A method and system for fire monitoring and early warning in highway tunnels

Abstract

This application relates to the field of fire safety warning, and particularly to a method and system for monitoring and warning of fire in highway tunnels. The method includes the steps of: calculating the learning rate of each pixel point; updating the Gaussian background modeling parameters through the learning rate of each pixel point at each moment, and extracting the foreground in the monitoring picture, where the foreground is smoke and / or flame; classifying the foreground through an image classification algorithm to obtain the fire category and the non-fire category, and giving a warning according to the fire category. This application has the effect of improving the accuracy of fire alarm.

Description

Technical Field

[0001] This application relates to the field of fire safety warning, and particularly to a method and system for monitoring and warning of fire in highway tunnels. Background Art

[0002] The fire monitoring and warning in highway tunnels is usually realized by temperature-sensing cables. When the temperature-sensing cables sense the heat generated by a fire, the fire is monitored and warned according to the temperature change of the temperature-sensing cables. However, in highway tunnel scenarios, fires often start from local areas of vehicles or equipment, and the increase in temperature at the initial stage of the fire is not obvious. When the temperature-sensing cables generate an induction, the fire has often developed to a certain extent. Therefore, there is a phenomenon of alarm lag in monitoring and warning of highway tunnel fires by this method, which affects the rescue work of fires.

[0003] In the prior art, by collecting videos in highway tunnels and analyzing in real time whether there are early characteristics of tunnel fires in the videos, the phenomenon of alarm lag is alleviated to achieve faster fire warning. This method requires subtracting the background in the video to better monitor and warn of fires through the foreground of the video. Usually, Gaussian background modeling can be used to model the background in the video, separate the foreground and background of the image, so as to accurately detect abnormal phenomena in the tunnel (such as smoke, flame). For example, a video flame detection method and system disclosed in a Chinese patent application document with the publication number CN115909196A. The video flame detection method includes the following steps: S1, using a trained flame detector to detect the current frame of the video flame to obtain a preliminary selected flame target area; S2, screening the preliminary selected flame target area according to the RGB color model to obtain a more accurate reselected flame target area; S3, extracting the foreground flame of the reselected flame target area through Gaussian mixture background modeling to obtain a final selected flame target area without background information.

[0004] However, Gaussian background modeling needs to set a learning rate to update parameters. If the learning rate is set improperly, smoke or other fire characteristics in the image will be incorporated into the background, resulting in missed alarms, and thus may delay the alarm of tunnel fires. Summary of the Invention

[0005] To solve the above technical problem of how to set the learning rate, this application provides a method and system for monitoring and warning of fire in highway tunnels.

[0006] In a first aspect, this application provides a method for monitoring and warning of fire in highway tunnels, adopting the following technical solutions:

[0007] A method for monitoring and warning of fire in a highway tunnel includes the steps of: calculating the learning rate of each pixel point; updating the Gaussian background modeling parameters through the learning rate of each pixel point at each moment, and extracting the foreground in the monitoring image, where the foreground is smoke and / or flame; classifying the foreground through an image classification algorithm to obtain a fire category and a non-fire category, and giving a warning according to the fire category; the calculation formula of the learning rate is: ; in the formula, is the learning rate of the th pixel point in the monitoring image at the th moment, is the local similarity degree of the th pixel point in the monitoring image at the th moment, is the jitter amplitude of the monitoring device at the th moment, is the acceleration of the monitoring device at the th moment in the th direction, represents the first direction, represents the second direction, represents the third direction, is the exponential function with the natural constant as the base, is a preset parameter, is the maximum value function.

[0008] The beneficial effects are as follows: By calculating the learning rate of each pixel point, the Gaussian background modeling parameters can be dynamically adjusted according to factors such as the local similarity degree of the pixel point, the jitter amplitude of the monitoring device, and the acceleration. This enables the model to better adapt to dynamic changes in a complex tunnel environment, enhancing the sensitivity and robustness of the model. In the case of changes in tunnel lighting and jitter of the image caused by vehicle driving, the model can still accurately extract the foreground (smoke and / or flame) in the monitoring image, thereby improving the real-time performance and accuracy of the fire monitoring system and reducing false negative situations.

[0009] Optionally, the calculation of the jitter amplitude includes the steps of: performing feature point matching on the monitoring images of adjacent two seconds through a feature point matching algorithm, and the two matched feature points form a feature point pair; calculating the Euclidean distance between the coordinates of the two feature points in the feature point pair, using the Euclidean distance as the distance of the feature point pair, and clustering the distance feature point pairs to obtain two types of feature points; calculating the jitter amplitude, and the expression of the jitter amplitude is: ; in the formula, is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all the first-type feature point pairs between the monitoring image at this moment and the monitoring image of the previous second, It is the average value of the distances of all second - type feature - point pairs between the monitoring screen at this moment and the monitoring screen one second before.

[0010] The beneficial effects are as follows: By calculating the jitter amplitude of the monitoring screen through the coordinates of the feature points in each frame of the monitoring video, the model can better adapt to the slight jitter of the monitoring device, accurately reflect the jitter conditions of the monitoring device in different directions, thereby providing more accurate jitter - amplitude parameters for the calculation of the learning rate, helping to accurately extract the foreground even when the monitoring device has jitter, avoiding misjudgment caused by jitter, and further improving the reliability of the fire - monitoring system.

[0011] Optionally, the calculation of the jitter amplitude includes the steps: performing feature - point matching on the monitoring screens of adjacent two seconds through a feature - point matching algorithm, and the two matched feature points form a feature - point pair; calculating the Euclidean distance between the coordinates of the two feature points in the feature - point pair, taking the Euclidean distance as the distance of the feature - point pair, clustering the distance feature - point pairs to obtain two types of feature points; calculating the jitter amplitude, and the expression of the jitter amplitude is: , ; In the formula, is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all first - type feature - point pairs between the monitoring screen at this moment and the monitoring screen one second before, is the average value of the distances of all second - type feature - point pairs between the monitoring screen at this moment and the monitoring screen one second before, is a preset sensitivity parameter.

[0012] The beneficial effects are as follows: Compared with the previous jitter - amplitude calculation method, this method introduces a preset sensitivity parameter α, making the calculation of the jitter amplitude more flexible. By adjusting the value of α, the calculation result of the jitter amplitude can be more finely controlled, so as to better adapt to the jitter characteristics of different monitoring devices and the changes in the tunnel environment. For example, in the case where the jitter of the monitoring device is more obvious, the adaptability of the model to jitter can be enhanced by adjusting the value of α, further improving the robustness of the fire - monitoring system.

[0013] Optionally, the feature - point matching algorithm is the ORB algorithm.

[0014] Optionally, the clustering algorithm is the k - means clustering algorithm.

[0015] Optionally, the calculation formula of the local similarity degree is: ; In the formula, is the local similarity degree of the th pixel point in the monitoring screen at the th moment, is the local similarity degree of the th pixel point in the monitoring screen at the The grayscale value of a pixel point, is the average value of the grayscale values of all pixel points within the local range of the th pixel point in the monitoring screen at the th moment. is the variance of the grayscale values of the th pixel point in the monitoring screen from the initial moment to the th moment. is an exponential function with the natural constant as the base.

[0016] The beneficial effect is that by calculating the difference between the grayscale value of a pixel point and the average value of the grayscale values of all pixel points within its local range, and combining the variance of the grayscale values of the pixel point from the initial moment to the current moment, the local similarity degree of the pixel point can be accurately reflected. This calculation method can effectively distinguish the background and the foreground, avoiding misjudgment caused by factors such as background noise. In the fire monitoring and early warning of highway tunnels, by calculating the local similarity degree in this way, the accuracy of foreground detection can be improved, thereby improving the overall performance of the fire monitoring system.

[0017] In a second aspect, the present application provides a highway tunnel fire monitoring and early warning system, adopting the following technical solution:

[0018] A highway tunnel fire monitoring and early warning system includes: a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the highway tunnel fire monitoring and early warning method described above is implemented.

[0019] The beneficial effect is that the above-mentioned highway tunnel fire monitoring and early warning method is generated into a computer program and stored in the memory to be loaded and executed by the processor. Thus, a system is made according to the memory and the processor, which is convenient to use.

[0020] The present application has the following technical effects:

[0021] By calculating the jitter amplitude of the monitoring screen based on the coordinates of the feature points in each frame of the monitoring video, the model can better adapt to the slight jitter of the monitoring device. By calculating the local similarity degree of each pixel point based on the jitter amplitude of the monitoring screen and the gray value of the pixel point, and adjusting the learning rate through the local similarity degree, the stability of the background model can be maintained in the stable area, and it can adapt to changes faster in the dynamic area to highlight the possible fire areas. By calculating the learning rate of each pixel point based on the local similarity degree of each pixel point, the jitter amplitude of the monitoring device, and the vibration data of the monitoring device, the sensitivity and robustness of the model are enhanced, enabling the model to better adapt to the dynamic changes in the complex tunnel environment. By updating the Gaussian background modeling parameters through the learning rate of each pixel point, using Gaussian background modeling for foreground detection, and monitoring and warning for highway tunnel fires through the foreground of the monitoring screen, the real-time performance and accuracy of the fire monitoring system are improved. Description of the Drawings

[0022] By referring to the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present application will become readily understood. In the drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0023] Figure 1 It is a flowchart of a method for monitoring and warning highway tunnel fires according to an embodiment of the present application.

[0024] Figure 2 It is a flowchart of step S1 in a method for monitoring and warning highway tunnel fires according to an embodiment of the present application. Detailed Embodiments

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present application.

[0026] It should be understood that when terms such as "first" and "second" are used in the claims, specifications, and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "including" and "comprising" used in the specifications and claims of the present application indicate the existence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their collections.

[0027] The embodiments of the present application disclose a method for monitoring and warning of fire in highway tunnels. The parameters of Gaussian background modeling are updated through the learning rate of each pixel point. Gaussian background modeling is used for foreground detection, and the monitoring and warning of fire in highway tunnels is carried out through the foreground of the monitoring screen. By using this method for monitoring and warning of fire in highway tunnels, the complex changes in the monitoring screen can be more flexibly responded to, so that the model can not only adapt to the background update requirements caused by equipment jitter, but also maintain a high sensitivity to foreground features such as flames and smoke, improving the accuracy of monitoring and warning of fire in highway tunnels. Specifically, referring to Figure 1 , the method includes steps S1 - step S2, which are specifically as follows:

[0028] S1: Calculate the learning rate of each pixel point.

[0029] Calculate the jitter amplitude of the monitoring screen through the coordinates of the feature points in each frame of the monitoring video. Calculate the local similarity degree of each pixel point through the jitter amplitude of the monitoring screen and the gray value of the pixel point. Calculate the learning rate of each pixel point through the local similarity degree of each pixel point, the jitter amplitude of the monitoring device, and the vibration data of the monitoring device.

[0030] During the acquisition process of the tunnel monitoring video, there will be a situation where the monitoring device jitters, causing certain changes in the monitoring screen. In this case, although the monitoring screen has changed, the content of the screen has not changed substantially. In order to enable Gaussian background modeling to better adapt to the changes in the monitoring screen brought about by this phenomenon, and thus better distinguish between the foreground and background of the monitoring screen, the present application calculates the jitter amplitude of the monitoring screen according to the coordinates of the feature points in each frame of the monitoring video.

[0031] In one embodiment, referring to Figure 2 , the calculation of the jitter amplitude includes steps S10 - step S11, which are specifically as follows:

[0032] S10: Perform feature point matching on the monitoring screens of two adjacent seconds through a feature point matching algorithm, and the two matched feature points form a feature point pair.

[0033] The feature point matching algorithm can be ORB (Oriented FAST and Rotated BRIEF) or other feature matching algorithms in the existing technologies applicable to the application scenario of the present application, which will not be elaborated here.

[0034] S11: Calculate the Euclidean distance between the coordinates of the two feature points in the feature point pair. Use the Euclidean distance as the distance of the feature point pair, cluster the distance feature point pairs to obtain two types of feature points, and calculate the jitter amplitude.

[0035] The clustering algorithm can adopt the K-means algorithm or the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm, and the prior art will not be elaborated herein.

[0036] In one embodiment, the expression of the jitter amplitude is: ; where is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all first-type feature point pairs between the monitoring screen at this moment and the monitoring screen of the previous second, is the average value of the distances of all second-type feature point pairs between the monitoring screen at this moment and the monitoring screen of the previous second.

[0037] When the monitoring screen does not change at all, the Euclidean distance between the coordinates of the feature point pairs between the two monitoring screens is 0. When the monitoring device jitters, the jitter of the monitoring device will cause a slight change in the shooting angle, so that the same part in the tunnel will have a certain offset in the two images. and The greater the and the smaller the offset of the image, the smaller the jitter amplitude of the monitoring device.

[0038] Since there may be movement of vehicles or objects in the monitoring screen, the distance between the two points in the feature point pairs generated by the moving objects is caused by the movement of the objects, rather than the jitter of the monitoring device. Therefore, the distance of the feature point pairs generated by the movement of the objects should be greater than the distance of the feature point pairs generated by the jitter of the monitoring device. Therefore, the clustering algorithm uses the distance of the feature point pairs to cluster the feature point pairs to weaken the calculation of the jitter amplitude due to the movement of the objects. When there are feature point pairs generated by the movement of the objects among the two types of feature points, the greater the difference between the distances of the two types of feature point pairs. and the greater the gap between them, then and the greater the gap between them, making as 's weight, and making as 's weight, which can make and the smaller value in them has a greater weight, thereby weakening the influence of the feature point pairs generated by the movement of the objects on the calculation of the jitter amplitude.

[0039] In one embodiment, to make the calculation of the jitter amplitude more suitable for the current application scenario, the expression of the jitter amplitude can also be: , ; where, is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all first-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before, is the average value of the distances of all second-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before, is a preset sensitivity parameter.

[0040] Exemplarily, , when it is larger, and the weight difference between them is more obvious, and the implementer can adjust the value of to make the calculation of the jitter amplitude more suitable for the current application scenario.

[0041] The change of the monitoring screen caused by the jitter of the monitoring device will not cause large changes in all pixel points in the image. For example, if a pixel point in the image is in the same area before and after jitter (for example, within a brick), the color of the pixel point will not change greatly. To reduce unnecessary parameter updates and avoid the instability of the model, the present application calculates the local similarity degree of each pixel point through the jitter amplitude of the monitoring screen and the gray value of the pixel point.

[0042] In one embodiment, the calculation formula of the local similarity degree is:

[0043] ; where, is the local similarity degree of the th pixel point in the monitoring screen at the th moment, is the gray value of the th pixel point in the monitoring screen at the th moment, is the average value of the gray values of all pixel points within the local range of the th pixel point in the monitoring screen at the th moment, is the variance of the gray values of the th pixel point in the monitoring screen from the initial moment to the th moment, is the exponential function with the natural constant as the base.

[0044] represents the The local difference value of the gray value of the k-th pixel in the monitoring screen at a certain moment reflects the static local difference of this pixel. The larger this value is, the less similar the k-th pixel is to the surrounding pixels, and the smaller the local similarity degree of the pixel; the smaller this value is, the more similar the k-th pixel is to the surrounding pixels, and the larger the local similarity degree of the pixel.

[0045] represents the -th pixel's gray value change degree during the period from the initial moment to the -th moment, reflecting the dynamic local difference of this pixel. The larger this value is, the greater the change of the -th pixel in multiple moments, the more significant the fluctuation of this pixel in the time dimension. During the acquisition of the monitoring video, the more frequent the change of this pixel, the smaller the local similarity degree of this pixel; the smaller this value is, the smaller the change of the

[0046] -th pixel in multiple moments, the smoother the fluctuation of this pixel in the time dimension. During the acquisition of the monitoring video, the less frequent the change of this pixel, the larger the local similarity degree of this pixel.

[0046] In the process of Gaussian background modeling, parameters will be updated through the learning rate, enabling the model to distinguish static (background) areas and dynamic (foreground) areas in the monitoring screen. However, in the scenario of tunnel fire monitoring, due to the possible jitter of the monitoring device caused by environmental vibration or external impact, resulting in a slight offset of the overall picture, a fixed learning rate may not be able to effectively adapt to the subtle changes caused by this jitter. At the same time, when an incipient fire appears in the monitoring scenario, flames and smoke usually show slow and local dynamic changes with little gray value fluctuation. If the learning rate adjustment is unreasonable, it is very likely to misidentify the fire characteristics as the background area, thereby reducing the accuracy of fire detection. In order to more flexibly handle the complex changes in the monitoring screen, enabling the model to not only adapt to the background update requirements caused by device jitter and light changes, but also maintain a high sensitivity to foreground features such as flames and smoke, this application calculates the learning rate of each pixel according to the local similarity degree of each pixel, the jitter amplitude of the monitoring device, and the vibration data of the monitoring device.

[0047] In one embodiment, the calculation formula of the learning rate is:

[0048] ; where is the learning rate of the -th pixel in the monitoring screen at the -th moment, is the local similarity degree of the -th pixel in the monitoring screen at the -th moment, For the Monitor the jitter amplitude of the device at each moment. For the The monitoring device at the moment The acceleration in each direction, represents the first direction, Indicates the second direction, Indicates the third direction, is an exponential function with a natural constant as base, For the preset parameters, exemplary, , implementers can choose the size of this parameter according to actual conditions. is the maximum value function.

[0049] The larger the value is, the more stable the regional environment corresponding to the pixel point is at that moment. In order to maintain the stability of Gaussian background modeling, a lower learning rate should be used to update the Gaussian background modeling parameters of the pixel point. The smaller the value, the more dynamic the regional environment corresponding to the pixel point at that moment is. In order to adapt to sudden fluctuations in pixel values and avoid misjudging environmental interference as foreground, a higher learning rate should be used to update the Gaussian background modeling parameters of the pixel point.

[0050] The larger the value, the more obvious the overall jitter of the monitoring device is, which may cause the overall image offset. In order to avoid misidentifying the changes caused by jitter as the foreground, a larger learning rate should be used to quickly update the background model to avoid misidentifying the changes caused by jitter as the foreground. The smaller the value, the smoother the overall jitter of the monitoring equipment or even no jitter, which may cause the overall image offset. In order to maintain the stability of Gaussian background modeling, a smaller learning rate should be used to avoid fire features being mistakenly identified as background.

[0051] The larger the value, the more severe the vibration of the monitoring equipment. The grayscale value of the pixel in the monitoring image may change more. A larger learning rate should be used to quickly update the Gaussian background model to adapt to the image changes caused by vibration. The smaller it is, the more likely it is that the monitoring equipment has no vibration changes, and the grayscale value change of the pixel in the monitoring image is likely to be smaller. A smaller learning rate should be used to prevent excessive updating of the Gaussian background model parameters and avoid misjudging the initial fire characteristics (flame or smoke) as background.

[0052] S2: Update the Gaussian background modeling parameters through the learning rate of each pixel at each moment, extract the foreground in the monitoring image, and the foreground is smoke and / or flame; classify the foreground through the image classification algorithm to obtain the fire category and the no-fire category, and issue an early warning based on the fire category.

[0053] In the process of highway tunnel fire monitoring, the learning rate of each pixel point at each moment is calculated by the above method, and the parameters of Gaussian background modeling (the mean value, variance of each pixel point, and the weight of each distribution) are updated through the learning rate of each pixel point at each moment. The foreground in the monitoring screen is extracted in real time through the updated Gaussian background modeling parameters, and the foreground is classified by an image classification algorithm. One category is that there is a fire, and the other category is that there is no fire. When a fire is detected, a fire alarm is issued to remind passing vehicles to pay attention and notify the fire department to handle it.

[0054] The image classification algorithm can be the VGG (Visual Geometry Group) algorithm or other image classification algorithms in the prior art that can be applied to this application, which will not be elaborated here.

[0055] An embodiment of this application also discloses a highway tunnel fire monitoring and early warning system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the highway tunnel fire monitoring and early warning method according to this application is implemented.

[0056] The above system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art, so they will not be elaborated here.

[0057] In this application, the aforementioned memory can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. For example, the computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium. For example, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application program, module, or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0058] Although this specification has shown and described multiple embodiments of the present application, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and concept of the present application. It should be understood that various alternatives to the embodiments of the present application described herein may be adopted in the practice of the present application.

[0059] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereby. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A method for monitoring and warning of fire in a highway tunnel, characterized in that, Including the steps: Calculating the learning rate of each pixel point; updating the Gaussian background modeling parameters with the learning rate of each pixel point at each moment, and extracting the foreground in the monitoring screen, where the foreground is smoke and / or flame; classifying the foreground through an image classification algorithm to obtain the fire category and the non-fire category, and giving an early warning according to the fire category. The calculation formula for the learning rate is as follows: ; where is the learning rate of the th pixel point in the monitoring screen at the th moment, is the local similarity degree of the th pixel point in the monitoring screen at the th moment, is the jitter amplitude of the monitoring device at the th moment, is the acceleration of the monitoring device at the th moment in the th direction, represents the first direction, represents the second direction, represents the third direction, is the exponential function with the natural constant as the base, , is the maximum value function; The calculation formula for the local similarity degree is as follows: ; where is the gray value of the th pixel point in the monitoring screen at the th moment, is the average value of the gray values of all pixel points within the local range of the th pixel point in the monitoring screen at the th moment, is the variance of the gray values of the th pixel point in the monitoring screens from the initial moment to the th moment.

2. The highway tunnel fire situation monitoring and early warning method according to claim 1, characterized in that The calculation of the jitter amplitude includes the steps: Performing feature point matching on the monitoring screens of two adjacent seconds through a feature point matching algorithm, and the two matched feature points form a feature point pair. Calculate the Euclidean distance between the coordinates of two feature points in a feature point pair. Use the Euclidean distance as the distance of the feature point pair, and cluster the distance feature point pairs to obtain two types of feature points; calculate the jitter amplitude, and the expression for the jitter amplitude is: ; In the formula, is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all first-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before, is the average value of the distances of all second-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before.

3. The highway tunnel fire monitoring and early warning method according to claim 1, characterized in that, The calculation of the jitter amplitude includes the steps: Performing feature point matching on the monitoring screens of two adjacent seconds through a feature point matching algorithm, and the two matched feature points form a feature point pair. Calculate the Euclidean distance between the coordinates of two feature points in a feature point pair, and use the Euclidean distance as the distance of the feature point pair. Cluster the distance feature point pairs to obtain two types of feature points; calculate the jitter amplitude, and the expression of the jitter amplitude is: , ; In the formula, is the jitter amplitude of the monitoring device at any moment, is the average value of the distances of all first-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before, is the average value of the distances of all second-type feature point pairs between the monitoring screen at this moment and the monitoring screen one second before, is the preset sensitivity parameter.

4. The highway tunnel fire monitoring and early warning method according to claim 2, characterized in that, The feature point matching algorithm is the ORB algorithm.

5. The highway tunnel fire monitoring and early warning method according to claim 2, wherein The clustering algorithm is the k-means clustering algorithm.

6. A highway tunnel fire monitoring and early warning system, characterized in that, Including: A processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the highway tunnel fire monitoring and early warning method according to any one of claims 1-5 is implemented.

Images