Multi-source information fusion fire early warning method and system for intelligent Internet of Things

Abstract

The invention discloses a fire early warning method and system for multi-source information fusion of intelligent Internet of Things, and relates to the technical field of fire early warning, which can accurately extract a smoke region and improve the robustness of smoke detection through the construction of a local abnormal value index and a candidate abnormal cluster in combination with the spatial proximity adjustment of a fire source candidate region. Through combined use of the temperature and humidity sensor, the smoke sensor and image monitoring data, image analysis can be started in time in the early stage of a fire, misjudgment of a single sensor is avoided, a gray level enhancement method based on brightness compensation and a fire source candidate area screening mechanism based on edge fluctuation frequency are further provided, and the fire source candidate area screening accuracy is improved. According to the method, real flames are effectively distinguished from interference sources such as illumination and reflection by utilizing the dynamic characteristics of flame edge jitter, so that leak detection and false alarm are remarkably reduced, and the robustness of smoke detection is improved by constructing a local abnormal value index and a candidate abnormal cluster in combination with spatial proximity adjustment of a fire source candidate region.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fire warning, in particular to a multi-source information fusion fire warning method and system for intelligent Internet of Things. BACKGROUND

[0002] The fire warning is an intelligent safety system based on multi-sensor fusion and artificial intelligence, which monitors the environment abnormity (such as sudden increase of smoke concentration, sudden rise of temperature, flame radiation, etc.) in real time by deploying smoke detectors, temperature sensors, infrared thermal imagers and other devices, dynamically analyzes and identifies patterns of data by using deep learning algorithms, accurately distinguishes real fire from false alarms (such as dust interference), and thus issues sound and light alarms, mobile phone push or automatically links the fire fighting system at the initial stage of fire or potential risk stage, reduces the probability of fire spreading and loss, and effectively protects the safety of life and property.

[0003] The current fire warning method has many deficiencies. First, the traditional system usually relies on a single temperature and humidity sensor or smoke detector for alarm, which is easily disturbed by environmental temperature fluctuations, humidity changes or dust, steam and other factors, leading to false alarms or missed alarms, and cannot timely detect fire at the early stage of fire. Second, the flame detection based on images relies on static gray scale, color threshold or simple feature extraction, which is difficult to distinguish real flame from light changes, object reflections or highlight objects, and has high false alarm rate and missed detection rate. Third, the smoke detection method usually only focuses on overall pixel outliers or fixed thresholds, ignoring local outliers and their spatial proximity, resulting in inaccurate smoke area extraction and insufficient robustness. Finally, the existing system lacks real-time linkage mechanism of fire source and smoke, and the alarm information is isolated, which lags in response and is difficult to ensure the timeliness and reliability of the warning, which is particularly evident in complex scenarios. SUMMARY

[0004] To solve the above technical problems, the present application provides a multi-source information fusion fire warning method and system for intelligent Internet of Things, which solves the problem that the traditional system usually relies on a single temperature and humidity sensor or smoke detector for alarm.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] A multi-source information fusion fire warning method for intelligent Internet of Things, comprising:

[0007] acquiring temperature and humidity sensor data and smoke sensor data, obtaining a fire potential risk score according to the temperature and humidity sensor data and the smoke sensor data, and if the fire potential risk score is greater than a preset threshold, acquiring a monitoring picture;

[0008] According to the gray image of the monitoring picture, the average brightness of the gray image corresponding to the monitoring picture is obtained, and the brightness of the monitoring picture is adjusted according to the average brightness;

[0009] According to the gray image corresponding to the monitoring picture, a region with a gray value greater than a preset threshold is defined as a fire source candidate region, and an algorithm is used to obtain the edge of all fire source candidate regions and the perimeter of the edge of the corresponding fire source candidate region;

[0010] According to the perimeter of the edge of all fire source candidate regions, the edge fluctuation frequency corresponding to the fire source candidate region is obtained, and the fire source candidate region that meets the high-frequency jitter of the real flame edge is screened out based on the edge fluctuation frequency;

[0011] The monitoring picture is divided into regions of a fixed size, the local outlier index of each region is obtained, the candidate abnormal clusters are generated by clustering the abnormal pixels based on the local outlier index, and the smoke candidate region is obtained according to the spatial position relationship between the candidate abnormal clusters and the fire source candidate region;

[0012] A target detection model is obtained, the fire source candidate region and the smoke candidate region are input into the target detection model, and it is judged whether the monitoring picture contains a fire source or smoke;

[0013] A smart object connection system is obtained, and if the monitoring picture contains a fire source or smoke, an alarm is given through the smart object connection system.

[0014] Preferably, the temperature and humidity sensor data and the smoke sensor data are obtained, and the fire potential risk score is obtained according to the temperature and humidity sensor data and the smoke sensor data, specifically including:

[0015] A time window of 24 hours is taken, the smoke intensity mean value, smoke intensity standard deviation and current smoke intensity of the monitoring scene within the time window are obtained;

[0016] According to the smoke intensity mean value, smoke intensity standard deviation and current smoke intensity of the monitoring scene within the time window, a first risk score is obtained;

[0017] According to the temperature and humidity sensor data, noise data is removed based on data preprocessing, and a temperature and humidity heat map of the monitoring scene is obtained;

[0018] According to the temperature and humidity heat map of the monitoring scene, temperature and humidity abnormal data is obtained, the temperature and humidity abnormal data being data greater than a corresponding preset threshold;

[0019] According to the temperature and humidity abnormal data, a second risk score is obtained;

[0020] According to the first risk score and the second risk score, a fire potential risk score is obtained;

[0021] The first risk score is specifically:

[0022] ;

[0023] In the formula, The first risk score is, The function is, The current smoke intensity is, And The smoke intensity mean and the smoke intensity standard deviation of the monitoring scene in a time window are respectively;

[0024] The second risk score is specifically:

[0025] ;

[0026] ;

[0027] ;

[0028] In the formula, And The temperature coefficient and the humidity coefficient are respectively, The highest temperature in the monitoring scene is, The lowest temperature in the monitoring scene is, The highest humidity in the monitoring scene is, The lowest humidity in the monitoring scene is, The second risk score is;

[0029] The fire potential risk score is specifically:

[0030] ;

[0031] In the formula, The fire potential risk score is, The first risk score is, The second risk score is.

[0032] Preferably, the average brightness of the corresponding gray image of the monitoring picture is obtained according to the gray image of the monitoring picture, and the brightness of the monitoring picture is adjusted according to the average brightness, and specifically includes:

[0033] The average brightness of the corresponding gray image of the monitoring picture is obtained according to the gray image of the monitoring picture;

[0034] The historical monitoring picture is obtained, the historical monitoring picture is analyzed, and the reference brightness value is obtained;

[0035] The ratio of the average brightness of the corresponding gray image of the monitoring picture to the reference brightness value is taken as a compensation coefficient;​

[0036] adjusting the brightness of the monitoring picture according to a compensation coefficient;

[0037] The average brightness of the monitoring picture corresponding to the gray-scale image is specifically:

[0038]

[0039] In the formula, is the average brightness of the monitoring picture corresponding to the gray-scale image, is the width of the monitoring picture, is the height of the monitoring picture, is the gray-scale value of a point on the monitoring picture;

[0040] The compensation coefficient is specifically:

[0041]

[0042] In the formula, is the compensation coefficient, is a reference brightness value;

[0043] The adjusting of the brightness of the monitoring picture is specifically:

[0044]

[0045] In the formula, is the gray-scale value of a point on the adjusted monitoring picture.

[0046] Preferably, the edge fluctuation frequency corresponding to the fire source candidate region is obtained according to the perimeters of the edges of all the fire source candidate regions, and the fire source candidate region that meets the high-frequency jitter of the real flame edge is screened out based on the edge fluctuation frequency, specifically including:

[0047] obtaining monitoring pictures within 3 seconds, and converting the monitoring pictures within 3 seconds into gray-scale images;

[0048] defining a region with a gray-scale value greater than a preset threshold in the gray-scale image as a fire source candidate region;

[0049] extracting the circumscribed rectangle of all the fire source candidate regions, and obtaining the edge contour corresponding to all the fire source candidate regions based on a centroid tracking algorithm and a Canny edge algorithm;

[0050] obtaining the perimeters of the edge contours corresponding to all the fire source candidate regions;

[0051] regarding the perimeter data of the edge contours corresponding to the same fire source candidate region as a group of data, and obtaining a group of fire perimeter data corresponding to each fire source candidate region; ​​​

[0052] performing fast Fourier transform on the fire source perimeter data corresponding to each fire source candidate region to obtain edge fluctuation frequencies corresponding to all fire source candidate regions;

[0053] obtaining a fluctuation threshold according to the edge fluctuation frequencies corresponding to all fire source candidate regions;

[0054] adjusting the fire source candidate regions according to a comparison between the edge fluctuation frequencies corresponding to all fire source candidate regions and the corresponding fluctuation threshold;

[0055] The obtaining of the edge fluctuation frequencies corresponding to all fire source candidate regions is specifically:

[0056]

[0057]

[0058] In the formula, is the data of the i-th fire source candidate region in the monitoring picture within 3 seconds, is the i-th data of the j-th fire source candidate region, is the perimeter of the edge contour corresponding to the i-th fire source candidate region in the j-th monitoring picture, is the average value of the perimeter of the edge contour corresponding to the i-th fire source candidate region in the monitoring picture, is the edge fluctuation frequency of the i-th fire source candidate region, is the fast Fourier transform, is the j-th data of the i-th fire source candidate region; The fluctuation threshold is specifically: In the formula, is the fluctuation threshold of the i-th fire source candidate region, is the standard deviation of the edge fluctuation frequency of the i-th fire source candidate region,

[0059] is the average area of the i-th fire source candidate region.

[0060]

[0061] In the formula, is the fluctuation threshold of the i-th fire source candidate region, is the standard deviation of the edge fluctuation frequency of the i-th fire source candidate region, is the average area of the i-th fire source candidate region. Preferably, the obtaining of the monitoring picture within 3 seconds comprises converting the monitoring picture within 3 seconds into a gray-scale image, specifically comprising:

[0062]

[0063] ​​​​​​​​​Dividing the monitoring picture into 8*8 regions, obtaining the mean and standard deviation of the gray value of all 8*8 regions;

[0064] According to the average gray value and the standard deviation of each region, obtaining the local outlier index corresponding to each region;

[0065] Based on the connected domain search algorithm and the local outlier index, if two abnormal pixels have a common edge, the two abnormal pixels are regarded as a candidate abnormal cluster, and the abnormal pixels are the regions whose local outlier index is not in the preset interval;

[0066] If there is an abnormal pixel whose distance to a candidate abnormal cluster is less than 8 pixels, and the absolute value of the difference between its local outlier index and the preset threshold is less than 0.1, it is added to the corresponding candidate abnormal cluster;

[0067] According to the obtained fire source candidate region, the candidate abnormal cluster is adjusted;

[0068] The adjusted candidate abnormal cluster is taken as a smoke candidate region;

[0069] Among them, the local outlier index is specifically:

[0070] ;

[0071] In the formula, and Take 1-8, is the local outlier index of the i-th 8*8 region, is the local outlier index of the i-th 8*8 region, is the local outlier index of the i-th 8*8 region, is the local outlier index of the i-th 8*8 region, is the gray value of the i-th 8*8 region, is the gray value of the i-th 8*8 region, is the gray value of the i-th 8*8 region, is the gray value of the i-th 8*8 region, is the mean of the gray value of the i-th 8*8 region, is the standard deviation of the gray value of the i-th 8*8 region.

[0072] Preferably, the adjustment of the candidate abnormal cluster according to the obtained fire source candidate region specifically includes:

[0073] According to the gray value of the fire source candidate region, the centroid of the fire source is obtained;

[0074] According to the gray value of the fire source candidate region and the centroid of the fire source, the inflation radius is obtained;

[0075] Based on the algorithm and the inflation radius, the fire source candidate region is morphologically dilated to obtain an expanded fire source candidate region; ​​

[0076] If the Euclidean distance between the expanded candidate fire source region and the candidate anomaly cluster is less than a preset threshold, then the candidate anomaly cluster is retained.

[0077] Obtain the area change curves of all candidate anomaly clusters within 3 seconds. If the curves continue to increase, retain the corresponding candidate anomaly clusters.

[0078] Specifically, the centroid of the fire source is:

[0079] ;

[0080] ;

[0081] ;

[0082] In the formula, The x-coordinate of the centroid The ordinate of the centroid is y. This represents the number of pixels contained in the candidate fire source region. For the first The grayscale value of each pixel For the first The x-coordinate of each pixel For the first The ordinate of each pixel The center of mass of the fire source;

[0083] Specifically, the expansion radius is:

[0084] ;

[0085] In the formula, Where is the radius of expansion. This represents the number of pixels contained in the candidate fire source region. The first candidate region for fire source The grayscale value of each pixel The first candidate region for fire source The x-coordinate of each pixel The first candidate region for fire source The ordinate of each pixel and These are the x-coordinate and y-coordinate of the centroid of the fire source, respectively.

[0086] Furthermore, a fire early warning system based on multi-source information fusion for the Internet of Things (IoT) is proposed to realize the fire early warning method based on multi-source information fusion for the IoT as described above, including:

[0087] A main control module is configured to acquire a fire potential risk score, a fire source candidate area and a smoke candidate area according to the temperature and humidity sensor data and the smoke sensor data, input the fire source candidate area and the smoke candidate area into a target detection model, determine whether the monitoring picture contains a fire source or smoke, and if the monitoring picture contains a fire source or smoke, alarm through the intelligent object connection system.

[0088] A risk score module is configured to acquire a first risk score and a second risk score according to the temperature and humidity sensor data and the smoke sensor data, and acquire a fire potential risk score according to the first risk score and the second risk score.

[0089] A fire source candidate area acquisition module is configured to define an area with a gray value greater than a preset threshold in a gray image as a fire source candidate area, and adjust the fire source candidate area according to a comparison between an edge fluctuation frequency corresponding to all fire source candidate areas and a corresponding fluctuation threshold.

[0090] Optionally, the main control module specifically comprises:

[0091] A data acquisition unit is configured to acquire the temperature and humidity sensor data and the smoke sensor data, acquire the fire potential risk score, and acquire an average brightness of a corresponding gray image of the monitoring picture according to a gray image of the monitoring picture.

[0092] A warning unit is configured to input the fire source candidate area and the smoke candidate area into the target detection model, determine whether the monitoring picture contains a fire source or smoke, and if the monitoring picture contains a fire source or smoke, alarm through the intelligent object connection system.

[0093] Optionally, the risk score module specifically comprises:

[0094] A first risk score acquisition unit is configured to acquire the first risk score according to a smoke intensity mean value, a smoke intensity standard deviation and a current smoke intensity of a monitoring scene within a time window.

[0095] A second risk score acquisition unit is configured to acquire temperature and humidity anomaly data according to a temperature and humidity thermal map of the monitoring scene, and acquire the second risk score according to the temperature and humidity anomaly data.

[0096] A risk score acquisition unit is configured to acquire the fire potential risk score according to the first risk score and the second risk score.

[0097] Optionally, the fire source candidate area acquisition module specifically comprises:

[0098] A perimeter acquisition unit is configured to define a region with a gray value greater than a preset threshold in the gray image as a fire source candidate region, and acquire a perimeter of an edge contour corresponding to all fire source candidate regions.

[0099] A fire source candidate region adjustment unit is configured to acquire an edge fluctuation frequency corresponding to all fire source candidate regions, and adjust the fire source candidate regions according to a comparison between the edge fluctuation frequency corresponding to all fire source candidate regions and a corresponding fluctuation threshold.

[0100] Compared with the prior art, the present application has the following beneficial effects:

[0101] The present application provides a fire warning method and system for intelligent Internet of Things multi-source information fusion, which uses temperature and humidity sensors, smoke sensors and image monitoring data in combination, can start image analysis in time in the early stage of fire, avoids misjudgment of a single sensor, and proposes a gray scale enhancement method based on brightness compensation and a fire source candidate region screening mechanism based on edge fluctuation frequency, uses the dynamic characteristics of flame edge jitter to effectively distinguish real flames from interference sources such as light and reflection, thereby significantly reducing missed detection and false positives, and through the construction of local outliers and candidate abnormal clusters, the spatial proximity adjustment of the fire source candidate region is combined to improve the robustness of smoke detection. BRIEF DESCRIPTION OF DRAWINGS

[0102] Figure 1 A flow chart of the fire warning method for intelligent Internet of Things multi-source information fusion is provided in the present application.

[0103] Figure 2 A flow chart of the fire potential risk score acquisition in the present application is provided.

[0104] Figure 3 A flow chart of the brightness adjustment of the monitoring picture in the present application is provided.

[0105] Figure 4 A fire source candidate region binary graph in the present application is provided.

[0106] Figure 5 An extended fire source candidate region binary graph in the present application is provided.

[0107] Figure 6 A structure block diagram of the fire warning system for intelligent Internet of Things multi-source information fusion is provided in the present application. DETAILED DESCRIPTION

[0108] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only used as examples, and other obvious modifications can be thought of by those skilled in the art.

[0109] REFERENCE Figure 1- Figure 3 The fire warning method for intelligent Internet of Things multi-source information fusion in the embodiment of the application shown in the embodiment of the application comprises:

[0110] obtaining temperature and humidity sensor data and smoke sensor data, obtaining a fire potential risk score according to the temperature and humidity sensor data and the smoke sensor data, and obtaining a monitoring picture if the fire potential risk score is greater than a preset threshold value;

[0111] obtaining the average brightness of the corresponding gray image of the monitoring picture according to the gray image of the monitoring picture, and adjusting the brightness of the monitoring picture according to the average brightness;

[0112] defining a region with a gray value greater than a preset threshold value as a fire source candidate region according to the corresponding gray image of the monitoring picture, and obtaining the edges of all the fire source candidate regions and the perimeters of the edges of the corresponding fire source candidate regions using an algorithm;

[0113] obtaining the edge fluctuation frequency of the fire source candidate region according to the perimeters of the edges of all the fire source candidate regions, and screening out the fire source candidate region that meets the high-frequency jitter of the real flame edge based on the edge fluctuation frequency;

[0114] dividing the monitoring picture into regions of a fixed size, obtaining a local outlier indicator of each region, clustering the abnormal pixels based on the local outlier indicator to generate a candidate abnormal cluster, and obtaining a smoke candidate region according to the spatial positional relationship between the candidate abnormal cluster and the fire source candidate region;

[0115] obtaining a target detection model, inputting the fire source candidate region and the smoke candidate region into the target detection model, and judging whether the monitoring picture contains a fire source or smoke;

[0116] obtaining an intelligent Internet of Things system, and alarming through the intelligent Internet of Things system if the monitoring picture contains a fire source or smoke;

[0117] In the embodiment, in the step of obtaining temperature and humidity sensor data and smoke sensor data, obtaining a fire potential risk score according to the temperature and humidity sensor data and the smoke sensor data, and obtaining a monitoring picture if the fire potential risk score is greater than a preset threshold value, the preset threshold value is 0.4;

[0118] In the step of defining a region with a gray value greater than a preset threshold value as a fire source candidate region according to the corresponding gray image of the monitoring picture, and obtaining the edges of all the fire source candidate regions and the perimeters of the edges of the corresponding fire source candidate regions using an algorithm, the preset threshold value is 150.

[0119] Specifically, the step of obtaining temperature and humidity sensor data and smoke sensor data, obtaining a fire potential risk score according to the temperature and humidity sensor data and the smoke sensor data, specifically comprises:

[0120] obtaining a smoke intensity mean value, a smoke intensity standard deviation and a current smoke intensity of the monitoring scene in a time window of 24 hours;

[0121] obtaining a first risk score according to the smoke intensity mean value, the smoke intensity standard deviation and the current smoke intensity of the monitoring scene in a time window;

[0122] obtaining a temperature and humidity heat map of the monitoring scene based on data preprocessing to remove noise data according to the temperature and humidity sensor data;

[0123] obtaining temperature and humidity abnormal data according to the temperature and humidity heat map of the monitoring scene, the temperature and humidity abnormal data being data greater than a corresponding preset threshold;

[0124] obtaining a second risk score according to the temperature and humidity abnormal data;

[0125] obtaining a fire potential risk score according to the first risk score and the second risk score;

[0126] wherein the first risk score is specifically:

[0127] ;

[0128] wherein, is the first risk score, is a function, is the current smoke intensity, and are the smoke intensity mean value and the smoke intensity standard deviation of the monitoring scene in a time window, respectively;

[0129] wherein the second risk score is specifically:

[0130] ;

[0131] ;

[0132] ;

[0133] wherein, and are a temperature coefficient and a humidity coefficient, respectively, is a maximum temperature in the monitoring scene, is a minimum temperature in the monitoring scene, is a maximum humidity in the monitoring scene, is a minimum humidity in the monitoring scene, is the second risk score;

[0134] wherein the fire potential risk score is specifically:

[0135] ;

[0136] In the formula, is a fire potential risk score, is a first risk score, is a second risk score;

[0137] In this embodiment, the temperature corresponding to the preset threshold is 50℃, and the humidity corresponding to the preset threshold is 40%.

[0138] It should be noted that the temperature and humidity data after preprocessing can be mapped to a visual temperature and humidity heat map, which can more intuitively reflect the temperature and humidity conditions of different positions in the region. The "spatial dimension" of the heat map corresponds to the physical region of the monitoring scene (such as a room, a floor, a park, etc.), and each pixel or region block represents the temperature and humidity state of a specific position in the monitoring scene. The "numerical dimension" of the heat map distinguishes the temperature and humidity size through the color gradient (such as high temperature / high humidity areas marked in red and orange, and low temperature / low humidity areas marked in blue and green), which intuitively presents the distribution difference of temperature and humidity in space, facilitates quick positioning of numerical anomaly areas, and combines the color marking and spatial position of the heat map to screen out two types of data that meet the "greater than the preset threshold" condition, that is, in the temperature heat map, the temperature data corresponding to the area with a temperature value greater than 50℃ is temperature abnormal data (for example, a room corner heat map is displayed in deep red, and the temperature value is 55℃, so the temperature data of 55℃ is abnormal data), and the humidity data corresponding to the area with a humidity value greater than 40% is humidity abnormal data (for example, the humidity heat map around a certain device is displayed in orange, and the humidity value is 48%, so the humidity data of 48% is abnormal data). The temperature and humidity data greater than the preset threshold extracted as temperature and humidity abnormal data.

[0139] Specifically, the average brightness of the gray-scale image corresponding to the monitoring picture is obtained according to the gray-scale image of the monitoring picture, and the brightness of the monitoring picture is adjusted according to the average brightness, specifically including:

[0140] According to the gray-scale image of the monitoring picture, the average brightness of the gray-scale image corresponding to the monitoring picture is obtained;

[0141] The reference brightness value is obtained by analyzing the historical monitoring picture;

[0142] The ratio of the average brightness of the gray-scale image corresponding to the monitoring picture to the reference brightness value is taken as a compensation coefficient;

[0143] The brightness of the monitoring picture is adjusted according to the compensation coefficient;

[0144] wherein the average brightness of the monitoring picture corresponding to the gray scale image is specifically:

[0145] ;

[0146] wherein, is the average brightness of the monitoring picture corresponding to the gray scale image, is the width of the monitoring picture, is the height of the monitoring picture, is the gray scale value of the point on the monitoring picture;

[0147] wherein the compensation coefficient is specifically:

[0148] ;

[0149] wherein, is the compensation coefficient, is the reference brightness value;

[0150] wherein the adjusting the brightness of the monitoring picture is specifically:

[0151] ;

[0152] wherein, is the gray scale value of the point on the adjusted monitoring picture.

[0153] In the scheme, the accuracy and robustness of fire warning are significantly improved through multi-source information fusion and dynamic brightness compensation mechanism. Compared with the traditional single sensor warning method, the comprehensive risk scoring mechanism based on temperature, humidity and smoke data effectively avoids false positives caused by environmental interference. For example, in a kitchen environment, the traditional system may misjudge cooking steam (high humidity) and oil smoke (similar to smoke) as fire. However, the scheme automatically reduces the risk score when the humidity exceeds 40% or the temperature is lower than 50℃ by analyzing the 24-hour smoke intensity average, standard deviation and temperature and humidity heat map, thereby avoiding misjudgment. At the same time, the gray scale enhancement method based on brightness compensation solves the image processing problem under different lighting conditions. The traditional system uses a fixed brightness threshold and cannot adapt to changes in light. However, the method obtains a reference brightness value from historical monitoring pictures, uses the average brightness of the historical monitoring picture corresponding to the gray scale image as the reference brightness value, dynamically adjusts the brightness of the current picture, makes the flame edge features clearer, and effectively distinguishes real flames from light reflection interference. For example, in a scene where the sun is shining directly, the traditional system may misjudge the window reflection as a flame. However, after brightness compensation, the method can accurately identify the jitter characteristics of the flame edge, reducing the false positive rate.

[0154] Specifically, the step of obtaining the edge fluctuation frequency corresponding to the edge of all candidate fire source regions based on the perimeter of the edge of each candidate fire source region, and filtering out candidate fire source regions that conform to the high-frequency jitter of the edge of a real flame based on the edge fluctuation frequency, specifically includes:

[0155] Acquire monitoring images within 3 seconds and convert them into grayscale images.

[0156] Regions in a grayscale image with grayscale values ​​greater than a preset threshold are defined as candidate fire source regions.

[0157] Extract the bounding rectangles of all fire source candidate regions, and obtain the edge contours of all fire source candidate regions based on the centroid tracking algorithm and the Canny edge algorithm;

[0158] Obtain the perimeter of the edge contours corresponding to all candidate fire source regions;

[0159] Take the edge contour perimeter data corresponding to the same fire source candidate area as a group of data, and obtain the fire source perimeter data group corresponding to each fire source candidate area.

[0160] Perform a fast Fourier transform on the perimeter data set corresponding to each candidate fire source region to obtain the edge fluctuation frequency corresponding to all candidate fire source regions.

[0161] The fluctuation threshold is obtained based on the edge fluctuation frequency corresponding to all candidate fire source regions;

[0162] The fire source candidate regions are adjusted based on a comparison of the edge fluctuation frequency and the corresponding fluctuation threshold of all fire source candidate regions.

[0163] Specifically, obtaining the edge fluctuation frequency corresponding to all candidate fire source regions involves:

[0164] ;

[0165] ;

[0166] In the formula, To monitor the image within 3 seconds Data on candidate fire source areas, For the first The first candidate area for fire source The first frame in the monitored image The perimeter of the edge contour corresponding to each candidate fire source region. To monitor the first in the image The average perimeter of the edge contour corresponding to each candidate fire source region For the first The edge fluctuation frequency of each candidate ignition source region For Fast Fourier Transform, For the first The first candidate area for fire source One data point;

[0167] Specifically, the fluctuation threshold is:

[0168] ;

[0169] In the formula, For the first Fluctuation threshold of each fire source candidate region For the first The standard deviation of the edge fluctuation frequency of each candidate ignition source region For the first The average area of ​​each candidate fire source region.

[0170] In this embodiment, the preset threshold corresponding to the gray value of the fire source candidate region is defined as 150.

[0171] Specifically, the process involves dividing the monitored image into regions of fixed size, obtaining local outlier indicators for each region, clustering outlier pixels based on these indicators to generate candidate outlier clusters, and then obtaining smoke candidate regions based on the spatial relationship between the candidate outlier clusters and the fire source candidate regions. This process specifically includes:

[0172] The monitoring image was divided into 8*8 regions, and the mean and standard deviation of the gray values ​​of all 8*8 regions were obtained.

[0173] Based on the average gray level and standard deviation of each region, obtain the local outlier index corresponding to each region;

[0174] Based on the connected component search algorithm and the local outlier index, if two outlier pixels have a common edge, then these two outlier pixels are regarded as a candidate outlier cluster. The outlier pixels are the regions where the local outlier index is not within the preset range.

[0175] If an anomalous pixel is less than 8 pixels away from a candidate anomalous cluster, and the absolute difference between its local outlier index and a preset threshold is less than 0.1, then it is added to the corresponding candidate anomalous cluster.

[0176] Adjust the candidate anomaly clusters based on the already obtained candidate fire source regions;

[0177] The adjusted candidate anomaly clusters are used as smoke candidate regions;

[0178] Specifically, the local outlier index is:

[0179] ;

[0180] In the formula, and Take 1-8, For the first The first 8*8 area Line 1 Local outlier index of the column, For the first The first 8*8 area Line 1 The grayscale value of the column, For the first The average grayscale value of an 8x8 region For the first The standard deviation of gray values ​​in an 8x8 area;

[0181] In this embodiment, based on the connected component search algorithm and the local outlier index, if two outlier pixels have a common edge, then these two outlier pixels are regarded as a candidate outlier cluster. The outlier pixels are in the region where the local outlier index is not within a preset range, which is 0.3-0.8.

[0182] If an anomalous pixel is less than 8 pixels away from a candidate anomaly cluster, and the absolute difference between its local anomaly index and a preset threshold is less than 0.1, then it is added to the corresponding candidate anomaly cluster. The preset threshold is 0.3 or 0.8, and the threshold is chosen to be the smaller of the difference between the anomalous pixel's local anomaly index and the absolute values ​​of 0.3 and 0.8. For example, if an anomalous pixel is 5 pixels away from a candidate anomaly cluster and its local anomaly index is 0.35, then the preset threshold is 0.3 (0.35-0.3=0.05, 0.8-0.35=0.55, 0.05 is less than 0.55, therefore...). If the preset threshold is 0.3, and the absolute value difference between the local outlier index of the abnormal pixel and the preset threshold is 0.05, which is less than 0.1, then the abnormal pixel is added to the candidate anomaly cluster. If the distance between an abnormal pixel and a candidate anomaly cluster is 5 pixels, and the local outlier index is 0.68, then the preset threshold is 0.8 (0.68-0.3=0.38, 0.8-0.68=0.18, 0.18 is less than 0.38, so the preset threshold is 0.8). The absolute value difference between the local outlier index of the abnormal pixel and the preset threshold is 0.12, which is greater than 0.1, so the abnormal pixel is not added to the candidate anomaly cluster.

[0183] Specifically, adjusting the candidate anomaly clusters based on the already acquired candidate fire source regions includes:

[0184] Based on the grayscale values ​​of the candidate fire source region, obtain the centroid of the fire source;

[0185] The expansion radius is obtained based on the grayscale value of the candidate fire source region and the centroid of the fire source;

[0186] Based on the algorithm and the expansion radius, morphological expansion processing is performed on the candidate fire source region to obtain an expanded candidate fire source region;

[0187] If the Euclidean distance between the expanded candidate fire source region and the candidate anomaly cluster is less than a preset threshold, then the candidate anomaly cluster is retained.

[0188] Obtain the area change curves of all candidate anomaly clusters within 3 seconds. If the curves continue to increase, retain the corresponding candidate anomaly clusters.

[0189] Specifically, the centroid of the fire source is:

[0190] ;

[0191] ;

[0192] ;

[0193] In the formula, The x-coordinate of the centroid The ordinate of the centroid is y. This represents the number of pixels contained in the candidate fire source region. For the first The grayscale value of each pixel For the first The x-coordinate of each pixel For the first The ordinate of each pixel The center of mass of the fire source;

[0194] Specifically, the expansion radius is:

[0195] ;

[0196] In the formula, Where is the radius of expansion. This represents the number of pixels contained in the candidate fire source region. The first candidate region for fire source The grayscale value of each pixel The first candidate region for fire source The x-coordinate of each pixel The first candidate region for fire source The ordinate of each pixel and These are the x-coordinate and y-coordinate of the centroid of the fire source, respectively;

[0197] In this embodiment, the morphological dilation of the fire source candidate region based on the algorithm and the dilation radius is performed to obtain an expanded fire source candidate region. The reason for performing morphological dilation on the fire source candidate region is as follows:

[0198] 1. The flame edges jitter strongly, which may cause some pixels to be missed in a certain frame (insufficient grayscale value, occlusion, lighting effect). The dilation operation can avoid missed detection and make the flame area a whole.

[0199] 2. In actual fires, smoke often appears around flames. If you only look at the original area of ​​the flame, you may find that the "flame" and "smoke candidate area" do not intersect in the image. After the flame expands, it is easier to capture the spatial proximity relationship between the smoke and the flame, thereby reducing false alarms.

[0200] 3. The brightness of edge pixels may fluctuate greatly (sometimes being judged as not flames). After expansion, the radius and centroid calculations are smoother and will not cause drastic fluctuations due to individual missing points.

[0201] Reference Figure 4 As shown, it is understandable that if the grayscale image of the fire source candidate region is further binarized, where 1 represents the flame pixel and 0 represents the background, and the region corresponding to 1 is used as the fire source candidate region, it is easy to see that the edge of the fire source candidate region is missing. It may have been filtered out because the grayscale value is less than a preset threshold. Therefore, it is necessary to perform dilation processing on the above-mentioned fire source candidate region. The grayscale image of the dilated fire source candidate region is then binarized as follows: Figure 5 As shown, it can be clearly seen that the flame edge of the expanded fire source candidate area is filled in. If there is a candidate anomalous cluster near the fire source candidate area and the Euclidean distance between it and the cluster is less than the preset threshold, the candidate anomalous cluster can be retained as a smoke candidate area.

[0202] If the Euclidean distance between the expanded candidate fire source region and the candidate anomaly cluster is less than a preset threshold, then the preset threshold in the step of retaining the candidate anomaly cluster is specifically one-quarter of the height of the monitoring image. For example, if the height of the monitoring image is 480 pixels, then the preset threshold is 120 pixels.

[0203] In the scheme, a multi-level, dynamic collaborative visual feature analysis and verification framework is constructed. Traditional methods usually rely on the static color or brightness threshold of the flame, which is easily disturbed by light changes and high-light reflective objects (such as lanterns and sunlight reflections), resulting in high false alarm rate. The scheme improves the early warning accuracy in complex scenes through a three-level progressive mechanism of dynamic frequency feature analysis, local anomaly clustering and spatial proximity verification. The frequency domain analysis based on the edge contour perimeter sequence is introduced to capture the high-frequency jittering feature of the flame, which distinguishes real flames (such as flickering candles) from static interference sources (such as stationary light bulbs) from a dynamic dimension. The connected component clustering algorithm based on the local outlier index is used to divide the image into 8*8 sub-regions for microscopic analysis, effectively identifying the sparse smoke that is missed by traditional global threshold methods in the early stage of diffusion. Finally, a systematic fusion verification mechanism is introduced to calculate the fire source centroid and inflation radius, perform morphological inflation on the fire source candidate area, and determine the spatial proximity of the smoke candidate area to the fire source. Smoke regions adjacent to the fire source or with continuous area growth are considered as smoke candidate regions, reducing the false detection rate of mist outside doors and windows or dust in the room being mistaken for fire smoke.

[0204] Referring to Figure 6 Further, in combination with the above-mentioned fire warning method based on multi-source information fusion of smart Internet of Things, a fire warning system based on multi-source information fusion of smart Internet of Things is proposed, which comprises:

[0205] A main control module is configured to obtain a fire potential risk score, a fire source candidate area and a smoke candidate area according to the temperature and humidity sensor data and the smoke sensor data, input the fire source candidate area and the smoke candidate area into a target detection model, and determine whether the monitoring picture contains a fire source or smoke. If the monitoring picture contains a fire source or smoke, an alarm is triggered through the smart Internet of Things system.

[0206] A risk score module is configured to obtain a first risk score and a second risk score according to the temperature and humidity sensor data and the smoke sensor data, and obtain a fire potential risk score according to the first risk score and the second risk score.

[0207] A fire source candidate area acquisition module is configured to define the area with a gray value greater than a preset threshold in the gray image as a fire source candidate area, and adjust the fire source candidate area according to the comparison between the edge fluctuation frequency corresponding to all fire source candidate areas and the corresponding fluctuation threshold.

[0208] The main control module specifically comprises:

[0209] The data acquisition unit is configured to acquire temperature and humidity sensor data and smoke sensor data, acquire a fire potential risk score, and acquire an average brightness of a gray image corresponding to the monitoring picture according to a gray image of the monitoring picture.

[0210] The early warning unit is configured to input the fire source candidate area and the smoke candidate area to a target detection model, judge whether the monitoring picture contains a fire source or smoke, and if the monitoring picture contains a fire source or smoke, alarm through the intelligent object system.

[0211] The risk score module specifically comprises:

[0212] The first risk score acquisition unit is configured to acquire a first risk score according to a smoke intensity average value, a smoke intensity standard deviation and a current smoke intensity of a monitoring scene within a time window.

[0213] The second risk score acquisition unit is configured to acquire temperature and humidity anomaly data according to a temperature and humidity thermal map of the monitoring scene, and acquire a second risk score according to the temperature and humidity anomaly data.

[0214] The risk score acquisition unit is configured to acquire a fire potential risk score according to the first risk score and the second risk score.

[0215] The fire source candidate area acquisition module specifically comprises:

[0216] The perimeter acquisition unit is configured to define an area with a gray value greater than a preset threshold in the gray image as a fire source candidate area, and acquire a perimeter of an edge contour corresponding to all fire source candidate areas.

[0217] The fire source candidate area adjustment unit is configured to acquire an edge fluctuation frequency corresponding to all fire source candidate areas, and adjust the fire source candidate area according to a comparison between the edge fluctuation frequency corresponding to all fire source candidate areas and a corresponding fluctuation threshold.

[0218] In summary, the advantages of the present application are that the temperature and humidity sensor, the smoke sensor and the image monitoring data are jointly used, the image analysis can be started in time in the early stage of the fire, the single sensor misjudgment is avoided, the gray scale enhancement method based on brightness compensation and the fire source candidate area screening mechanism based on the edge fluctuation frequency are proposed, the dynamic characteristics of the flame edge jitter are utilized to effectively distinguish the real flame from the interference sources such as light and reflection, so that the missed detection and false alarm are significantly reduced, and the robustness of the smoke detection is improved through the construction of the local outlier index and the candidate abnormal cluster and the adjustment of the spatial proximity of the fire source candidate area.

[0219] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A fire early warning method based on multi-source information fusion for smart IoT, characterized in that, include: Acquire data from temperature and humidity sensors and smoke sensors. Based on the temperature and humidity sensor data and smoke sensor data, obtain a potential fire risk score. If the potential fire risk score is greater than a preset threshold, acquire monitoring images. Based on the grayscale image of the monitored image, obtain the average brightness of the corresponding grayscale image, and adjust the brightness of the monitored image according to the average brightness; Based on the grayscale image corresponding to the monitored image, the area with a grayscale value greater than a preset threshold is defined as a fire source candidate area. The algorithm is used to obtain the edge of all fire source candidate areas and the perimeter of the corresponding fire source candidate area edge. Based on the perimeter of the edges of all candidate fire source regions, the corresponding edge fluctuation frequency of the candidate fire source regions is obtained, and candidate fire source regions that conform to the high-frequency jitter of the real flame edge are selected based on the edge fluctuation frequency. The monitoring image is divided into regions of fixed size, and local outlier indicators are obtained for each region. Based on the local outlier indicators, abnormal pixels are clustered to generate candidate outlier clusters. Then, based on the spatial relationship between the candidate outlier clusters and the fire source candidate regions, smoke candidate regions are obtained. Obtain the target detection model, input the fire source candidate region and smoke candidate region into the target detection model, and determine whether the monitored image contains a fire source or smoke. If the monitored image contains a fire source or smoke, an alarm will be triggered through the smart IoT system.

2. The fire early warning method based on multi-source information fusion for smart IoT as described in claim 1, characterized in that, The acquisition of temperature and humidity sensor data and smoke sensor data, and the subsequent determination of a potential fire risk score based on these data, specifically include: Using a 24-hour time window, obtain the mean smoke intensity, standard deviation of smoke intensity, and current smoke intensity of the monitored scene within the time window; The first risk score is obtained based on the mean smoke intensity, standard deviation of smoke intensity and current smoke intensity of the monitored scene within a time window; Based on temperature and humidity sensor data, and after data preprocessing to remove noise, a temperature and humidity heat map of the monitoring scene is obtained. Based on the temperature and humidity heat map of the monitoring scene, abnormal temperature and humidity data are obtained, wherein the abnormal temperature and humidity data are data that are greater than the corresponding preset threshold. A second risk score is obtained based on abnormal temperature and humidity data; Based on the first risk score and the second risk score, obtain the potential fire risk score; The first risk score is as follows: ； In the formula, As the first risk score, for function, Given the current smoke intensity, and These represent the mean and standard deviation of smoke intensity in the monitored scene within a time window, respectively. The second risk score is specifically as follows: ； ； ； In the formula, and These are the temperature coefficient and humidity coefficient, respectively. To monitor the highest temperature in the scene, To monitor the lowest temperature in the scene, To monitor the highest humidity in the scene, To monitor the lowest humidity in the scene, This is the second risk score; Specifically, the fire potential risk score is as follows: ； In the formula, Assess potential fire risks. As the first risk score, This is the second risk score.

3. A fire early warning method based on multi-source information fusion for smart IoT as described in claim 1, characterized in that, The step of obtaining the average brightness of the grayscale image corresponding to the monitored image, and adjusting the brightness of the monitored image based on the average brightness, specifically includes: Based on the grayscale image of the monitored image, obtain the average brightness of the corresponding grayscale image; Acquire historical monitoring images, analyze the historical monitoring images, and obtain reference brightness values; The ratio of the average brightness of the grayscale image corresponding to the monitored image to the reference brightness value is used as the compensation coefficient; Adjust the brightness of the monitored image based on the compensation coefficient; Specifically, the average brightness of the grayscale image corresponding to the monitored image is: ； In the formula, To monitor the average brightness of the corresponding grayscale image, To monitor the width of the image, To monitor the height of the image, To monitor points on the image grayscale value; Specifically, the compensation coefficient is: ； In the formula, For compensation coefficient, For reference brightness value; Specifically, adjusting the brightness of the monitored image involves: ； In the formula, To adjust the monitoring points on the image The grayscale value.

4. A fire early warning method based on multi-source information fusion for smart IoT as described in claim 1, characterized in that, The step of obtaining the edge fluctuation frequency corresponding to the edge of all candidate fire source regions based on the perimeter of the edge of each candidate fire source region, and filtering out candidate fire source regions that conform to the high-frequency jitter of the real flame edge based on the edge fluctuation frequency, specifically includes: Acquire monitoring images within 3 seconds and convert them into grayscale images. Regions in a grayscale image with grayscale values ​​greater than a preset threshold are defined as candidate fire source regions. Extract the bounding rectangles of all fire source candidate regions, and obtain the edge contours of all fire source candidate regions based on the centroid tracking algorithm and the Canny edge algorithm; Obtain the perimeter of the edge contours corresponding to all candidate fire source regions; Take the edge contour perimeter data corresponding to the same fire source candidate area as a group of data, and obtain the fire source perimeter data group corresponding to each fire source candidate area. Perform a fast Fourier transform on the perimeter data set corresponding to each candidate fire source region to obtain the edge fluctuation frequency corresponding to all candidate fire source regions. The fluctuation threshold is obtained based on the edge fluctuation frequency corresponding to all candidate fire source regions; The fire source candidate regions are adjusted based on a comparison of the edge fluctuation frequency and the corresponding fluctuation threshold of all fire source candidate regions. Specifically, obtaining the edge fluctuation frequency corresponding to all candidate fire source regions involves: ； ； In the formula, To monitor the image within 3 seconds Data on candidate fire source areas, For the first The first candidate area for fire source The first frame in the monitored image The perimeter of the edge contour corresponding to each candidate fire source region. To monitor the first in the image The average perimeter of the edge contour corresponding to each candidate fire source region For the first The edge fluctuation frequency of each candidate ignition source region For Fast Fourier Transform, For the first The first candidate area for fire source One data point; Specifically, the fluctuation threshold is: ； In the formula, For the first Fluctuation threshold of each fire source candidate region For the first The standard deviation of the edge fluctuation frequency of each candidate ignition source region For the first The average area of ​​each candidate fire source region.

5. A fire early warning method based on multi-source information fusion for smart IoT as described in claim 1, characterized in that, The process involves dividing the monitored image into regions of fixed size, obtaining local outlier indicators for each region, clustering outlier pixels based on these indicators to generate candidate outlier clusters, and then obtaining smoke candidate regions based on the spatial relationship between the candidate outlier clusters and the fire source candidate regions. Specifically, this includes: The monitoring image was divided into 8*8 regions, and the mean and standard deviation of the gray values ​​of all 8*8 regions were obtained. Based on the average gray level and standard deviation of each region, obtain the local outlier index corresponding to each region; Based on the connected component search algorithm and the local outlier index, if two outlier pixels have a common edge, then these two outlier pixels are regarded as a candidate outlier cluster. The outlier pixels are the regions where the local outlier index is not within the preset range. If an anomalous pixel is less than 8 pixels away from a candidate anomalous cluster, and the absolute difference between its local outlier index and a preset threshold is less than 0.1, then it is added to the corresponding candidate anomalous cluster. Adjust the candidate anomaly clusters based on the already obtained candidate fire source regions; The adjusted candidate anomaly clusters are used as smoke candidate regions; Specifically, the local outlier index is: ； In the formula, and Take 1-8, For the first The first 8*8 area Line number Local outlier index of the column, For the first The first 8*8 area Line number The grayscale value of the column, For the first The average grayscale value of an 8x8 region For the first The standard deviation of gray values ​​in an 8x8 area.

6. A fire early warning method based on multi-source information fusion for smart IoT as described in claim 5, characterized in that, The adjustment of candidate anomaly clusters based on the already acquired candidate fire source regions specifically includes: Based on the grayscale values ​​of the candidate fire source region, obtain the centroid of the fire source; The expansion radius is obtained based on the grayscale value of the candidate fire source region and the centroid of the fire source; Based on the algorithm and the expansion radius, morphological expansion processing is performed on the candidate fire source region to obtain an expanded candidate fire source region; If the Euclidean distance between the expanded candidate fire source region and the candidate anomaly cluster is less than a preset threshold, then the candidate anomaly cluster is retained. Obtain the area change curves of all candidate anomaly clusters within 3 seconds. If the curves continue to increase, retain the corresponding candidate anomaly clusters. Specifically, the centroid of the fire source is: ； ； ； In the formula, The x-coordinate of the centroid The ordinate of the centroid is y. This represents the number of pixels contained in the candidate fire source region. For the first The grayscale value of each pixel For the first The x-coordinate of each pixel For the first The ordinate of each pixel The center of mass of the fire source; Specifically, the expansion radius is: ； In the formula, Where is the radius of expansion. This represents the number of pixels contained in the candidate fire source region. The first candidate region for fire source The grayscale value of each pixel The first candidate region for fire source The x-coordinate of each pixel The first candidate region for fire source The ordinate of each pixel and These are the x-coordinate and y-coordinate of the centroid of the fire source, respectively.

7. A fire early warning system based on multi-source information fusion for smart IoT, used to implement the fire early warning method based on multi-source information fusion for smart IoT as described in any one of claims 1-6, characterized in that, include: The main control module is used to obtain a potential fire risk score, fire source candidate area and smoke candidate area based on temperature and humidity sensor data and smoke sensor data. The fire source candidate area and smoke candidate area are input into the target detection model to determine whether the monitoring image contains fire source or smoke. If the monitoring image contains fire source or smoke, an alarm is triggered through the smart IoT system. The risk scoring module is used to obtain a first risk score and a second risk score based on temperature and humidity sensor data and smoke sensor data, and to obtain a potential fire risk score based on the first risk score and the second risk score. The fire source candidate region acquisition module is used to define the region in the grayscale image with a grayscale value greater than a preset threshold as the fire source candidate region, and to adjust the fire source candidate region based on the comparison between the edge fluctuation frequency corresponding to all fire source candidate regions and the corresponding fluctuation threshold.

8. A fire early warning system for smart IoT based on multi-source information fusion, as described in claim 7, is characterized in that... The main control module specifically includes: The data acquisition unit is used to acquire temperature and humidity sensor data, smoke sensor data, acquire fire potential risk score, and acquire the average brightness of the grayscale image corresponding to the monitoring image based on the grayscale image of the monitoring image. The early warning unit is used to input the candidate areas of fire source and smoke into the target detection model to determine whether the monitoring image contains fire source or smoke. If the monitoring image contains fire source or smoke, an alarm is triggered through the smart IoT system.

9. A fire early warning system for smart IoT based on multi-source information fusion, as described in claim 7, is characterized in that... The risk scoring module specifically includes: The first risk score acquisition unit is used to acquire a first risk score based on the average smoke intensity, standard deviation of smoke intensity and current smoke intensity of the monitored scene within a time window. The second risk score acquisition unit is used to acquire abnormal temperature and humidity data based on the temperature and humidity heat map of the monitoring scene, and to acquire a second risk score based on the abnormal temperature and humidity data. A risk scoring acquisition unit is used to acquire a potential fire risk score based on a first risk score and a second risk score.

10. A fire early warning system for smart IoT based on multi-source information fusion, as described in claim 7, is characterized in that... The fire source candidate area acquisition module specifically includes: The perimeter acquisition unit is used to define the region in the grayscale image with a grayscale value greater than a preset threshold as a fire source candidate region, and to acquire the perimeter of the edge contours corresponding to all fire source candidate regions. A fire source candidate area adjustment unit is used to obtain the edge fluctuation frequency corresponding to all fire source candidate areas, and adjust the fire source candidate areas based on the comparison between the edge fluctuation frequency corresponding to all fire source candidate areas and the corresponding fluctuation threshold.

Images