A fire inspection drone control system

Abstract

This invention relates to the field of image feature analysis technology, specifically to a fire inspection drone control system. The system determines the fire matching direction through feature point matching, and then determines the effective adjustment amount of the image detection frequency based on the difference between the fire matching direction and the drone's flight direction. During feature point matching, it analyzes the associated region information of the fire area in historical images, determines the distribution of single important information, quantifies the degree of aggregation of original important information, and thus determines the accurate matching region and obtains the matching result. This invention analyzes the distribution of important information of the fire area in historical images, quantifies the characteristics of fire spread, and obtains effective matching points. After obtaining the fire matching direction, the effective adjustment amount of the image detection frequency can be determined by the difference between it and the drone's flight direction, avoiding excessive energy consumption due to over-adjustment that affects flight endurance.

Description

Technical Field

[0001] This invention relates to the field of image feature analysis technology, and specifically to a fire inspection drone control system. Background Technology

[0002] Using drones for fire monitoring can quickly and effectively pinpoint fire areas within a large inspection area. During the inspection, the drone's onboard infrared imaging detection device collects infrared images of the inspected area, and the temperature information revealed by these images can then be used to determine the fire zone. Drone fire inspections are commonly used in large, fire-prone areas such as forests and grasslands.

[0003] Because the inspection area is large and the drone spends a considerable amount of time operating over it, the image detection frequency is typically set low to minimize energy consumption and ensure long-term endurance. This allows the image acquisition equipment to operate with low power consumption. However, in fire scenarios, where fire spreads rapidly, a high detection frequency is required for image acquisition in fire-affected areas to ensure timely and complete fire information transmission. Current technology cannot adaptively control the detection frequency of drones during fire image acquisition, making it impossible to simultaneously balance endurance and fire information transmission efficiency. Summary of the Invention

[0004] To address the technical problem that existing technologies cannot adaptively control the detection frequency of drones during fire image acquisition, thus failing to simultaneously balance endurance and fire information transmission efficiency, the present invention aims to provide a fire inspection drone control system. The specific technical solution adopted is as follows:

[0005] This invention proposes a fire inspection drone control system, the system comprising:

[0006] The fire inspection image acquisition module is used to acquire fire images taken by drones in the inspection area; the fire images contain multiple fire areas;

[0007] The historical area comparison module is used to obtain the corresponding associated areas of the inspection area and the fire area in multiple historical images of the inspection area when no fire occurred; on each historical image, important information gray values ​​are filtered out and the corresponding information pixels are determined according to the frequency of occurrence of gray values ​​contained in the associated areas; the original important information aggregation degree of the fire area is obtained according to the density of information pixels in the associated areas and the degree of information uniformity in all historical images.

[0008] The fire area matching module is used to match two adjacent fire images. During the matching process, for a feature point in one fire image, the matching area is obtained according to the size of the fire area and the degree of aggregation of the original important information, and the matching point is determined in the matching area of ​​the other fire image.

[0009] The detection frequency adjustment module is used to connect mutually matched feature points to obtain the fire matching direction, and adjust the image detection frequency of the UAV according to the difference between the fire matching direction and the flight direction of the UAV.

[0010] Furthermore, the method for filtering the grayscale values ​​of the important information includes:

[0011] For a historical image, any grayscale value is selected as the target grayscale value. The percentage of pixels with the target grayscale value in each associated region is obtained, and the area weight is obtained based on the area of ​​the associated region. The area weight is multiplied by the percentage of pixels to obtain the information content of the target grayscale value in each associated region. The information content of all associated regions in the historical image is averaged to obtain the information importance of the target grayscale value in the historical image. Grayscale values ​​with information importance greater than a preset importance threshold are selected as important information grayscale values ​​in the historical image.

[0012] Furthermore, the method for obtaining the degree of information uniformity includes: performing a negative correlation mapping and normalizing the information entropy in the associated regions to obtain the degree of information uniformity.

[0013] Furthermore, the method for obtaining the degree of aggregation of the original important information includes:

[0014] For all associated regions corresponding to a fire area, the product of the information pixel density and the degree of information uniformity in each associated region is used as the initial information aggregation degree of each associated region. The initial information aggregation degrees of all associated regions corresponding to the fire area are averaged to obtain the original important information aggregation degree of the fire area.

[0015] Furthermore, the method for filtering the feature points includes:

[0016] The saliency of each pixel is obtained using a saliency algorithm, and pixels with saliency greater than a preset saliency threshold are selected as feature points.

[0017] Furthermore, the method for obtaining the matching region includes:

[0018] Matching the previous frame of the fire image with the next frame of the fire image: For a feature point in the previous frame of the fire image, obtain the first minimum bounding rectangle of the fire area to which the feature point belongs, and obtain the second minimum bounding rectangle of the fire area to which the corresponding pixel of the feature point belongs in the next frame of the fire image; obtain the intersection-union ratio (IUR) of the first and second minimum bounding rectangles; use the product of the IUR and the long side of the first minimum bounding rectangle as the matching reference side length; increase the matching reference side length by utilizing the original important information aggregation degree of the fire area where the feature point is located, and obtain the matching area size; construct the matching area in the next frame of the fire image with the corresponding pixel of the feature point as the center based on the matching area size.

[0019] Furthermore, the method for obtaining the matching point includes:

[0020] The pixels in the matching region of the next frame fire image are taken as the points to be matched of the feature point; the first pixel value change feature of the feature point in each direction in the neighborhood is obtained, and the second pixel value change feature of the point to be matched in the same direction in the neighborhood is obtained. The matching degree between the feature point and the point to be matched is obtained based on the similarity between the first pixel value change feature and the second pixel value change feature in the same direction. The point to be matched with the highest matching degree is selected as the matching point of the feature point.

[0021] Furthermore, the method for obtaining the matching degree includes:

[0022] The neighborhood is an eight-neighborhood. The pixel value difference between the feature point and the neighboring pixels in each direction within the eight-neighborhood is used as the first pixel value change feature, and the pixel value difference between the point to be matched and the neighboring pixels in each direction within the eight-neighborhood is used as the second pixel value change feature. For each direction, the difference between the first pixel value change feature and the second pixel value change feature is negatively correlated and normalized to obtain the pixel value change similarity in each direction. The average pixel value change similarity in all directions is used as the matching degree.

[0023] Furthermore, the method for obtaining the fire intensity matching direction includes:

[0024] For the time when the drone image detection frequency needs to be adjusted, the preset time period before the time to be adjusted is taken as the time period to be analyzed. The pixels in the two adjacent fire images that have matching relationships in the time period to be analyzed are connected to obtain multiple initial directions. The direction with the highest frequency of initial directions is selected as the fire matching direction at the time to be adjusted.

[0025] Furthermore, adjusting the image detection frequency of the UAV based on the difference between the fire matching direction and the UAV's flight direction includes:

[0026] The angle between the fire matching direction and the UAV flight direction is normalized to obtain the adjustment weight. The sum of the positive integer 1 and the adjustment weight is used as the increase ratio. The product of the increase ratio and the image detection frequency is used as the increased image detection frequency.

[0027] The present invention has the following beneficial effects:

[0028] To effectively improve image detection frequency without blindly increasing it and causing drone battery life issues, this invention uses feature point matching to determine the fire matching direction. The difference between the fire matching direction and the drone's flight direction is then used to determine the effective adjustment amount for the image detection frequency, avoiding excessive energy consumption due to over-adjustment that impacts battery life. In the feature point matching process, considering that inspection areas are often covered by single-information elements such as vegetation, where fire spreads significantly, accurate matching areas are needed to ensure matching accuracy. This invention analyzes the associated region information of fire areas in historical images to determine the distribution of single important information, quantifying the degree of original important information aggregation. A higher aggregation degree indicates a greater fire spread, requiring a larger matching area to accurately determine the matching point. This invention analyzes the distribution of important information in fire areas in historical images, quantifies the characteristics of fire spread, and obtains effective matching points. After obtaining the fire matching direction, the difference between this direction and the drone's flight direction can be used to determine the effective adjustment amount for the image detection frequency, avoiding excessive energy consumption due to over-adjustment that impacts battery life. Attached Figure Description

[0029] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a block diagram of a fire inspection drone control system provided in one embodiment of the present invention. Detailed Implementation

[0031] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a fire inspection drone control system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] The specific solution of a fire inspection drone control system provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0034] Please see Figure 1 The diagram illustrates a fire inspection drone control system according to an embodiment of the present invention. The system includes: a fire inspection image acquisition module 101, a historical area comparison module 102, a fire area matching module 103, and a detection frequency adjustment module 104.

[0035] The fire inspection image acquisition module 101 acquires fire images taken by the drone in the inspection area; the fire images contain multiple fire areas. In this embodiment of the invention, the initial detection frequency of the drone is set to once every 3 seconds, and a detection frequency adjustment cycle of 30 seconds is set during the inspection process. The detection frequency is adjusted when the drone image detection frequency needs to be adjusted.

[0036] In this embodiment of the invention, the fire area can be directly obtained using existing region analysis methods. For example, the infrared image corresponding to the fire image can be used, where the high-temperature region is the fire area. Similarly, the fire area in the fire image can be directly segmented using image segmentation or semantic segmentation networks. These are specific techniques well-known to those skilled in the art and will not be elaborated upon here.

[0037] In order to analyze the direction of fire movement within an inspection area, this invention employs a feature point matching method. By matching multiple consecutive frames of fire images, the direction of feature point movement, i.e., the direction of fire movement, can be determined. To ensure matching accuracy, this invention considers that fires can spread more rapidly in areas of the inspection area covered by vegetation or other single-information sources. Therefore, the matching area needs to be precisely located during the matching process, and effective matching needs to be performed within that area.

[0038] The historical region comparison module 102 is used to analyze the associated regions corresponding to each fire area in multiple historical images of the inspection area when the fire occurred. That is, the associated regions are areas of the same location, size, and shape as the fire areas, differing only in the time of acquisition. One fire area in a single fire image frame corresponds to multiple associated regions, each representing an image obtained from a historical image at a specific historical moment. Since single important information such as vegetation is key information in the inspection area and its distribution is relatively wide and abundant, important information gray values ​​can be directly filtered out from the historical images based on the frequency of gray values ​​appearing in the associated regions, thereby determining the information pixels corresponding to the important information gray values ​​in each associated region.

[0039] In this embodiment of the invention, the associated region can be extracted by constructing a mask from the fire area. This is a technical method well known to those skilled in the art and will not be described in detail here.

[0040] For a given associated region, the more information pixels it contains and the more uniform its overall information, the more important the information is considered to be concentrated within that region. Therefore, we statistically analyze all associated regions corresponding to each fire area and determine the original degree of concentration of important information within the fire area based on the pixel density and information uniformity of these regions across all historical images. In other words, the greater the original degree of concentration of important information, the more concentrated the fire area is in its original state, indicating a concentration of single, important information such as vegetation. Consequently, the larger the matching region required during the matching process, the more important the fire area will be.

[0041] The matching process is further executed in the fire area matching module 103, matching the acquired fire images of two adjacent frames. In this embodiment of the invention, the previous fire image is used as a reference to obtain matching points for feature points in the subsequent fire image. During the matching process, for a feature point in the previous fire image, it is first necessary to determine the matching area of ​​that feature point in the subsequent fire image. Specifically, the matching area can be obtained by combining the size of the fire area where the feature point is located and the degree of aggregation of original important information. That is, the larger the fire area where the feature point is located, the greater the spread of the fire, and the more likely a matching point needs to be found in the subsequent fire image within a larger matching area; the greater the degree of aggregation of original important information in the fire area where the feature point is located, the greater the spread of the fire in that area, and the more likely a matching point needs to be found in the subsequent fire image within a larger matching area. After determining the matching area, matching can be performed to obtain the matching points for the feature points.

[0042] Once the matching relationship is established, the detection frequency adjustment module 104 connects the matching feature points, and the fire matching direction is obtained based on the direction of the connecting lines. Then, the image detection frequency of the drone is adjusted according to the difference between the fire matching direction and the drone's flight direction. That is, the greater the difference between the two directions, the more likely the drone is moving away from the fire's spread, and a higher detection frequency is needed to acquire images to avoid missing fire information; conversely, a smaller difference indicates that the drone is flying along the fire's direction, and the original detection frequency or a slightly enhanced detection frequency is sufficient to acquire effective fire information.

[0043] In summary, this invention determines the fire matching direction through feature point matching, and then determines the effective adjustment amount of the image detection frequency by the difference between the fire matching direction and the drone's flight direction. During feature point matching, the associated region information of the fire area in historical images is analyzed to determine the distribution of single important information, quantify the degree of aggregation of original important information, and thus determine the accurate matching region and obtain the matching result. This invention analyzes the distribution of important information of the fire area in historical images, quantifies the characteristics of fire spread, and obtains effective matching points. After obtaining the fire matching direction, the effective adjustment amount of the image detection frequency can be determined by the difference between it and the drone's flight direction, avoiding excessive energy consumption due to over-adjustment that affects flight endurance.

[0044] Preferably, in this embodiment of the invention, the method for filtering the grayscale values ​​of important information includes:

[0045] The important information described in this embodiment of the invention refers to the important information formed by single pixel values ​​such as vegetation and shrubs. Because the fire spreads to a greater extent in areas such as vegetation and shrubs during the fire's spread, it is necessary to determine the grayscale values ​​of important information in historical images of the inspection area, and then filter out information pixels to evaluate the properties of various related areas in the historical images.

[0046] For a historical image, any grayscale value is selected as the target grayscale value. The percentage of pixels corresponding to the target grayscale value in each associated region is obtained, and the area weight is obtained based on the area of ​​the associated region. Since the associated region belongs to the historical region corresponding to the fire area before the fire occurred, for a target grayscale value, the more pixels it corresponds to in an associated region, the greater the probability that the target grayscale value is an important information grayscale value. Furthermore, an area weight is introduced; that is, the larger the area weight, the larger the fire area at the time of the fire, and the greater the information confidence of the associated region. Therefore, the area weight is multiplied by the percentage of pixels to obtain the information content of the target grayscale value in each associated region.

[0047] In this embodiment of the invention, the area weight is the result of normalizing the area of ​​the associated region. This embodiment of the invention uses maximum maximization for normalization, that is, the area of ​​the associated region is used as the numerator, and the area of ​​the largest associated region is used as the denominator to obtain the area weight.

[0048] The information intensity of all associated regions in the historical image is averaged to obtain the information importance of the target gray value in the historical image; gray values ​​with information importance greater than a preset importance threshold are selected as important information gray values ​​in the historical image. In this embodiment of the invention, after information importance normalization processing, the threshold is set to 0.68, and the normalization method can be range standardization. It should be noted that after analysis, each historical image corresponds to a set of important information gray values, and the important information gray values ​​in different historical images may be the same or may only have overlap.

[0049] Preferably, in this embodiment of the invention, considering that information entropy can characterize the degree of disorder in the distribution of information in a related region, the information entropy can be negatively correlated and normalized to obtain the degree of information uniformity. The negative correlation mapping and normalization method used in this embodiment of the invention is the function mapping method, which takes the negative of the information entropy as the power of an exponential function with the natural constant as the base, and the mapping result of the exponential function is the result after negative correlation mapping and normalization.

[0050] Preferably, in this embodiment of the invention, the method for obtaining the degree of aggregation of original important information includes:

[0051] For all associated regions corresponding to a fire area, the product of the information pixel density and the degree of information uniformity in each associated region is used as the initial information clustering degree of each associated region. The initial information clustering degrees of all associated regions corresponding to the fire area are averaged to obtain the original important information clustering degree of the fire area. It should be noted that the average here is applied to the associated regions corresponding to the fire area in different historical images, not all associated regions in a single historical image. By averaging the information of associated regions in all historical images, the original important information clustering degree is obtained. The larger the original important information clustering degree, the more likely the fire area was covered by vegetation, bushes, etc. before the fire occurred, indicating a strong fire spread capability and requiring a larger matching area in the subsequent matching process.

[0052] It should be noted that the information pixel density is the ratio of the number of information pixels in the associated region to the total number of pixels in the region.

[0053] Preferably, in this embodiment of the invention, the feature point selection method includes:

[0054] The saliency of each pixel is obtained using a saliency algorithm, and pixels with a saliency greater than a preset saliency threshold are selected as feature points. In this embodiment, the CA saliency algorithm can be used; other existing saliency algorithms can also be used in other embodiments, which will not be elaborated further. In this embodiment, after saliency normalization, the saliency threshold is set to 0.7.

[0055] Preferably, in this embodiment of the invention, the method for obtaining the matching region includes:

[0056] The preceding and following fire images are matched. For a feature point in the preceding fire image, the first minimum bounding rectangle of the fire region to which the feature point belongs is obtained, and the second minimum bounding rectangle of the fire region to which the corresponding pixel in the following fire image belongs is obtained. The intersection-union ratio (IUR) of the first and second minimum bounding rectangles is then calculated. The IUR represents the similarity in shape, position, and area between two regions. A larger IUR indicates a stronger correspondence between the two regions in the preceding and following fire images. Therefore, the IUR can be used as a weight, and the product of the IUR and the longer side of the first minimum bounding rectangle is used as the matching reference side length.

[0057] The matching region size is obtained by increasing the matching baseline side length by utilizing the original important information aggregation degree of the fire area where the feature point is located. In this embodiment of the invention, the original important information aggregation degree is a value between 0 and 1. Therefore, the sum of the positive integer 1 and the original important information aggregation degree can be directly used as the amplification factor. Multiplying the amplification factor by the matching baseline side length yields the matching region size.

[0058] Based on the size of the matching region, a matching region is constructed on the subsequent frame of the fire image, centered on the pixel corresponding to the feature point. In this embodiment of the invention, the matching region is a square with a side length equal to the size of the matching region. It should be noted that, to ensure successful construction of the matching region, the obtained matching region size needs to be rounded up before construction.

[0059] Furthermore, in this embodiment of the invention, after determining the matching region, the method for obtaining the matching point includes:

[0060] The pixels in the matching region of the subsequent fire image frame are used as the target points for the feature point. The first pixel value change feature of the feature point in each direction within its neighborhood is obtained, and the second pixel value change feature of the target point in the same direction within its neighborhood is obtained. The degree of matching between the feature point and the target point is determined based on the similarity between the first and second pixel value change features in the same direction. That is, the more similar the neighborhood pixel value change features of the feature point and the target point, the more matched the two points are, and they may be of the same type, such as protruding points on the boundary of a fire area. Then, the target point with the highest degree of matching is selected as the matching point for the feature point within the matching region.

[0061] Furthermore, methods for obtaining the degree of matching include:

[0062] The neighborhood analyzed for the aforementioned feature points and the points to be matched is an eight-neighborhood, meaning that there is only one neighboring pixel in each direction within the neighborhood. Therefore, the pixel value difference between the feature point and the neighboring pixels in each direction within the eight-neighborhood is used as the first pixel value change feature, and the pixel value difference between the point to be matched and the neighboring pixels in each direction within the eight-neighborhood is used as the second pixel value change feature. In this embodiment of the invention, the pixel value difference is set as the difference between the neighborhood center point and the neighboring pixels, meaning the pixel value difference has a positive or negative sign to indicate the direction of the change feature.

[0063] For each direction, the difference between the first pixel value change feature and the second pixel value change feature is negatively correlated and normalized to obtain the pixel value change similarity in each direction; the average pixel value change similarity in all directions is used as the matching degree.

[0064] In this embodiment of the invention, the difference between the first pixel value change feature and the second pixel value change feature is the absolute value of the difference between the two features. The negative correlation mapping and normalization method also adopts the above-mentioned exponential function mapping method, which will not be described in detail here.

[0065] Preferably, in this embodiment of the invention, the method for obtaining the fire matching direction includes:

[0066] For the moment when the drone image detection frequency needs to be adjusted, a preset time period before the moment to be adjusted is taken as the analysis time period. Pixels with matching relationships in two adjacent fire images participating in the analysis time period are connected to obtain multiple initial directions. The direction with the highest frequency of initial directions is selected as the fire matching direction at the moment to be adjusted. In this embodiment of the invention, the preset time period is the entire monitoring frequency adjustment cycle, that is, the analysis time period can be set to 30 seconds.

[0067] In other embodiments of the present invention, after connecting the pixels that have a matching relationship, multiple direction vectors are obtained, and the direction of the sum of all direction vectors can be used as the fire matching direction.

[0068] Preferably, in this embodiment of the invention, adjusting the image detection frequency of the UAV based on the difference between the fire matching direction and the UAV's flight direction includes:

[0069] The angle between the fire matching direction and the drone's flight direction is normalized to obtain an adjustment weight. The sum of the positive integer 1 and the adjustment weight is used as the amplification ratio, and the product of the amplification ratio and the image detection frequency is used as the amplified image detection frequency. In this embodiment of the invention, the included angle is the smallest of the two included angles between the two directions, i.e., the maximum included angle is 180 degrees; the normalization operation is maximization, i.e., the included angle value is the numerator, and 180 is the denominator, to obtain the adjustment weight.

[0070] It should be noted that in actual implementation, the obtained image detection frequency also needs to be rounded for easy adjustment. In this embodiment of the invention, the product of the increase ratio and the image detection frequency is processed by rounding up.

[0071] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A fire inspection unmanned aerial vehicle control system, characterized in that, The system comprises: a fire inspection image acquisition module, configured to acquire fire images taken by a UAV under an inspection area; the fire images contain a plurality of fire areas; a historical area comparison module, configured to acquire associated areas corresponding to the fire areas in a plurality of historical images of the inspection area when no fire occurs; the associated areas and the fire areas are areas of the same position, size and shape, and are acquired at different time periods, one fire area in one frame of fire image corresponds to a plurality of associated areas; in each historical image, important information gray values are screened out and corresponding information pixel points are determined according to the frequency of occurrence of the gray values contained in the associated areas; and the original important information aggregation degree of the fire area is obtained according to the information pixel point density and the information uniformity degree in the associated areas in all historical images; a fire area matching module, configured to match two adjacent frames of fire images; in the matching process, for one feature point in one frame of fire image, a matching area is obtained according to the size of the fire area where the feature point is located and the original important information aggregation degree, and a matching point is determined in the matching area of another frame of fire image; a detection frequency adjustment module, configured to connect the feature points matched with each other to obtain a fire matching direction, and adjust the image detection frequency of the UAV according to the difference between the fire matching direction and the flight direction of the UAV; the information uniformity degree is obtained by negatively correlating and normalizing the information entropy in the associated areas.

2. The fire inspection unmanned aerial vehicle control system according to claim 1, wherein, The screening method of the important information gray values comprises: for one historical image, an arbitrary gray value is selected as a target gray value, the pixel point proportion of the target gray value in each associated area is obtained, and an area weight is obtained according to the area of the associated area; the area weight and the pixel point proportion are multiplied to obtain the information degree of the target gray value in each associated area, the information degrees of all associated areas in the historical image are averaged to obtain the information importance degree of the target gray value in the historical image, and the gray value with the information importance degree greater than a preset importance threshold is selected as the important information gray value in the historical image.

3. The fire inspection drone control system of claim 1, wherein, The acquisition method of the original important information aggregation degree comprises: for all associated areas corresponding to one fire area, the product of the information pixel point density and the information uniformity degree in each associated area is taken as the initial information aggregation degree of each associated area, and the initial information aggregation degrees of all associated areas corresponding to the fire area are averaged to obtain the original important information aggregation degree of the fire area.

4. The fire inspection drone control system of claim 1, wherein, The screening method of the feature points comprises: the saliency of each pixel point is obtained by using a saliency algorithm, and the pixel point with the saliency greater than a preset saliency threshold is selected as the feature point.

5. The fire inspection drone control system of claim 1, wherein, The acquisition method of the matching area comprises: The previous frame of fire image is matched with the next frame of fire image, for a feature point on the previous frame of fire image, a first minimum circumscribed rectangle of a fire area to which the feature point belongs is obtained, and a second minimum circumscribed rectangle of a fire area to which a corresponding position pixel point of the feature point on the next frame of fire image belongs is obtained; an intersection-over-union of the first minimum circumscribed rectangle and the second minimum circumscribed rectangle is obtained; a product of the intersection-over-union and a long side of the first minimum circumscribed rectangle is taken as a matching reference edge length; the matching reference edge length is increased by using an original important information aggregation degree of the fire area where the feature point is located, and a matching area size is obtained; and a matching area is constructed on the next frame of fire image with the corresponding position pixel point of the feature point as the center according to the matching area size.

6. The fire inspection drone control system of claim 5, wherein, The matching point acquisition method comprises: The pixel point in the matching area in the next frame of fire image is taken as a to-be-matched point of the feature point; a first pixel value change feature of the feature point in each direction in a neighborhood is obtained, a second pixel value change feature of the to-be-matched point in the same direction in the neighborhood is obtained, a matching degree between the feature point and the to-be-matched point is obtained according to the similarity of the first pixel value change feature and the second pixel value change feature in the same direction, and the to-be-matched point with the largest matching degree is selected as the matching point of the feature point.

7. The fire inspection drone control system of claim 6, wherein, The matching degree acquisition method comprises: The neighborhood is an eight-neighborhood, the pixel value difference between the feature point and the neighborhood pixel point in each direction in the eight-neighborhood is taken as the first pixel value change feature, and the pixel value difference between the to-be-matched point and the neighborhood pixel point in each direction in the eight-neighborhood is taken as the second pixel value change feature; for each direction, the difference between the first pixel value change feature and the second pixel value change feature is negatively correlated and normalized to obtain the pixel value change similarity in each direction; and the average pixel value change similarity in all directions is taken as the matching degree.

8. The fire inspection drone control system of claim 1, wherein, The fire matching direction acquisition method comprises: For an unmanned aerial vehicle image detection frequency adjustment time, a preset time period before the adjustment time is taken as an analysis time period, the pixel points having a matching relationship in the adjacent two frames of fire images in the analysis time period are connected to obtain a plurality of initial directions, and a direction with the most initial direction frequency is selected as the fire matching direction at the adjustment time.

9. The fire inspection drone control system of claim 1, wherein, The image detection frequency of the unmanned aerial vehicle is adjusted according to the difference between the fire matching direction and the flight direction of the unmanned aerial vehicle, comprising: The angle between the fire matching direction and the flight direction of the unmanned aerial vehicle is normalized to obtain an adjustment weight, the sum of the positive integer 1 and the adjustment weight is taken as an increase ratio, and the product of the increase ratio and the image detection frequency is taken as the increased image detection frequency.

Images