A fire risk early warning method and system based on image

Abstract

The present invention relates to the field of image monitoring technology, and specifically discloses an image-based fire risk warning method and system, the method comprising obtaining object distribution information within a warning scene, inserting object contours containing risk values ​​into a scene map according to the object distribution information; determining the basic risk probability of each position based on the object contours containing risk values, and determining the monitoring frequency of each monitoring device according to the basic risk probability; receiving a scene image obtained by each monitoring device based on the monitoring frequency, identifying the scene image, and generating warning information according to the identification result; while identifying the scene image, synchronously locating the risk object, and updating the monitoring frequency of each monitoring device according to the risk object. The present invention adjusts the use frequency of the monitoring device according to the risk probability of each position, converts the original real-time monitoring process into an intermittent detection process, and increases the service life of related equipment and modules.

Description

Technical Field

[0001] The present invention relates to the technical field of image monitoring, and in particular to an image-based fire risk early warning method and system. Background Art

[0002] Image-based fire risk warning systems use image processing and computer vision technology to automatically identify fire features from video surveillance or image streams, thereby achieving early warning. Such systems usually combine methods such as color, shape, motion, and deep learning to improve detection accuracy and real-time performance.

[0003] In the existing image-based fire risk warning scheme, in order to ensure real-time performance, all image acquisition devices and recognition processes are carried out in real time. However, fires are rare, and continuous real-time warnings will reduce the use time of related equipment. Once a problem occurs with the equipment, the fire monitoring process will be ineffective, which will affect real-time performance. How to optimize the fire risk warning process and increase the service life of related equipment is the technical problem that the technical solution of the present invention aims to solve. Summary of the invention

[0004] The purpose of the present invention is to provide an image-based fire risk warning method and system to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A fire risk early warning method based on an image, the method comprising:

[0007] Obtain the object distribution information in the warning scene, and insert the object outline containing the risk value into the scene map according to the object distribution information;

[0008] Determine the basic risk probability of each position based on the object contour containing the risk value, and determine the monitoring frequency of each monitoring device according to the basic risk probability;

[0009] Receiving scene images acquired by each monitoring device based on the monitoring frequency, identifying the scene images, and generating warning information according to the identification results;

[0010] When the scene image is identified, the risk object is located synchronously, and the monitoring frequency of each monitoring device is updated according to the risk object.

[0011] As a further solution of the present invention: the step of obtaining the object distribution information in the warning scene and inserting the object contour containing the risk value into the scene map according to the object distribution information comprises:

[0012] Acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types;

[0013] For each region, query a preset feature table according to the region type, identify objects in the region based on the feature table, and obtain the object contour and object type;

[0014] Count all object outlines and object types, obtain object distribution information, and insert the object distribution information into the scene map;

[0015] determining a risk value of the object contour according to the object type of the object contour;

[0016] The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

[0017] As a further solution of the present invention: the step of determining the basic risk probability of each position based on the object contour containing the risk value, and determining the monitoring frequency of each monitoring device according to the basic risk probability includes:

[0018] Determine the basic risk probability of each location in the scene map based on the object outline containing the risk value;

[0019] Mark the monitoring area corresponding to the monitoring device in the scene map;

[0020] Query the basic risk probability of all points in the monitoring area, and determine the monitoring frequency of the monitoring equipment according to the basic risk probability;

[0021] The process of determining the basic risk probability is as follows:

[0022] ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours;

[0023] The process of determining the monitoring frequency is as follows:

[0024] ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

[0025] As a further solution of the present invention: the step of receiving the scene image acquired by each monitoring device based on the monitoring frequency, identifying the scene image, and generating warning information according to the identification result includes:

[0026] Receiving scene images acquired by each monitoring device based on the monitoring frequency;

[0027] Arrange the scene images in time order to obtain an image sequence;

[0028] Extracting feature values ​​of each image in the image sequence, and marking the scene image according to the feature values;

[0029] Identify the marked scene image and generate warning information based on the identification result;

[0030] The marking process is as follows:

[0031] ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

[0032] As a further solution of the present invention: the step of identifying the marked scene image and generating warning information according to the identification result includes:

[0033] Extract the HSV space color value of the scene image;

[0034] Locating the flame area in the scene image according to a preset flame threshold;

[0035] Based on the Canny edge detection algorithm, edge points are extracted in the flame area;

[0036] Identify the extracted edge points and determine the probability of fire;

[0037] When the probability of a fire is greater than a preset probability threshold, an early warning message is generated.

[0038] The flame threshold is: ;

[0039] The calculation process of fire probability is:

[0040] ; In the formula, is the fire probability, is the area of ​​the region composed of edge points, is the perimeter of the region composed of edge points, is the preset adjustment factor.

[0041] As a further solution of the present invention: when identifying the scene image, the step of synchronously locating the risk object and updating the monitoring frequency of each monitoring device according to the risk object includes:

[0042] When the probability of a fire is less than a preset probability threshold, query the monitoring device corresponding to the scene image;

[0043] updating the monitoring frequency of the monitoring device according to the probability;

[0044] The update process is as follows:

[0045] ; Indicates the updated monitoring frequency. Indicates the monitoring frequency before the update. Represents the probability of fire.

[0046] The technical solution of the present invention also provides an image-based fire risk early warning system, the system comprising:

[0047] The object information acquisition module is used to obtain the object distribution information in the warning scene, and insert the object outline containing the risk value into the scene map according to the object distribution information;

[0048] A monitoring frequency determination module, used to determine the basic risk probability of each position based on the object contour containing the risk value, and determine the monitoring frequency of each monitoring device according to the basic risk probability;

[0049] An image recognition module is used to receive scene images acquired by each monitoring device based on the monitoring frequency, recognize the scene images, and generate warning information according to the recognition results;

[0050] The monitoring frequency updating module is used to synchronously locate risk objects when identifying the scene image, and update the monitoring frequency of each monitoring device according to the risk objects.

[0051] As a further solution of the present invention: the object information acquisition module includes:

[0052] A segmentation information acquisition unit, used to acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types;

[0053] An object recognition and positioning unit is used to query a preset feature table for each area according to the area type, recognize objects in the area based on the feature table, and obtain the object contour and object type;

[0054] An information insertion unit is used to count all object contours and object types, obtain object distribution information, and insert the object distribution information into the scene map;

[0055] a risk value determination unit, configured to determine a risk value of the object contour according to the object type of the object contour;

[0056] The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

[0057] As a further solution of the present invention: the monitoring frequency determination module includes:

[0058] A basic determination unit, used to determine the basic risk probability of each position in the scene map based on the object contour containing the risk value;

[0059] An area marking unit, used to mark a monitoring area corresponding to a monitoring device in a scene map;

[0060] An accumulation unit, used to query the basic risk probability of all points in the monitoring area, and determine the monitoring frequency of the monitoring equipment according to the basic risk probability;

[0061] The process of determining the basic risk probability is as follows:

[0062] ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours;

[0063] The process of determining the monitoring frequency is as follows:

[0064] ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

[0065] As a further solution of the present invention: the image recognition module includes:

[0066] An image receiving unit, used to receive scene images acquired by each monitoring device based on the monitoring frequency;

[0067] An image arrangement unit, used for arranging scene images in time sequence to obtain an image sequence;

[0068] An image marking unit, used to extract a feature value of each image in the image sequence and mark the scene image according to the feature value;

[0069] The recognition execution unit is used to recognize the marked scene image and generate warning information according to the recognition result;

[0070] The marking process is as follows:

[0071] ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

[0072] Compared with the prior art, the beneficial effects of the present invention are: the present invention calculates the risk probability of each location in the scene, adjusts the use frequency of the monitoring equipment according to the risk probability, and converts the original real-time monitoring process into an intermittent detection process. On the basis of ensuring the real-time monitoring as much as possible, the service life of the monitoring equipment and the identification module is greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention.

[0074] Figure 1 This is a flowchart of the image-based fire risk warning method.

[0075] Figure 2 This is the first sub-process flowchart of the image-based fire risk warning method.

[0076] Figure 3 This is a flowchart of the second sub-process of the image-based fire risk warning method.

[0077] Figure 4 This is the third sub-process flowchart of the image-based fire risk warning method.

[0078] Figure 5 This is the fourth sub-process flowchart of the image-based fire risk warning method.

[0079] Figure 6 This is a structural block diagram of the image-based fire risk warning system. DETAILED DESCRIPTION

[0080] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0081] Figure 1 The flowchart of the image-based fire risk warning method is as follows. In an embodiment of the present invention, an image-based fire risk warning method is provided, and the method includes:

[0082] Step S100: Obtaining object distribution information in the warning scene, and inserting object contours containing risk values ​​into the scene map according to the object distribution information;

[0083] The warning scene is a scene where a fire risk warning is required, usually a workshop. The object distribution information in the warning scene is obtained. The object distribution information indicates which objects are placed in which positions. In order to facilitate the analysis and display processes, after the object distribution information is obtained, the object distribution information is inserted into the scene map. The insertion method is to mark the outlines corresponding to the objects in the scene map, which are called object outlines. Then, according to the possibility of the object causing a fire, an analysis is performed and a risk value is inserted into the object outline. The higher the possibility of causing a fire, the greater the risk value.

[0084] Step S200: determining the basic risk probability of each position based on the object contour containing the risk value, and determining the monitoring frequency of each monitoring device according to the basic risk probability;

[0085] The risk value indicates the possibility of causing a fire. After determining the risk value of each object outline in the scene map, the risk value is expanded to calculate the possibility of each location in the entire scene map being affected by the fire, which is called the basic risk probability. The monitoring frequency of each monitoring device is determined by the basic risk probability. The meaning of this process is that the greater the basic risk probability, the higher the possibility of causing a fire. At this time, the monitoring frequency of the monitoring device is higher, and more images need to be obtained within the same length of time to identify a fire. Among them, the monitoring device is a camera.

[0086] Step S300: receiving scene images acquired by each monitoring device based on the monitoring frequency, identifying the scene images, and generating warning information according to the identification results;

[0087] The monitoring equipment obtains the video in the scene based on the monitoring frequency. The video is a set of images with time tags, and each image is called a scene image. By identifying the scene image, it can be determined whether there is a fire, and then determine whether to generate warning information.

[0088] Step S400: when identifying the scene image, synchronously locate the risk object, and update the monitoring frequency of each monitoring device according to the risk object;

[0089] The focus of the technical solution of the present invention is the monitoring frequency of the monitoring equipment. Fire is almost impossible to occur under normal conditions. Therefore, the monitoring frequency does not need to be too high. If the monitoring frequency is too high, a large amount of computing resources will be consumed. The life of the monitoring equipment is limited. The more computing resources are consumed, the easier it is to be damaged. Once damaged, the monitoring frequency is equivalent to zero, which also reduces the speed of risk identification in disguise.

[0090] In fact, in the existing technical solutions, smoke alarms can also be used for rapid fire monitoring, but they can only quickly determine whether there is a fire and cannot reflect specific information. On this basis, the present application provides an image-based fire recognition method, which can generate early warning information when a fire is identified and upload on-site data to assist management personnel in selecting subsequent response methods.

[0091] It should be noted that the image-based fire detection process does not conflict with existing smoke alarms and can be used in conjunction with them.

[0092] Figure 2 The first sub-process flowchart of the image-based fire risk warning method is as follows: the step of obtaining object distribution information in the warning scene and inserting object contours containing risk values ​​into the scene map according to the object distribution information includes:

[0093] Step S101: Acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types;

[0094] Step S102: for each region, query a preset feature table according to the region type, identify objects in the region based on the feature table, and obtain the object contour and object type;

[0095] Step S103: Count all object outlines and object types to obtain object distribution information, and insert the object distribution information into the scene map;

[0096] Step S104: determining a risk value of the object contour according to the object type of the object contour;

[0097] The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

[0098] The early warning scene is generally a workshop, which is divided into multiple areas for completing different tasks, such as production or warehousing. The partition information is set independently by the management party and is regarded as known data in the technical solution of the present invention. The regional segmentation information in the early warning scene is obtained, and then each area is analyzed separately. The possible object features are determined according to the area type of the area, which is called a feature table. The object features are generally image features. In this way, by using a camera to capture the area image, the corresponding object can be identified in the area image to obtain the object contour and object type. All object contours and object types are called object distribution information. The object distribution information is inserted into the scene map. The insertion process is very simple. First, a map of the early warning scene is obtained, which is called a scene map. The contours corresponding to the objects are marked in the scene map. When marking, a risk value is determined in each contour according to the object type. The risk value is used to describe the possibility of causing a fire.

[0099] It should be noted that the relationship between the object type and the risk value is pre-set and can be stored in a table. For example, the risk value of wood is higher than that of iron. In the technical solution of the present invention, the risk value can be read in the table according to the object type. The table includes an object type item and a risk value item.

[0100] Figure 3 The second sub-flow chart of the image-based fire risk warning method, wherein the steps of determining the basic risk probability of each position based on the object contour containing the risk value, and determining the monitoring frequency of each monitoring device according to the basic risk probability include:

[0101] Step S201: determining the basic risk probability of each position in the scene map based on the object contour containing the risk value;

[0102] Step S202: marking the monitoring area corresponding to the monitoring device in the scene map;

[0103] Step S203: querying the basic risk probability of all points in the monitoring area, and determining the monitoring frequency of the monitoring equipment according to the basic risk probability;

[0104] The resulting scene map includes multiple object outlines, each of which has a risk value. By analyzing the risk values, the basic risk probability of each location in the scene map can be determined. This process is equivalent to determining the risk values ​​at all locations in the entire scene based on the distribution of objects, which is called the basic risk probability. For each monitoring device, its monitoring area is queried. The monitoring area is a range. The basic risk probability of all points within the range is obtained, and the risk situation of the entire range can be calculated, thereby determining the monitoring frequency of the monitoring device. The greater the risk probability, the higher the monitoring frequency.

[0105] The process of determining the basic risk probability is as follows:

[0106] ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours;

[0107] The calculation principle of the basic risk probability is to select the object contour in turn and calculate the predicted risk value of the object contour at a certain position. The calculation process is to calculate the sum of the risk values ​​of all points in the object contour and then divide it by the distance. Of course, an index can also be introduced on the distance to adjust the influence of the distance; the distance uses the distance between the center point of the contour and the position, accumulates the predicted risk values ​​of all object contours at the position, and then converts it into data in the range of 0 to 1 to obtain the basic risk probability; the conversion process is very simple, just calculate the difference between it and the minimum value, and then divide it by the difference between the maximum and minimum values.

[0108] The process of determining the monitoring frequency is as follows:

[0109] ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

[0110] The monitoring frequency is the working frequency of the monitoring equipment. After the basic risk probability has been obtained, the monitoring area of ​​the monitoring equipment is obtained, the basic risk probabilities of all points in the monitoring area are accumulated, and the final monitoring frequency is obtained after correction and rounding up.

[0111] Figure 4 The third sub-process flowchart of the image-based fire risk warning method is as follows: the steps of receiving the scene image acquired by each monitoring device based on the monitoring frequency, identifying the scene image, and generating warning information according to the identification result include:

[0112] Step S301: receiving a scene image acquired by each monitoring device based on a monitoring frequency;

[0113] Step S302: Arrange the scene images in chronological order to obtain an image sequence;

[0114] Step S303: extracting feature values ​​of each image in the image sequence, and marking the scene image according to the feature values;

[0115] Step S304: Identify the marked scene image and generate warning information according to the identification result;

[0116] In a preferred embodiment of the technical solution of the present invention, a specific description is given of the scene image recognition process, and scene images acquired by each monitoring device based on the monitoring frequency are received. When acquiring the scene images, each monitoring device records the acquisition time, and the scene images are arranged in chronological order to obtain an image sequence; the characteristic values ​​of each image in the image sequence are extracted, and the characteristic values ​​are analyzed to determine whether there is a mutation phenomenon in the monitoring area. If there is a mutation, the corresponding scene image is marked; then, the marked scene image is further identified to generate early warning information.

[0117] The marking process is as follows:

[0118] ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

[0119] The marking principle is very simple, that is, when the change in eigenvalue is large enough, it is marked; the change in eigenvalue can be calculated by subtracting the previous eigenvalue from the current eigenvalue, or by subtracting the current eigenvalue from the next eigenvalue. After the calculation is completed, the absolute value is taken to make it a positive number.

[0120] Specifically, the step of identifying the marked scene image and generating warning information according to the identification result includes:

[0121] Extract the HSV space color value of the scene image;

[0122] Locating the flame area in the scene image according to a preset flame threshold;

[0123] Based on the Canny edge detection algorithm, edge points are extracted in the flame area;

[0124] Identify the extracted edge points and determine the probability of fire;

[0125] When the probability of a fire is greater than a preset probability threshold, an early warning message is generated.

[0126] The above content provides a specific solution for identifying fire through images. The colors of flames and smoke are relatively typical (such as flames are yellow, orange or red, and smoke are gray or black). Appropriate HSV color thresholds or RGB color thresholds can be set to filter out areas that may contain flames or smoke. In the technical solution of the present invention, the HSV color threshold is used. For the flame area, the HSV color threshold is: ;

[0127] For the smoke area, you can choose a gray range with low saturation and lower brightness.

[0128] In the technical solution of the present invention, the flame area is mainly located. After the flame area is located, edge points are extracted in the flame area based on the Canny edge detection algorithm. The edge of the flame or smoke is extracted using Canny edge detection. The edge of the flame is often irregular and the edge of the smoke is fuzzy. Therefore, the edge points are identified and analyzed to calculate the probability of fire. When the probability of fire is large enough, warning information is generated.

[0129] The calculation process of fire probability is:

[0130] ; In the formula, is the fire probability, is the area of ​​the region composed of edge points, is the perimeter of the region composed of edge points, is the preset adjustment factor.

[0131] The above content, It represents the roundness. For regular edges, the roundness value is close to 1 (such as a circle), while the roundness value of irregular edges deviates from 1.

[0132] Figure 5 The fourth sub-flow chart of the image-based fire risk warning method, wherein when the scene image is identified, the risk object is synchronously located, and the step of updating the monitoring frequency of each monitoring device according to the risk object includes:

[0133] Step S401: when the probability of fire is less than a preset probability threshold, query the monitoring device corresponding to the scene image;

[0134] Step S402: updating the monitoring frequency of the monitoring device according to the probability;

[0135] Since the fire probability calculation process is for the marked scene image, the marked scene image is the image when a mutation occurs. Therefore, when the probability of fire is high but not too high, it is necessary to increase the monitoring frequency of the monitoring equipment to allow it to monitor more frequently to ensure the timeliness of monitoring.

[0136] The update process is as follows:

[0137] ; Indicates the updated monitoring frequency. Indicates the monitoring frequency before the update. Represents the probability of fire.

[0138] The updating process of the monitoring frequency is very simple. The monitoring frequency can be amplified in proportion to the probability.

[0139] Figure 6 FIG. 1 is a structural block diagram of a fire risk warning system based on an image. In an embodiment of the present invention, an image-based fire risk warning system 10 includes:

[0140] The object information acquisition module 11 is used to obtain the object distribution information in the warning scene, and insert the object outline containing the risk value into the scene map according to the object distribution information;

[0141] A monitoring frequency determination module 12, configured to determine a basic risk probability of each location based on an object profile containing a risk value, and determine a monitoring frequency of each monitoring device according to the basic risk probability;

[0142] An image recognition module 13 is used to receive scene images acquired by each monitoring device based on the monitoring frequency, recognize the scene images, and generate warning information according to the recognition results;

[0143] The monitoring frequency updating module 14 is used to simultaneously locate risk objects when identifying the scene image, and update the monitoring frequency of each monitoring device according to the risk objects.

[0144] Furthermore, the object information acquisition module 11 includes:

[0145] A segmentation information acquisition unit, used to acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types;

[0146] An object recognition and positioning unit is used to query a preset feature table for each area according to the area type, recognize objects in the area based on the feature table, and obtain the object contour and object type;

[0147] An information insertion unit is used to count all object contours and object types, obtain object distribution information, and insert the object distribution information into the scene map;

[0148] a risk value determination unit, configured to determine a risk value of the object contour according to the object type of the object contour;

[0149] The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

[0150] Specifically, the monitoring frequency determination module 12 includes:

[0151] A basic determination unit, used to determine the basic risk probability of each position in the scene map based on the object contour containing the risk value;

[0152] An area marking unit, used to mark a monitoring area corresponding to a monitoring device in a scene map;

[0153] An accumulation unit, used to query the basic risk probability of all points in the monitoring area, and determine the monitoring frequency of the monitoring equipment according to the basic risk probability;

[0154] The process of determining the basic risk probability is as follows:

[0155] ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours;

[0156] The process of determining the monitoring frequency is as follows:

[0157] ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

[0158] Furthermore, the image recognition module 13 includes:

[0159] An image receiving unit, used to receive scene images acquired by each monitoring device based on the monitoring frequency;

[0160] An image arrangement unit, used for arranging scene images in time sequence to obtain an image sequence;

[0161] An image marking unit, used to extract a feature value of each image in the image sequence and mark the scene image according to the feature value;

[0162] The recognition execution unit is used to recognize the marked scene image and generate warning information according to the recognition result;

[0163] The marking process is as follows:

[0164] ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A fire risk early warning method based on an image, characterized in that: The method comprises: Obtain the object distribution information in the warning scene, and insert the object outline containing the risk value into the scene map according to the object distribution information; Determine the basic risk probability of each position based on the object contour containing the risk value, and determine the monitoring frequency of each monitoring device according to the basic risk probability; Receiving scene images acquired by each monitoring device based on the monitoring frequency, identifying the scene images, and generating warning information according to the identification results; When the scene image is identified, the risk object is synchronously located, and the monitoring frequency of each monitoring device is updated according to the risk object; The step of determining the basic risk probability of each position based on the object contour containing the risk value, and determining the monitoring frequency of each monitoring device according to the basic risk probability comprises: Determine the basic risk probability of each location in the scene map based on the object outline containing the risk value; Mark the monitoring area corresponding to the monitoring device in the scene map; Query the basic risk probability of all points in the monitoring area, and determine the monitoring frequency of the monitoring equipment according to the basic risk probability; The process of determining the basic risk probability is as follows: ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours; The process of determining the monitoring frequency is as follows: ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

2. The image-based fire risk early warning method according to claim 1, characterized in that: The step of obtaining object distribution information in the warning scene and inserting object contours containing risk values ​​into the scene map according to the object distribution information comprises: Acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types; For each region, query a preset feature table according to the region type, identify objects in the region based on the feature table, and obtain the object contour and object type; Count all object outlines and object types, obtain object distribution information, and insert the object distribution information into the scene map; determining a risk value of the object contour according to the object type of the object contour; The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

3. The image-based fire risk early warning method according to claim 1, characterized in that: The step of receiving the scene image acquired by each monitoring device based on the monitoring frequency, identifying the scene image, and generating warning information according to the identification result includes: Receiving scene images acquired by each monitoring device based on the monitoring frequency; Arrange the scene images in time order to obtain an image sequence; Extracting feature values ​​of each image in the image sequence, and marking the scene image according to the feature values; Identify the marked scene image and generate warning information based on the identification result; The marking process is as follows: ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

4. The image-based fire risk early warning method according to claim 3 is characterized in that: The step of identifying the marked scene image and generating warning information according to the identification result includes: Extract the HSV space color value of the scene image; Locating the flame area in the scene image according to a preset flame threshold; Based on the Canny edge detection algorithm, edge points are extracted in the flame area; Identify the extracted edge points and determine the probability of fire; When the probability of fire is greater than the preset probability threshold, an early warning message is generated; The flame threshold is: ; The calculation process of fire probability is: ; In the formula, is the fire probability, is the area of ​​the region composed of edge points, is the perimeter of the region composed of edge points, is the preset adjustment factor.

5. The image-based fire risk early warning method according to claim 4, characterized in that: The step of synchronously locating risk objects when identifying the scene image and updating the monitoring frequency of each monitoring device according to the risk objects comprises: When the probability of a fire is less than a preset probability threshold, query the monitoring device corresponding to the scene image; updating the monitoring frequency of the monitoring device according to the probability; The update process is as follows: ; Indicates the updated monitoring frequency. Indicates the monitoring frequency before the update. Represents the probability of fire.

6. An image-based fire risk warning system, characterized in that: The system comprises: The object information acquisition module is used to obtain the object distribution information in the warning scene, and insert the object outline containing the risk value into the scene map according to the object distribution information; A monitoring frequency determination module, used to determine the basic risk probability of each position based on the object contour containing the risk value, and determine the monitoring frequency of each monitoring device according to the basic risk probability; An image recognition module is used to receive scene images acquired by each monitoring device based on the monitoring frequency, recognize the scene images, and generate warning information according to the recognition results; A monitoring frequency updating module, used to synchronously locate risk objects when identifying the scene image, and update the monitoring frequency of each monitoring device according to the risk objects; The monitoring frequency determination module comprises: A basic determination unit, used to determine the basic risk probability of each position in the scene map based on the object contour containing the risk value; An area marking unit, used to mark a monitoring area corresponding to a monitoring device in a scene map; An accumulation unit, used to query the basic risk probability of all points in the monitoring area, and determine the monitoring frequency of the monitoring equipment according to the basic risk probability; The process of determining the basic risk probability is as follows: ; In the formula, Location in the scene map The basic risk probability at Location in the scene map The predicted risk value at is the minimum value of the predicted risk value of all locations, is the maximum value of the predicted risk value of all locations; Indicates The risk value of the object contour, For the The area of ​​the object's outline, For the The coordinates of the center point of the object contour; express and distance; Represents the total number of object contours; The process of determining the monitoring frequency is as follows: ; In the formula, Indicates the monitoring frequency, is the preset adjustment coefficient, Indicates the point in the monitoring area The basic risk probability at the location; and Respectively represent the length and width of the monitoring area.

7. The image-based fire risk warning system according to claim 6, characterized in that: The object information acquisition module includes: A segmentation information acquisition unit, used to acquire regional segmentation information within the warning scene; the regional segmentation information includes regional boundaries and regional types; An object recognition and positioning unit is used to query a preset feature table for each area according to the area type, recognize objects in the area based on the feature table, and obtain the object contour and object type; An information insertion unit is used to count all object contours and object types, obtain object distribution information, and insert the object distribution information into the scene map; a risk value determination unit, configured to determine a risk value of the object contour according to the object type of the object contour; The area boundary and object outline are both coordinate sets, and the coordinates use relative coordinates based on the warning scene.

8. The image-based fire risk warning system according to claim 6, characterized in that: The image recognition module comprises: An image receiving unit, used to receive scene images acquired by each monitoring device based on the monitoring frequency; An image arrangement unit, used for arranging scene images in time sequence to obtain an image sequence; An image marking unit, used to extract a feature value of each image in the image sequence and mark the scene image according to the feature value; The recognition execution unit is used to recognize the marked scene image and generate warning information according to the recognition result; The marking process is as follows: ; In the formula, is a marking parameter, which means marking when it is one and not marking when it is zero; is the change in the eigenvalue of the scene image, and the calculation process is to find the difference between two adjacent eigenvalues; is the preset feature value threshold.

Images