevent extremely small values ​​from being divided by zero. (In generating a resolution of...) After obtaining the binarized mask, the image processing device performs statistics on the region of interest (ROI) to be detected, determined in step 103 and scaled and mapped to the coordinate system of the mask. It calculates the total number of pixels marked as 1 within the ROI. .like ( If a preset first threshold (e.g., 5 pixels) is used, then it is determined that a real fire point exists in the area, and a fire alarm signal is generated. If If this is not the case, the area is determined to be interference or noise, and no fire point is indicated. This method... The original RAW image is decomposed into The method of processing the single-channel response map not only avoids the data smoothing and feature ambiguity caused by neighborhood interpolation, but also significantly reduces the amount of data processing (data volume reduced by 75%) and improves the real-time performance of fire point determination.

[0100] This application's embodiments, based on pure visual detection using a visual sensor, utilize a general-purpose image sensor (such as CMOS / CCD) combined with dual narrowband light-transmitting elements to extract and analyze specific spectral energy using the sensor's inherent photosensitive characteristics. This "full-vision" discrimination method avoids the introduction of high-cost hardware such as infrared thermal imaging, temperature sensing, or infrared / ultraviolet sensors, ensuring both detection resolution and positioning accuracy while significantly reducing system complexity and cost. Furthermore, by calculating the difference between the response values ​​of the B and R channels, the independent energies of the first specific band (near-infrared) and the second specific band (blue visible light) are separated in a single imaging session. Utilizing the fundamental difference in the radiance ratios of flames and sunlight reflections, and lamplight in the aforementioned first and second specific bands (i.e., flames have significantly higher energy in the infrared band than in the visible light band), a third threshold is used to effectively filter out high-brightness interference sources (such as specular reflections, vehicle headlights, and other lighting sources) that are prone to false alarms. Furthermore, this embodiment directly extracts response values ​​from the original image data for calculation, avoiding the disruption of spectral energy ratios caused by nonlinear transformations such as white balance, gamma correction, and tone mapping in conventional image signal processing (ISP). This allows the algorithm to obtain more realistic physical photosensitive values. Combined with the second threshold determination, it can accurately eliminate artificial interference targets with low radiation intensity, improving the detection robustness of extremely weak fire points and complex backgrounds. Finally, by statistically analyzing the number of effective fire point pixels, spatial scale discrimination is introduced across the entire visual image dimension. This not only effectively filters out single-point false alarms caused by sensor noise, bad pixels, or minor environmental reflections in the acquisition device, but also ensures that the finally determined fire points have realistic physical size characteristics, further improving the accuracy of the alarm.

[0101] The specific implementation process of the fire detection method provided in the embodiments of this application is described below.

[0102] See Figure 8 Fire detection methods include: A second acquisition device (bullet camera) with a wider field of view performs a panoramic scan of the scene of interest to obtain a wide-angle image. An image recognition model then identifies the wide-angle image with a first confidence level (bullet camera detection), yielding the bullet camera detection result. If a first suspected fire point area appears in the bullet camera detection result, its spatial coordinates are extracted, and the first acquisition device (PTZ camera) is guided to rotate and align with these coordinates, designating the first suspected fire point area as the detection area. In other words, the detection area is extracted based on the bullet camera detection result and then detected by the PTZ camera. Subsequently, the first acquisition device magnifies the image using optical zoom (through a fully transparent element) and uses a visual recognition model with a second confidence level (higher than the first confidence level) to perform a secondary verification of the target. Based on the PTZ camera detection result, the detection area is redefined, thereby eliminating distant false targets and pinpointing the precise core area of ​​the flame.

[0103] After visual locking is completed, physical property acquisition and analysis are performed. The first acquisition device switches from a full-pass light element to a specially designed dual-narrowband light element (e.g., allowing only 940nm infrared light and 450nm blue light to pass through), and directly acquires the raw image data, i.e., RAW data, of the area to be detected in this state. The image processing device acquires the raw image data obtained by the PTZ camera through the dual-narrowband light element, and analyzes the RAW data based on the spectral response characteristics of the acquisition device, separating the data of each pixel into R / B channel response maps, where the R channel represents infrared energy and the B channel represents the mixed value of blue light and infrared energy.

[0104] The system performs dual physical threshold verification on each pixel in parallel. On one hand, it determines whether the R-channel response value is greater than a preset second threshold to ensure that the target has sufficient infrared radiation intensity and eliminate noise. Pixels with an R-channel response value greater than the preset second threshold are marked as valid fire pixels, while pixels with an R-channel response value less than or equal to the preset second threshold are marked as non-fire pixels. On the other hand, it calculates the R / (BR) channel response value (i.e., the ratio of infrared to blue light energy) and marks pixels with an R / (BR) channel response value greater than a preset third threshold as valid fire pixels, while pixels with an R / (BR) channel response value less than or equal to the preset third threshold are marked as non-fire pixels. This verifies whether the target conforms to the physical characteristics of a flame: "extremely strong infrared radiation and relatively weak blue light radiation," thereby eliminating interference from sunlight.

[0105] Based on the intersection of the above judgment results, effective fire point pixels that simultaneously meet the intensity and spectral conditions are selected, and the number of effective fire point pixels is counted to determine the fire point.

[0106] If the number of valid fire point pixels is greater than the preset first threshold, it indicates that the target has a certain spatial scale and consistent pixel features, thus ultimately determining that there is a fire point in the area to be detected; if the number of valid fire point pixels is less than or equal to the preset first threshold, it is determined that there is no fire point in the area to be detected, thus completing the confirmation of the fire situation.

[0107] The following description continues to illustrate the exemplary structure of the fire detection device 255 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 2 As shown, the software module stored in the fire detection device 255 in the memory 250 may include: The original image acquisition module 2551 is used to acquire the original image data of the area to be detected. The original image data is acquired by the first acquisition device through a dual narrowband light transmission element. The dual narrowband light transmission element is used to filter out the energy of bands other than the first specific band and the second specific band. The response value extraction module 2552 is used to extract the response value of the R channel and the response value of the B channel of each pixel in the original image data. The response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. The effective fire point pixel determination module 2553 is used to identify pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions as effective fire point pixels. The fire point confirmation module 2554 is used to count the number of valid fire point pixels in the area to be detected. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the area to be detected.

[0108] This application provides a computer program product, which includes a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions from the computer-readable storage medium and executes the computer-executable instructions, causing the electronic device to perform the fire detection method described above in this application.

[0109] This application provides a computer-readable storage medium storing computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, the processor will execute the fire detection method provided in this application. For example, ... Figure 3 The fire detection method is shown.

[0110] In some embodiments, the computer-readable storage medium may be a memory such as RAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0111] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0112] As an example, computer-executable instructions can be deployed to execute on a single electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0113] In summary, the embodiments of this application extract the R-channel and B-channel response values ​​of each pixel from the original image data obtained through the dual narrowband light transmission element. By utilizing the different effects of different types of light sources on the R-channel and B-channel, the independent energies of the first specific band and the second specific band are decoupled from the original image. This allows for the effective differentiation of flames from other interfering light sources based on their physical properties, taking advantage of the different spectra of real flames in different bands, thus improving the accuracy of fire detection.

[0114] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for detecting fire points, characterized in that, The method includes: Acquire raw image data of the area to be detected, wherein the raw image data is acquired by a first acquisition device through a dual narrowband light transmission element, the dual narrowband light transmission element being used to filter out energy of bands other than the first specific band and the second specific band; Extract the response value of the R channel and the response value of the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. Pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions are identified as valid fire point pixels. The number of valid fire point pixels in the original image data is counted. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the detection area.

2. The method according to claim 1, characterized in that, The step of identifying pixels whose response values ​​of the R channel and the B channel satisfy the set recognition conditions as valid fire points includes: Calculate the difference between the response value of channel B and the response value of channel R; A pixel that satisfies the condition that the response value of the R channel is greater than a preset second threshold and the ratio between the response value of the R channel and the difference is greater than a preset third threshold is identified as a valid fire point pixel, wherein the second threshold and the third threshold are determined according to the set identification conditions.

3. The method according to claim 1, characterized in that, Before acquiring the original image data of the region to be detected, the method further includes: A wide-angle image including the area to be detected is acquired, wherein the wide-angle image is acquired by at least one second acquisition device, and the field of view coverage of the second acquisition device is greater than that of the first acquisition device; Using an image recognition model, a first suspected fire point region in the wide-angle image is identified based on a set first confidence level, and the spatial coordinates of the first suspected fire point region are determined. The orientation of the first acquisition device is controlled to align with the position corresponding to the spatial coordinates, and the first suspected fire point area is taken as the area to be detected.

4. The method according to claim 3, characterized in that, The first acquisition device also includes a fully optically transparent element; After controlling the orientation of the first acquisition device to align with the position corresponding to the spatial coordinates, the method further includes: A first suspected fire point image is obtained, wherein the first suspected fire point image is obtained by the first acquisition device, when aligned with the position corresponding to the spatial coordinates, by magnifying the first suspected fire point area in the field of view of the first acquisition device to a preset image size and then passing it through the full-transmission light element; the first suspected fire point image is an image including the first suspected fire point area. The image recognition model identifies a second suspected fire point region in the first suspected fire point image based on a set second confidence level, wherein the second confidence level is higher than the first confidence level. If the second suspected fire point area is identified, the second suspected fire point area is taken as the area to be detected, triggering the step of acquiring the original image data of the area to be detected.

5. The method according to claim 4, characterized in that, The first acquisition device also includes a switcher; Before triggering the step of acquiring the raw image data of the region to be detected, the method further includes: If the second suspected fire point area is identified, the second suspected fire point area is taken as the area to be detected, and the full-transmission element set in the first acquisition device is switched to the dual narrow-band transmission element through the switcher.

6. The method according to claim 1, characterized in that, The step of extracting the R-channel response value and B-channel response value of each pixel in the original image data includes: For each pixel, the original image data is analyzed based on the spectral response characteristic curve of the first acquisition device to obtain the response value of the R channel and the response value of the B channel.

7. A fire detection system, characterized in that, include: First acquisition equipment and image processing equipment; The first acquisition device is used to acquire raw image data of the area to be detected while transmitting light through dual narrowband light transmission elements, wherein the dual narrowband light transmission elements are used to filter out energy of bands other than the first specific band and the second specific band. The image processing device is configured to extract the response values ​​of the R channel and the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band; and, Pixels whose response values ​​of the R channel and the B channel satisfy the set recognition conditions are identified as valid firepoint pixels; and, The number of valid fire point pixels in the original image data is counted. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the detection area.

8. A fire detection device, characterized in that, The device includes: The original image acquisition module is used to acquire the original image data of the area to be detected. The original image data is acquired by the first acquisition device through a dual narrowband light transmission element. The dual narrowband light transmission element is used to filter out the energy of bands other than the first specific band and the second specific band. The response value extraction module is used to extract the response value of the R channel and the response value of the B channel of each pixel in the original image data, wherein the response value of the R channel is used to characterize the energy of the first specific band, and the response value of the B channel is used to characterize the superposition energy of the energy of the first specific band and the energy of the second specific band. The effective fire point pixel determination module is used to identify pixels whose response values ​​of the R channel and the B channel meet the set recognition conditions as effective fire point pixels. The fire point confirmation module is used to count the number of valid fire point pixels in the original image data. If the number of valid fire point pixels exceeds a preset first threshold, it is determined that there is a fire point in the area to be detected.

9. An electronic device, characterized in that, The electronic device includes: Memory is used to store executable instructions or computer programs. A processor, when executing computer-executable instructions or computer programs stored in the memory, implements the fire detection method according to any one of claims 1 to 6.

10. A computer-readable storage medium storing computer-executable instructions or a computer program, characterized in that, When the computer-executable instructions or computer program are executed by a processor, the fire detection method according to any one of claims 1 to 6 is implemented.

Images