Tunnel fire early warning system

Abstract

The invention provides a tunnel fire early warning system, and relates to the technical field of fire detection. The system comprises a video acquisition module used for acquiring video images in a tunnel; the background extraction module is used for identifying a suspicious motion area containing flame / smoke from the video; the flame extraction module is used for screening flame color feature regions by using an HSI model and extracting related feature data; the flame judgment module is used for fusing the data through a three-layer BP neural network to obtain flame judgment information; the smoke judgment module is used for checking whether the main motion direction of the suspicious area is consistent with a preset direction or not, and performing texture feature detection when the main motion direction is consistent with the preset direction to obtain smoke judgment information; and the fire judgment module is used for judging whether a fire occurs or not by combining the judgment results of the smoke and the flames and sending out a corresponding early warning signal. According to the scheme, the accuracy and the real-time performance of fire monitoring can be improved, the false alarm rate and the missing report rate are reduced, and therefore safe operation of a tunnel is more effectively guaranteed.

Description

Technical Field

[0001] The present application relates to the field of fire detection technology, and in particular to a tunnel fire early warning system. Background Art

[0002] In modern transportation, tunnels serve as crucial passages connecting cities, crossing mountains, and across waterways, making their safety paramount. However, due to the unique structure of tunnels, smoke and flames can spread rapidly in the event of a fire, making rescue and escape extremely difficult. Therefore, the development and implementation of tunnel fire warning systems is particularly important.

[0003] Currently, most tunnel fire warning systems utilize a combination of sensor technology and image processing. Sensor technology determines fire conditions by monitoring parameters such as smoke concentration and temperature within the tunnel, while image processing technology detects fire characteristics such as flames and smoke by collecting and analyzing video images within the tunnel. However, existing tunnel fire warning systems still have several issues and shortcomings in their application. First, while sensor technology can relatively accurately monitor fire parameters, it is limited by the number and placement of sensors, making it difficult to achieve comprehensive coverage and real-time monitoring. Second, image processing technology is susceptible to factors such as lighting changes and shadow interference when detecting flames and smoke, resulting in high rates of false alarms and missed alarms.

[0004] Therefore, there is an urgent need for a tunnel fire warning system that can improve the accuracy and real-time performance of fire monitoring, reduce the false alarm and missed alarm rates, and thus more effectively ensure the safe operation of the tunnel, in order to address the above-mentioned many problems and shortcomings that still exist in the application of existing tunnel fire warning systems. Summary of the Invention

[0005] The purpose of this application is to provide a tunnel fire warning system that can improve the accuracy and real-time performance of fire monitoring, reduce false alarm and missed alarm rates, and thus more effectively ensure the safe operation of the tunnel.

[0006] This application is implemented as follows:

[0007] In a first aspect, the present application provides a tunnel fire warning system, comprising: a video acquisition module for acquiring video images in a tunnel; a background extraction module for performing motion foreground target extraction processing on the video images to extract suspicious motion areas including flames and / or smoke; a flame extraction module for segmenting an area that meets flame color characteristics from the suspicious motion area using an HSI color space model, and extracting flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape change based on the segmented flame suspected area; a flame judgment module for sending the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; a smoke judgment module for judging whether the main motion direction of the suspicious motion area is consistent with a preset bottom-up direction, and performing smoke texture feature detection when it is consistent to obtain smoke judgment result information; a fire judgment module for judging whether a fire has occurred based on the smoke judgment result information and the flame judgment result information, so as to issue a corresponding warning signal.

[0008] In a second aspect, the present application provides a tunnel fire early warning method, which comprises the following steps:

[0009] Collect video images in the tunnel; perform motion foreground target extraction on the video images to extract suspicious motion areas including flames and / or smoke; use the HSI color space model to segment areas that meet flame color characteristics from the suspicious motion areas, and extract flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape changes based on the segmented flame suspected areas; send the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; based on the suspicious motion area, determine whether its main motion direction is consistent with the preset bottom-up direction, and if it is consistent, perform smoke texture feature detection to obtain smoke judgment result information; based on the smoke judgment result information and the flame judgment result information, determine whether a fire has occurred and issue a corresponding early warning signal.

[0010] In a third aspect, the present application provides an electronic device comprising a memory for storing one or more programs; a processor; and when the one or more programs are executed by the processor, the method as described in any one of the second aspects is implemented.

[0011] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of the second aspects above.

[0012] Compared with the prior art, this application has at least the following advantages or beneficial effects:

[0013] This application proposes a tunnel fire early warning system that, by performing motion foreground object extraction processing on video images, can rapidly identify suspicious moving areas, including flames and / or smoke, thereby enabling comprehensive monitoring of fire conditions within tunnels. Furthermore, by utilizing the HSI color space model, the system accurately segments areas that meet flame color characteristics and extracts detailed flame feature data, such as color, number of sharp corners, roundness, flame transition frequency, and shape variation. This feature data provides a rich information foundation for subsequent flame identification, helping to reduce false alarms. Furthermore, the system not only monitors flames but also preliminarily screens possible smoke areas by determining whether the main motion direction of suspicious moving areas matches a preset bottom-up direction. This step effectively avoids false alarms caused by other moving objects (such as vehicles and pedestrians) and improves the accuracy of smoke identification. Furthermore, by utilizing a pre-trained three-layer BP neural network for flame feature fusion, this step fully considers the multiple characteristics of flames. Leveraging the powerful learning capabilities of the neural network, the system improves the accuracy of flame identification and reduces the false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0015] Figure 1 This is a structural block diagram of an embodiment of a tunnel fire warning system of the present application;

[0016] Figure 2 This is a structural block diagram of another embodiment of a tunnel fire warning system of the present application;

[0017] Figure 3 This is a structural block diagram of another embodiment of a tunnel fire warning system of the present application;

[0018] Figure 4 This is a flow chart of an embodiment of a tunnel fire early warning method of the present application;

[0019] Figure 5 This is a structural block diagram of an electronic device provided in an embodiment of the present application.

[0020] Icon: 201, processor; 202, memory; 203, communication interface. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Generally, the components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0022] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present application without creative work are within the scope of protection of the present application.

[0023] The following describes some embodiments of the present application in detail with reference to the accompanying drawings. In the absence of conflict, the following embodiments and features thereof may be combined with each other.

[0024] Example 1

[0025] The embodiment of the present application provides a tunnel fire warning system, which can improve the accuracy and real-time performance of fire monitoring, reduce false alarm and missed alarm rates, and thus more effectively ensure the safe operation of the tunnel.

[0026] See also Figure 1 The tunnel fire warning system includes: a video acquisition module for acquiring video images in the tunnel; a background extraction module for performing motion foreground target extraction processing on the video images to extract suspicious motion areas including flames and / or smoke; a flame extraction module for segmenting an area that meets the flame color characteristics from the suspicious motion area using the HSI color space model, and extracting flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape change based on the segmented flame suspected area; a flame judgment module for sending the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; a smoke judgment module for judging whether the main motion direction of the suspicious motion area is consistent with a preset bottom-up direction, and performing smoke texture feature detection if it is consistent to obtain smoke judgment result information; a fire judgment module for judging whether a fire has occurred based on the smoke judgment result information and the flame judgment result information, so as to issue a corresponding warning signal.

[0027] In the above embodiment, the video capture module is responsible for capturing video images within the tunnel, providing raw data for subsequent fire warnings. For example, a high-definition camera installed at the top of the tunnel can capture dynamic images of the tunnel in real time. This captures subtle changes in the early stages of a fire, providing timely information support for fire warnings.

[0028] Next, the background extraction module extracts moving foreground objects from the video image. By comparing the current frame with the background frame, it identifies suspicious moving areas, including flames and / or smoke. This effectively filters out static background and highlights moving objects, providing a clear image foundation for subsequent flame and smoke detection. This step helps reduce false alarms and improves early warning accuracy.

[0029] Next, the flame extraction module uses the HSI color space model to further analyze suspicious motion areas. The HSI model decomposes color information into three components: hue, saturation, and intensity. This allows the system to more accurately identify flame color characteristics. Furthermore, the module extracts other flame characteristics, such as the number of sharp corners, circularity, flame jump frequency, and shape changes. These features provide rich information for subsequent flame identification.

[0030] The flame identification module then receives the feature data output by the flame extraction module and feeds it into a pre-trained three-layer BP neural network for flame feature fusion. The BP neural network is a machine learning algorithm that, through training and learning, can accurately identify whether the input features belong to flames. This step further improves the accuracy and reliability of flame identification.

[0031] Simultaneously, the smoke detection module analyzes suspicious motion areas. This module determines whether the main direction of movement in the area matches a preset bottom-up direction, based on the fact that smoke typically rises from the ground. If so, it then performs smoke texture feature detection to generate smoke detection results. By determining the direction of movement, it effectively distinguishes smoke from other moving objects, improving smoke detection accuracy.

[0032] Finally, the fire detection module combines the flame and smoke detection results to determine whether a fire has occurred. If both meet the fire characteristics, the system issues a corresponding warning signal, notifying relevant personnel to take appropriate countermeasures.

[0033] In summary, this tunnel fire warning system, through motion foreground object extraction processing in video images, can rapidly identify suspicious motion areas, including flames and / or smoke, achieving comprehensive monitoring of fire conditions within tunnels. Furthermore, by utilizing the HSI color space model, the system accurately segments areas meeting flame color characteristics and extracts detailed flame feature data, such as color, number of sharp corners, roundness, flame transition frequency, and shape variations. This feature data provides a rich foundation for subsequent flame identification, helping to reduce false alarms. Furthermore, the system not only focuses on flames but also preliminarily screens possible smoke areas by determining whether the main motion direction of suspicious motion areas matches a predefined bottom-up direction. This step effectively avoids false alarms caused by other moving objects (such as vehicles and pedestrians) and improves the accuracy of smoke detection. Furthermore, flame feature fusion is performed using a pretrained three-layer BP neural network. This step fully considers multiple flame characteristics and leverages the powerful learning capabilities of the neural network to improve flame detection accuracy and reduce false alarm rates.

[0034] Based on the above scheme, please refer to Figure 2 In some implementations of the present application, the system further includes a pre-processing module for performing image enhancement and filtering on the video image fed into the background extraction module.

[0035] In the above implementation, the preprocessing module performs image enhancement on the video image fed into the background extraction module, thereby improving the quality of the video image, making it clearer and easier to analyze. Image enhancement techniques may include contrast enhancement, brightness adjustment, and sharpening. These operations can enhance the detailed information in the image, making target features such as flames and smoke more distinct, thereby helping to improve the accuracy of subsequent background extraction and flame / smoke identification. At the same time, filtering is mainly used to remove noise and interference from video images. The environment inside the tunnel is complex, and there may be interference factors such as light changes, camera shake, and dust. These factors may cause unnecessary noise or blurred areas in the image. Through filtering processes such as Gaussian filtering, mean filtering, or median filtering, these noises can be effectively reduced, making the image smoother and cleaner, and facilitating subsequent processing and analysis.

[0036] Specifically, the preprocessing module performs image enhancement and filtering on the video images fed into the background extraction module, significantly improving the quality of the video images and making target features such as flames and smoke more prominent. This improves the accuracy of background extraction, flame extraction, and smoke detection, and reduces the risk of false alarms and missed alarms. Furthermore, the system is able to cope with complex and changing environmental conditions within tunnels, such as lighting changes and camera jitter. By reducing noise and interference, it enhances the system's stability and robustness, enabling it to maintain high performance in a variety of environments.

[0037] Based on the above scheme, please refer to Figure 3 In some implementations of the present application, the background extraction module further includes: a background extraction unit for reading each frame of the video image, performing grayscale histogram statistics on each pixel in each frame to find the point with the highest frequency, and calculating the energy contained in its neighborhood. If the energy exceeds 3 / 5 of the total energy of the entire histogram, the grayscale value corresponding to the point is used as the background grayscale value. Otherwise, another point with the highest frequency is searched outside the neighborhood as the background grayscale value to determine the background image information from the video image based on the background grayscale value. A motion area extraction unit for comparing the current frame of the video image with the background image information to extract suspicious motion areas including flames and / or smoke.

[0038] In the above implementation, the background extraction unit first reads each frame of the video image and performs a grayscale histogram analysis on each pixel in each frame. The grayscale histogram can reflect the distribution of grayscale values ​​in the image. Next, the unit searches for the point with the highest frequency, that is, the grayscale value that occurs the most frequently, which represents the main or stable part of the image. The unit then calculates the energy contained in the neighborhood of this point (such as a small window or area). Energy here is a measure of the statistics (such as variance, sum of squares, etc.) of the grayscale values ​​of the pixels in this neighborhood. If this energy exceeds 3 / 5 of the total energy of the entire histogram, it indicates that the point and its neighborhood dominate the image, and the grayscale value corresponding to this point is therefore used as the background grayscale value. If the energy is less than 3 / 5, the unit continues to search for another point with the highest frequency outside the neighborhood and repeats the above determination process until a suitable background grayscale value is found. Finally, based on the determined background grayscale value, background image information is constructed from the video image pixel by pixel.

[0039] The motion region extraction unit compares the current frame image with the background image pixel by pixel. By comparing the grayscale values ​​of corresponding pixels in the current frame and the background image, it can extract areas that show significant changes compared to the background. These areas are considered suspicious motion areas and may contain flames, smoke, or other moving objects.

[0040] In summary, through grayscale histogram statistics and energy judgment, the background extraction unit can more accurately determine the background grayscale value, especially in complex and changing scenes, such as tunnel lighting changes and camera jitter, while maintaining the stability of background extraction. The motion region extraction unit uses the contrast between the current frame and the background image to more sensitively capture suspicious moving areas such as flames and / or smoke, thereby improving the response speed and accuracy of the early warning system. In other words, through the background extraction unit and motion region extraction unit of the above implementation method, the system can more effectively distinguish between real targets such as flames and smoke and false targets such as noise and interference, thereby reducing the false alarm rate.

[0041] Based on the above scheme, please refer to Figure 3 In some implementations of the present application, the flame extraction module further includes: a flame suspected region extraction unit, configured to convert the suspicious motion region from an RGB image into an HSI image and feed it into an HSI color space model, thereby traversing each pixel in the HSI image and determining whether it meets the threshold conditions for flame color characteristics based on its H value, S value, and I value. If the H value, S value, and I value of a pixel simultaneously meet the set threshold conditions, it is deemed to be a pixel of flame color characteristics and retained in the segmentation result, so as to obtain a flame suspected region based on the segmentation result. A flame feature data extraction unit, configured to extract flame feature data including color, number of sharp corners, roundness, flame jump frequency, and shape change based on the flame suspected region.

[0042] It should be noted that the HSI color space model is a color space model in which H represents hue, S represents saturation, and I represents intensity. Compared to the RGB model, the HSI color space model is more intuitive and effective in describing color features. The suspected flame region extraction unit first converts the suspicious motion region from the RGB image into an HSI image. It then traverses each pixel in the HSI color space model and determines whether its H, S, and I values ​​meet the threshold conditions for flame color features. These threshold conditions are set based on the color characteristics of flames in HSI space. For example, flames typically have a specific hue range, high saturation, and a certain brightness range. If a pixel's H, S, and I values ​​all meet the threshold conditions, it is considered a pixel with flame color features and is retained in the segmentation result. By performing this judgment on all pixels, a segmentation result containing flame color features is obtained, namely the suspected flame region.

[0043] The flame feature data extraction unit further extracts flame feature data based on the suspected flame area. This feature data includes color, number of sharp corners, roundness, flame jump frequency, and shape changes, which can more comprehensively describe the morphological and dynamic characteristics of the flame. Among them, the color feature of the flame is one of its most basic characteristics and can be described by HSI values ​​or RGB values. The edge of the flame usually has irregular sharp corners, and the number of sharp corners can be used as a characteristic of the flame shape. Although the shape of the flame is usually irregular, its roundness can be approximately described by calculating the similarity between its circumscribed circle and the actual shape. The flickering and jumping of the flame is one of its dynamic characteristics, which can be described by calculating the frequency of change in the brightness of the flame area. The change of the flame shape over time is also one of its important characteristics and can be extracted by comparing between consecutive frames.

[0044] In summary, by converting the suspicious motion area into an HSI image and using the color features of the flame in the HSI space for judgment, the flame suspected area extraction unit can more accurately identify the flame area and reduce false positives and negatives.

[0045] Exemplarily, in some implementations of the present application, the H value, S value, and I value of a pixel point simultaneously satisfying the set threshold conditions include: the H value ∈ [0°, 60°], S value ∈ [0.4, 1], and I value ∈ [0.6, 1] of the pixel point.

[0046] It should be noted that the hue (H) of flames in the HSI color space is typically concentrated within the red to yellow range. In this application, the H value threshold condition is set to the pixel's H value ∈ [0°, 60°]. This range covers hues from red (0°) to yellow (60°), consistent with the primary color characteristics of flames. Furthermore, flames typically have a high saturation (S), meaning their colors are more vivid. In this application, the S value threshold condition is set to the pixel's S value ∈ [0.4, 1]. This range indicates that a pixel can only be considered a pixel with flame color characteristics if its saturation is at least 0.4 (i.e., 40% saturation). The upper limit is set to 1, indicating a fully saturated color. Furthermore, the brightness (I) of flames is typically high, but it should be noted that excessive brightness can also cause a pixel to be misidentified as a flame. Therefore, in this application, the I value threshold condition is set to the pixel's I value ∈ [0.6, 1]. This range means that a pixel can only be considered as a flame color feature if its brightness is at least 0.6 (i.e. 60% brightness). The upper limit is set to 1, indicating the maximum brightness.

[0047] Based on the above solution, in some implementations of the present application, the suspected flame region extraction unit is further configured to perform expansion, erosion, and edge detection processing on the segmentation result.

[0048] Among them, dilation is a morphological image processing operation whose main function is to expand the foreground area in the image (i.e., the suspected flame area). Through dilation, small holes or slits in the suspected flame area can be filled, making it more continuous and complete. This helps to reduce the problem of flame area fragmentation caused by image noise or segmentation errors. In contrast to dilation, erosion is an operation that shrinks the foreground area in the image. It is mainly used to remove small noise or irregular edges around the suspected flame area, making the flame area clearer and more regular. Erosion helps to reduce false positives, especially when there are interferences in the background with similar colors to the flame. In addition, edge detection is an important step in image processing. It is used to identify the boundaries between different areas in the image. In the extraction of suspected flame areas, edge detection can help determine the precise outline of the flame area, which is crucial for subsequent flame feature extraction and recognition. Through edge detection, the flame area and background area can be further distinguished, improving the accuracy of flame recognition.

[0049] Specifically, through dilation, erosion, and edge detection, the suspected flame region extraction unit can output a more accurate and clear flame region. These operations help reduce the impact of image noise and segmentation errors on flame identification, improving the integrity and continuity of the flame region. Furthermore, the optimized suspected flame region more accurately reflects the true shape and location of the flame, thereby improving the accuracy of flame identification. This helps reduce false alarms and missed alarms, improving the reliability of the fire warning system. Furthermore, these image processing operations can adapt to varying lighting conditions, flame shapes, and background complexity, enhancing the system's adaptability and robustness. Even in complex environments, the system can accurately identify the flame region and issue a warning.

[0050] Based on the above-mentioned scheme, in some implementation methods of the present application, the three-layer BP neural network includes an input layer with 5 nodes, a hidden layer with 4 nodes, and an output layer with 1 node; wherein the 5 input signals in the input layer are: the average value of the first-order color moment of a first preset number of consecutive suspected flame areas; the number of sharp corners of the suspected flame areas; the average value of the roundness of a second preset number of consecutive suspected flame areas; the number of high-frequency components greater than a first threshold value obtained by wavelet decomposition of the height change curve of a third preset number of consecutive suspected flame areas; and the average value of the shape similarity of a fourth preset number of consecutive suspected flame areas.

[0051] In this implementation, by incorporating input signals containing rich flame characteristics and combining them with the powerful classification capabilities of a three-layer BP neural network, the system can more accurately identify flames and reduce false positives and negatives. These input signals encompass multiple aspects of flame color, shape, and dynamic characteristics, and can adapt to varying lighting conditions, flame shape, and background complexity, improving the system's adaptability and robustness.

[0052] Among them, the detailed explanation of the 5-way signal of the input layer is as follows:

[0053] (1) Average of the first-order color moments of a first preset number of consecutive suspected flame regions: A color moment is a feature used to describe the color distribution of an image, where the first-order color moment is the average color of the image. Here, by calculating the average of the first-order color moments of a certain number (the first preset number) of consecutive suspected flame regions, the stability and consistency of the flame color can be reflected. This feature helps distinguish flames from other objects with similar colors.

[0054] (2) Number of sharp corners in the suspected flame area: The edge of a flame usually has irregular sharp corners. The number of sharp corners can be used as a characteristic of the flame morphology. By counting the number of sharp corners in the suspected flame area, information about the flame morphology can be obtained, which is helpful for flame recognition.

[0055] (3) Average circularity of a second preset number of consecutive suspected flame areas: Although the shape of a flame is generally irregular, its circularity can be approximated by calculating the similarity between its circumscribed circle and the actual shape. Here, by calculating the average circularity of a certain number (the second preset number) of consecutive suspected flame areas, the overall characteristics of the flame shape can be reflected.

[0056] (4) Using wavelet decomposition, the height change curves of the third preset number of consecutive flame suspected areas are decomposed, and the number of high-frequency components greater than the first threshold is obtained: The height of the flame usually changes over time, and this change can be described by the height change curve. Wavelet decomposition is an effective signal processing tool that can extract different frequency components in the signal. Here, the height change curves of a certain number (the third preset number) of consecutive flame suspected areas are decomposed by wavelet decomposition, and the number of high-frequency components greater than a certain threshold (the first threshold) is counted. This feature can reflect the dynamic characteristics of the flame, such as flickering and jumping.

[0057] (5) Average shape similarity of a fourth preset number of consecutive suspected flame regions: Shape similarity refers to the degree of similarity between suspected flame regions in consecutive frames. By calculating the average shape similarity of a certain number (the fourth preset number) of consecutive suspected flame regions, information about the stability of the flame shape can be obtained, which helps in continuous flame identification and tracking.

[0058] Based on the above solution, in some implementations of the present application, the hidden layer and the output layer of the three-layer BP neural network both use the sigmoid function as the excitation function of the neurons.

[0059] In the above implementation, using the sigmoid function as the excitation function in the hidden layer introduces nonlinear characteristics, enabling the neural network to learn and represent complex nonlinear relationships. This helps the neural network extract useful features from the input signal, preparing for subsequent classification tasks and thus improving the accuracy of flame recognition. Using the sigmoid function in the output layer constrains the neural network's output values ​​to the range (0, 1), allowing them to be interpreted as the probability of flame recognition. When the output value is close to 1, the probability of the input signal being recognized as flame is high; when the output value is close to 0, the probability of being recognized as non-flame is high. In other words, the use of the sigmoid function in the output layer allows the neural network's output value to be directly interpreted as the probability of flame recognition, helping to optimize classification performance and reduce false positives and false negatives.

[0060] Based on the above-mentioned scheme, in some implementations of the present application, the smoke texture feature detection includes judging the smoke texture based on three characteristics: the position of the smoke source is fixed, the early smoke blocks the background area so that the background edge is blurred, and the smoke is easy to spread.

[0061] It should be noted that smoke is typically released from a fixed source (such as a burning object). Therefore, the smoke's starting location (i.e., the source location) typically remains relatively fixed in consecutive frames of an image or video. This feature can be used to preliminarily determine whether blurred areas in an image or video are caused by smoke. Furthermore, as smoke spreads, it gradually obscures the background, blurring the background edges. This blurring effect is a key characteristic of smoke texture and can be quantified using image processing techniques (such as edge detection and blur calculation). Comparing the blur difference between the smoke region and the surrounding background can further confirm the presence of smoke. Furthermore, smoke is diffusive, gradually spreading from the source to the surrounding area. This characteristic manifests itself in an image or video as the gradual enlargement and morphological changes of the smoke region. By tracking the diffusion of the smoke region, the presence of smoke can be further verified and its potential spread can be predicted.

[0062] Therefore, in the above implementation, the three features above can be used to first extract the smoke texture in the suspicious motion area. This includes calculating the blur of the smoke area and tracking its diffusion process. Discriminant analysis is then performed based on the extracted features. If a region in the image or video meets two or more of the three features above (the specific threshold can be set according to the actual situation), it can be preliminarily judged as a smoke area.

[0063] Therefore, by combining the three characteristics of smoke source location stability, blurred background edges, and easy diffusion, smoke textures in images or videos can be more effectively identified, reducing false positives and false negatives. Furthermore, these characteristics are stable and robust under varying lighting conditions, smoke concentrations, and background complexity, enabling the system to adapt to diverse application scenarios and environmental changes.

[0064] Based on the aforementioned solution, in some implementations of the present application, the feature of the background edge being blurred due to the occlusion of the background area by the early smoke is detected by a spatial domain wavelet analysis algorithm.

[0065] In the above implementation, a two-dimensional wavelet transform is first performed on the image data corresponding to the suspicious motion region, decomposing it into low-frequency and high-frequency components. The low-frequency component contains the image's primary structure and contour information, while the high-frequency component contains edge, texture, and detail information. Next, the high-frequency component is analyzed. Because smoke obscures the background area, background edges become blurred. This blurring effect manifests as a weakening or disappearance of edge information in the high-frequency component. Therefore, the degree of edge blur can be quantified by calculating statistics such as the energy or variance of the high-frequency component. Then, based on the statistics of the high-frequency component, a threshold is set to determine which areas are blurred. If the high-frequency component statistics of a region are below the threshold, the region is considered blurred, likely due to smoke obscuration. Finally, the detected blurred regions undergo post-processing, such as morphological filtering and connected component analysis, to further confirm the presence of smoke and remove possible noise and interference.

[0066] Therefore, spatial wavelet analysis can extract multi-scale and multi-directional information from images, offering significant advantages for detecting blurred areas and edge features within them. By calculating statistics for high-frequency components and setting thresholds, blurred background edges caused by smoke obstruction can be more effectively identified. Furthermore, spatial wavelet analysis is robust to changes in lighting and noise, enabling the system to operate stably across diverse application scenarios and environmental variations.

[0067] Example 2

[0068] See also Figure 4, an embodiment of the present application provides a tunnel fire early warning method, which includes the following steps:

[0069] Step S101: Capture video images in the tunnel; Step S102: Perform motion foreground target extraction on the video images to extract suspicious motion areas including flames and / or smoke; Step S103: Use the HSI color space model to segment areas that meet flame color characteristics from the suspicious motion areas, and extract flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape changes based on the segmented flame suspected areas; Step S104: Send the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; Step S105: Based on the suspicious motion area, determine whether its main motion direction is consistent with the preset bottom-up direction, and if it is consistent, perform smoke texture feature detection to obtain smoke judgment result information; Step S106: Determine whether a fire has occurred based on the smoke judgment result information and the flame judgment result information to issue a corresponding early warning signal.

[0070] The specific implementation process of the above method can be referred to the tunnel fire warning system provided in Example 1, which will not be described in detail here.

[0071] It should be understood that although Figure 4 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. In addition, Figure 4 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0072] For example, step S103 and step S105 may be performed simultaneously or separately. If they are performed separately, the order of performing step S103 and step S105 is not limited.

[0073] Example 3

[0074] See also Figure 5The embodiment of the present application provides an electronic device, which includes at least one processor 201 and at least one memory 202; wherein the processor 201 and the memory 202 are directly connected to each other, or communicate with each other through a communication interface 203, or are electrically connected through one or more communication buses or signal lines to achieve data transmission or interaction; the memory 202 stores program instructions that can be executed by the processor 201, and the processor 201 calls the program instructions to execute a tunnel fire early warning method. For example, it implements:

[0075] Collect video images in the tunnel; perform motion foreground target extraction on the video images to extract suspicious motion areas including flames and / or smoke; use the HSI color space model to segment areas that meet flame color characteristics from the suspicious motion areas, and extract flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape changes based on the segmented flame suspected areas; send the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; based on the suspicious motion area, determine whether its main motion direction is consistent with the preset bottom-up direction, and if it is consistent, perform smoke texture feature detection to obtain smoke judgment result information; based on the smoke judgment result information and the flame judgment result information, determine whether a fire has occurred and issue a corresponding early warning signal.

[0076] Among them, the memory 202 can be, but is not limited to, random access memory (RAM), read only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0077] The processor 201 can be an integrated circuit chip with signal processing capabilities. The processor 201 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0078] I understand. Figure 5 The structure shown is only for illustration, and the electronic device may also include Figure 5 More or fewer components than shown, or with Figure 5 Different configurations shown. Figure 5 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0079] Example 4

[0080] The present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by the processor 201, a method for early warning of tunnel fire is implemented. For example, the method implements:

[0081] Collect video images in the tunnel; perform motion foreground target extraction on the video images to extract suspicious motion areas including flames and / or smoke; use the HSI color space model to segment areas that meet flame color characteristics from the suspicious motion areas, and extract flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape changes based on the segmented flame suspected areas; send the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; based on the suspicious motion area, determine whether its main motion direction is consistent with the preset bottom-up direction, and if it is consistent, perform smoke texture feature detection to obtain smoke judgment result information; based on the smoke judgment result information and the flame judgment result information, determine whether a fire has occurred and issue a corresponding early warning signal.

[0082] If the above functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0083] It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above and that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the present application is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A tunnel fire warning system, characterized in that: include: Video acquisition module, used to collect video images in the tunnel; A background extraction module is used to perform motion foreground target extraction processing on the video image to extract suspicious motion areas including flames and / or smoke; A flame extraction module is used to segment the area that meets the flame color characteristics from the suspicious motion area using the HSI color space model, and extract flame feature data including color, number of sharp corners, roundness, flame jump frequency and shape change based on the segmented flame suspected area; The flame judgment module is used to input the flame feature data into a pre-trained three-layer BP neural network for flame feature fusion to obtain flame judgment result information; a smoke determination module, configured to determine, based on the suspicious motion area, whether its main motion direction matches a preset bottom-up direction, and, if so, to perform smoke texture feature detection to obtain smoke determination result information; The fire determination module is used to determine whether a fire has occurred based on the smoke determination result information and the flame determination result information, so as to issue a corresponding early warning signal.

2. The system according to claim 1, wherein: The system also includes a pre-processing module for performing image enhancement and filtering processing on the video image sent to the background extraction module.

3. The system according to claim 1, wherein: The background extraction module further comprises: The background extraction unit is used to read each frame of the video image, perform grayscale histogram statistics on each pixel in each frame to find the point with the highest frequency, and calculate the energy contained in its neighborhood. If the energy exceeds 3 / 5 of the total energy of the entire histogram, the grayscale value corresponding to the point is used as the background grayscale value. Otherwise, another point with the highest frequency is searched outside the neighborhood as the background grayscale value, and the background image information is determined from the video image based on the background grayscale value. The motion region extraction unit is configured to compare the current frame image in the video image with the background image information to extract the suspicious motion region including flames and / or smoke.

4. The system according to claim 1, wherein: The flame extraction module further comprises: a flame suspected region extraction unit, configured to convert the suspicious motion region from an RGB image into an HSI image and feed the image into an HSI color space model, so as to traverse each pixel in the HSI image and determine whether it meets the threshold conditions of the flame color feature based on its H value, S value, and I value. If the H value, S value, and I value of a pixel simultaneously meet the set threshold conditions, the pixel is regarded as a pixel of the flame color feature and retained in the segmentation result, so as to obtain the flame suspected region based on the segmentation result; The flame characteristic data extraction unit is used to extract flame characteristic data including color, number of sharp corners, roundness, flame jump frequency and shape change according to the flame suspected area.

5. The system according to claim 4, characterized in that The H value, S value, and I value of a pixel point simultaneously satisfying the set threshold conditions include: the H value ∈ [0°, 60°], the S value ∈ [0.4, 1], and the I value ∈ [0.6, 1] of the pixel point.

6. The system according to claim 4, characterized in that The flame suspected area extraction unit is further used to perform expansion, erosion and edge detection processing on the segmentation result.

7. The system according to claim 1 or 4, characterized in that The three-layer BP neural network includes an input layer with 5 nodes, a hidden layer with 4 nodes, and an output layer with 1 node; Among them, the five input signals in the input layer are: the average value of the first-order color moment of the first preset number of consecutive suspected flame areas; the number of sharp corners of the suspected flame areas; the average value of the roundness of the second preset number of consecutive suspected flame areas; the number of high-frequency components greater than the first threshold obtained by wavelet decomposition of the height change curve of the third preset number of consecutive suspected flame areas; and the average value of the shape similarity of the fourth preset number of consecutive suspected flame areas.

8. The system according to claim 7, characterized in that The hidden layer and output layer of the three-layer BP neural network both use the sigmoid function as the neuron excitation function.

9. The system according to claim 1, wherein: The smoke texture feature detection includes distinguishing the smoke texture based on three features: the smoke source location is fixed; the early smoke blocks the background area so that the background edge is blurred; and the smoke is easy to diffuse.

10. The system according to claim 9, characterized in that The feature that the background edge is blurred due to the occlusion of the background area by the early smoke is detected by a spatial domain wavelet analysis algorithm.

Images