Substation fire point identification method and system based on morphological dynamic perception and light point library

Abstract

The present invention discloses a method and system for identifying fire points in a substation based on dynamic morphological perception and a light point library, and belongs to the technical field of fire point identification in electric power facilities. The method comprises: three-dimensional modeling of the substation area to obtain a three-dimensional scene model, and calibrating the positions and observation angles of different cameras in the three-dimensional scene model; respectively obtaining real-time images of the same area taken by cameras with different observation angles at the same time; respectively extracting scene images with the same observation angle as the real-time image from the three-dimensional scene model; and performing multi-directional dynamic identification of fire points based on real-time images and scene images at different observation angles. By constructing a three-dimensional scene model, the present invention can perform multi-directional collaborative analysis based on real-time images and scene images at various observation angles, thereby improving the accuracy of identifying fire points in substations.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer substation illegal behavior identification, and in particular relates to a transformer substation fire point identification method and system based on morphological dynamic perception and a light point library. Background Art

[0002] Substations at all levels are the hubs of the power transmission system. The long-term and stable operation of substations can effectively guarantee industrial production and social livelihood. Therefore, the protection and maintenance of substations is of vital importance. With the rapid development of artificial intelligence and image recognition technology in recent years, the use of video analysis and image recognition to achieve real-time monitoring of violations in substation areas can improve efficiency while ensuring the safety and stability of substations as much as possible. However, the recognition effect of substation area recognition based on traditional single-plane images is easily affected by the environment. Although multi-angle cameras can perform multi-directional observations from different angles, the image data collected by multiple cameras are isolated from each other and cannot be analyzed collaboratively.

[0003] Among the irregularities and violations of regulations at substations, man-made fires are considered the most common but most serious violations. Existing technologies are unable to accurately, intelligently and automatically monitor fire point information at substations under all-weather conditions.

[0004] Therefore, a more effective method for identifying substation violations is needed to automatically and accurately identify violations. Summary of the invention

[0005] In view of this, the present invention proposes a substation fire point identification method and system based on morphological dynamic perception and light point library, which are used to solve the problem of accurately identifying substation violations.

[0006] In a first aspect, the present invention discloses a method for identifying fire points in a substation based on dynamic morphological perception and a light point library, the method comprising:

[0007] Carry out 3D modeling of the substation area to obtain a 3D scene model, and calibrate the positions of different cameras in the 3D scene model;

[0008] Real-time images of the same area taken by cameras with different observation angles at the same time are obtained respectively;

[0009] Extracting scene images having the same observation angle as the real-time image from the three-dimensional scene model respectively;

[0010] Multi-directional fire point identification is performed based on real-time images and scene images from different observation angles.

[0011] On the basis of the above technical solution, preferably, the three-dimensional modeling of the substation area to obtain a three-dimensional scene model, and calibrating the positions and observation angles of different cameras in the three-dimensional scene model specifically includes:

[0012] The targets in the substation area are divided into constant targets and non-constant targets, where constant targets include buildings, fences, and substation equipment in the substation area; others are non-constant targets;

[0013] Establish a 3D model of a constant target and perform texture mapping to construct a 3D scene model;

[0014] The positions and observation angles of different cameras are calibrated so that each location in the substation area can be observed by at least two cameras.

[0015] On the basis of the above technical solution, preferably, the real-time images captured by cameras with different observation angles at the same time on the same area are obtained specifically including:

[0016] The observation environment is divided into three situations according to the sunshine conditions: normal observation, shadow observation and night observation;

[0017] For normal observation situations, obtain video stream data from multiple cameras that can observe the same area, and extract real-time images at the same time under different observation angles;

[0018] For shadow observation, obtain the video stream data of multiple cameras that can observe the same area, extract the real-time images at the same time under different observation angles, and divide each real-time image into a normal area and a shadow area;

[0019] For nighttime observations, cameras installed on top of the substation continuously record the positions and coordinates of lights in non-observed areas, and build a light spot library based on the continuously recorded positions and coordinates of lights in non-observed areas. Video stream data from multiple cameras that can observe the same area are obtained, and normal images and infrared images taken at the same time under different observation angles are extracted as real-time images. Image enhancement processing is performed on infrared images.

[0020] On the basis of the above technical solution, preferably, each real-time image is divided into a normal area and a shadow area specifically including:

[0021] Calculate the azimuth of the sun relative to the substation area based on the current time and date;

[0022] According to the position and coordinates of the constant target and the azimuth of the sun relative to the substation area, the real-time shadow line in the current observation area is calculated, and each real-time image is divided into a normal area and a shadow area according to the real-time shadow line.

[0023] On the basis of the above technical solution, preferably, the multi-directional fire point identification based on real-time images and scene images at different observation angles specifically includes:

[0024] For normal observation situations, the real-time image and scene image of each observation angle are compared, the structural similarity between the images is calculated, and the fire point is identified based on the structural similarity;

[0025] For shadow observation, for normal areas of real-time images, the same fire point identification method as that for normal observation is used to determine whether there are hidden dangers of fire points; for shadow areas of real-time images, the changes in the R component of the RGB values ​​of the pixels in the shadow areas are observed to determine whether there are hidden dangers of fire points;

[0026] For nighttime observations, the bright spot data of normal images are extracted and compared with the light spot library. After eliminating the bright spot data in the light spot library, it is determined whether there is a potential fire hazard. Other violations are identified through infrared images that have been processed through image enhancement.

[0027] On the basis of the above technical solutions, preferably, for normal observation situations, the real-time image and the scene image of each observation angle are compared respectively, the structural similarity between the images is calculated, and the fire point identification based on the structural similarity specifically includes:

[0028] If, in the calculation results of the structural similarity under all observation angles, there is a situation where the structural similarity is greater than the preset similarity threshold and less than or equal to the preset similarity threshold, the real-time image is refreshed, and the structural similarity between the real-time image and the scene image under each observation angle is recalculated;

[0029] If the structural similarity under all observation angles is greater than the preset similarity threshold, it is determined that there is no fire point;

[0030] If the structural similarities under all observation angles are less than or equal to the preset similarity threshold, the real-time image and the scene image are divided into grids, and the structural similarity corresponding to each grid is calculated respectively;

[0031] Grid areas whose structural similarity under all observation angles is less than or equal to the preset similarity threshold are screened out, the screened grid areas are continuously observed and the edge information change rate is calculated through edge detection. If the edge information change rate is greater than the preset change rate threshold, it is determined that there is a fire hazard.

[0032] On the basis of the above technical solution, preferably, for the shadow area of ​​the real-time image, judging whether there is a fire point hazard by observing the change of the R component of the RGB value of the pixel point in the shadow area specifically includes:

[0033] Acquire shadow area images of multiple consecutive frames of real-time images;

[0034] Filter out pixels whose R component of RGB value is greater than a preset component threshold value from each frame of shadow area image;

[0035] Starting from the second frame of the shadow area image, respectively calculate the position change of the pixel points selected in each frame of the shadow area image compared with the pixel points selected in the first frame of the shadow area image, and select the changed frame according to the position change;

[0036] If the number of changed frames exceeds a preset number threshold, it is determined that there is a potential fire point, otherwise it is determined that there is no fire point.

[0037] In a second aspect of the present invention, based on the first aspect of the present invention, a substation violation behavior identification system based on three-dimensional scene assisted perception is proposed, and the system comprises:

[0038] Scene modeling module: used to perform 3D modeling of the substation area, obtain a 3D scene model, and calibrate the positions and observation angles of different cameras in the 3D scene model;

[0039] Image acquisition module: used to obtain real-time images of the same area taken by cameras with different observation angles at the same time;

[0040] Scene extraction module: used to extract scene images with the same observation angle as the real-time image from the three-dimensional scene model;

[0041] Fire point identification module: used for multi-directional fire point identification based on real-time images and scene images from different observation angles.

[0042] In a third aspect of the present invention, an electronic device is disclosed, comprising: at least one processor, at least one memory, a communication interface and a bus;

[0043] Wherein, the processor, memory, and communication interface communicate with each other via the bus;

[0044] The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to the first aspect of the present invention.

[0045] According to a fourth aspect of the present invention, a computer-readable storage medium is disclosed, wherein the computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method described in the first aspect of the present invention.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1) The present invention establishes a three-dimensional scene model of the substation area and combines it with camera calibration at different observation angles to quickly extract scene images with the same observation angle as the real-time image from the three-dimensional scene model. Based on the real-time images and scene images at various observation angles, multi-directional observation and collaborative analysis are performed from different angles to achieve accurate recognition of the observed target, thereby identifying and perceiving the fire point and improving the recognition accuracy.

[0048] 2) The present invention divides the observation environment into three situations according to the sunlight conditions: normal observation, shadow observation and night observation. For each observation situation, real-time images at the same time under different observation angles are extracted respectively, and different strategies are used to identify the fire points, which can effectively improve the recognition accuracy and realize all-weather monitoring.

[0049] 3) The present invention introduces structural similarity calculation and grid grading methods in the fire point identification under normal observation conditions. By calculating the structural similarity of each grid area corresponding to each observation angle, the fire point potential area can be accurately found. Based on the characteristic of continuous and rapid changes in flame morphology, the fire point is identified by the changes in edge pixels in the fire point potential area, thereby improving the accuracy of identification.

[0050] 4) In the fire point recognition under shadow observation, the present invention senses the fire point through the change of R value of RGB data of the pixel point under the shadow, thereby realizing the fire point recognition in the shadow area.

[0051] 5) In the case of night observation, the present invention eliminates the influence of night lights on fire point judgment by constructing a light point library, thereby reducing the misjudgment rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 The present invention is a flow chart of the method for identifying fire points in a substation based on dynamic morphological perception and a light point library. DETAILED DESCRIPTION

[0054] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 The present invention proposes a substation fire point identification method based on morphological dynamic perception and light point library, the method comprising:

[0056] S1. Perform three-dimensional modeling of the substation area to obtain a three-dimensional scene model, and calibrate the positions and observation angles of different cameras in the three-dimensional scene model.

[0057] The targets in the substation area are divided into constant targets and non-constant targets. The constant targets include buildings, fences, and substation equipment in the substation area; gates and other objects in the ground area are non-constant targets, such as movable ladders, boxes, weeds, etc. Since the substation area generally does not change within a certain period of time, constant targets can be selected in advance and a 3D model can be built at one time.

[0058] The three-dimensional modeling of the present invention is mainly aimed at constant targets. First, a three-dimensional vector model is constructed through three-dimensional modeling software; then two-dimensional pictures of the constant target are taken from various angles, and the two-dimensional pictures are pasted as textures to the surface of the three-dimensional vector model to construct a three-dimensional scene model.

[0059] In addition, it is necessary to calibrate the positions and observation angles of different cameras in the 3D scene model so that each location in the substation area can be observed by at least two cameras.

[0060] At this point, the 3D modeling and camera calibration of the substation area have been completed.

[0061] S2. Real-time images of the same area taken by cameras with different observation angles at the same time are obtained respectively.

[0062] Multi-source video stream data collected by cameras at different observation angles need to be processed in order to extract real-time images captured at different observation angles.

[0063] Step S2 specifically includes the following sub-steps:

[0064] S21. Divide the observation environment into three situations according to sunlight conditions: normal observation, shadow observation and night observation.

[0065] Among them, the normal observation situation is the situation where the light in the observation area is good, such as the observation area in the daytime when it is sunny and there is no shadow of buildings or trees; the shadow observation situation is the situation where there is shadow in the observation area, such as the observation area in the daytime when it is cloudy and rainy or sunny but there are shadows of buildings or trees. The normal observation situation and the shadow observation situation can be calculated in real time based on the observation time point and the angle of the observation area. The night observation situation can be calculated based on the current observation time and the sunset time point of the area.

[0066] S22. For normal observation situations, obtain video stream data from multiple cameras that can observe the same area, and extract real-time images at the same time under different observation angles from the video stream data, that is, one real-time image for each observation angle, and these real-time images are at the same time.

[0067] S23. For shadow observation situations, obtain video stream data of multiple cameras that can observe the same area, extract real-time images at the same time under different observation angles, and divide each real-time image into a normal area and a shadow area.

[0068] According to relevant regulations, new buildings and tree planting are prohibited within a certain range around the substation area, so the shadow range of the substation area can be calculated.

[0069] Specifically, according to the current time and date, the azimuth of the sun relative to the substation area is calculated;

[0070] According to the position and coordinates of the constant target and the azimuth of the sun relative to the substation area, the real-time shadow line and shadow area in the current observation area are calculated;

[0071] Obtain video stream data from multiple cameras that can observe the same area, and extract real-time images at the same time from different observation angles;

[0072] Each real-time image is divided into a normal area and a shadow area according to the real-time shadow lines.

[0073] S24. For nighttime observation, obtain video stream data from multiple cameras that can observe the same area, extract normal shooting images and infrared images at the same time under different observation angles as real-time images, and build a light point library.

[0074] For nighttime observations, it is necessary to eliminate the influence of external light (moonlight and lights in non-observation areas). Since moonlight is generally dim and is at a downward angle, it will not affect subsequent data judgments. Lights in non-observation areas may be misjudged as fire points, so data processing is required in advance to facilitate accurate perception of subsequent fire points.

[0075] Specifically, the present invention arranges and calibrates a camera on the top of a substation building. Through the camera, the positions of lights in non-observation areas in three-dimensional space and their coordinates on a plane image taken by the camera at the top are recorded every day. The recording is continued for a certain period of time. A light spot library is constructed based on the continuously recorded positions and coordinates of lights in non-observation areas to eliminate light interference in subsequent nighttime observations.

[0076] For nighttime observation, real-time images include normal images and infrared images. Normal images are used for fire point identification, and infrared images are used for other illegal behavior identification. For infrared images at night, color restoration and image enhancement are required to improve recognition accuracy.

[0077] The specific process of color restoration and image enhancement of infrared images is as follows:

[0078] ①Build the TensorFlow deep learning platform;

[0079] ② Construct a sample library, where the samples are pairs of "nighttime infrared images - normal daytime images"; wherein, based on the acquired nighttime infrared images, the position and shooting angle of the camera can be inferred, and according to the observation angle (orientation, angle) of the camera, the normal daytime images with the same observation angle are extracted from the three-dimensional scene model constructed in step S1;

[0080] ③Build a generative adversarial network model, and expand the input parameters of the traditional generative adversarial network model from one-dimensional random noise to a three-dimensional vector, where the three dimensions correspond to the R, G, and B components in the RGB color space of the image;

[0081] ④ Use the generative adversarial network model in step ③ to train the samples in the sample library, and stop training when the model becomes stable.

[0082] At this point, we have obtained an artificial intelligence model that can convert input nighttime infrared images into normal daytime images.

[0083] The present invention divides the observation environment into three situations according to the sunlight conditions: normal observation, shadow observation and night observation. For each observation situation, real-time images at the same time under different observation angles are extracted respectively, and different strategies are adopted to identify the fire points, which can effectively improve the recognition accuracy and realize all-weather monitoring.

[0084] S3. Extracting scene images having the same observation angle as the real-time image from the three-dimensional scene model.

[0085] The three-dimensional scene model in step S1 has calibrated the position and observation angle of each camera. After obtaining the large real-time image, the corresponding observation position and observation angle can be reversed, so as to locate the camera in the three-dimensional scene model, and extract the image of the desired observation angle from the three-dimensional scene model as the scene image corresponding to the real-time image. The scene image is defined as: a plane image extracted from the three-dimensional scene model at the same observation angle as the camera in the real scene.

[0086] S4. Perform multi-directional fire point identification based on real-time images and scene images from different observation angles.

[0087] Step S4 specifically includes the following sub-steps:

[0088] S41. For normal observation situations, the real-time image and the scene image at each observation angle are compared respectively, the structural similarity between the images is calculated, and the fire point is identified based on the structural similarity.

[0089] Step S41 specifically includes the following sub-steps:

[0090] S411, first calculate the structural similarity SSIM between the real-time image and the scene image at a certain angle. If the SSIM is greater than a preset similarity threshold (e.g., 0.98), it is determined that the two images are similar and there is no fire point; if the SSIM is less than the preset similarity threshold, it indicates that there is a difference between the two images, and it is necessary to calculate the SSIM value between the real-time image and the corresponding scene image at other observation angles;

[0091] S412: if in the calculation results of the structural similarity under all observation angles, there is a situation where the structural similarity is greater than the preset similarity threshold and less than or equal to the preset similarity threshold, then refresh the real-time image after an interval of 5 seconds, and recalculate the structural similarity between the real-time image and the scene image under each observation angle;

[0092] S413: If the structural similarities at all observation angles are greater than a preset similarity threshold, it is determined that there is no fire point;

[0093] S414, if the structural similarities under all observation angles are less than or equal to a preset similarity threshold, the real-time image and the scene image are divided into grids, and the structural similarities between the grid area images corresponding to each grid are calculated respectively;

[0094] Grid area images whose structural similarities under all observation angles are less than or equal to a preset similarity threshold are screened out, and the screened grid area images are continuously observed for 100 frames. The 100 consecutive grid area images are converted into binary space or grayscale space, and the Canny algorithm is used for edge detection. The number of edge pixels in each frame of the grid area image is counted, and the edge information change rate of the 100 frames of the grid area images is calculated according to the ratio of the number of edge pixels of two consecutive frames of images in the binary space. If the edge information change rate is greater than a preset change rate threshold (for example, 20%), it is determined that there is a fire hazard.

[0095] The present invention introduces structural similarity calculation and grid grading methods in the fire point identification under normal observation conditions. By performing structural similarity calculation on each grid area corresponding to each observation angle, the fire point potential area can be accurately found, and the fire point can be identified by the edge pixel changes in the fire point potential area, thereby improving the accuracy of identification.

[0096] S42. For shadow observation situations, for normal areas of real-time images, the same fire point identification method as that for normal observation situations is used to determine whether there are potential fire point hazards; for shadow areas of real-time images, the changes in the R component of the RGB values ​​of the pixels in the shadow areas are observed to determine whether there are potential fire point hazards.

[0097] For the normal area of ​​the real-time image, the normal area is taken as the real-time image, and the same fire point identification method as the normal observation situation in step S41 is used to determine whether there is a fire point hazard.

[0098] For the shadow area of ​​the real-time image, since the fire point appears as a sudden change in brightness in the shadow area image (not obvious under normal sunlight), it can be simplified to calculate the RGB value of each pixel in the shadow area of ​​the real-time image. Since the shape of the flame presents a short-term and continuous change, it will be observed in the video stream that the bright spot area will change repeatedly over time. At the same time, considering that the R component in the RGB color space can best reflect the color characteristics of the flame, it is necessary to continuously observe the bright spots in the continuous frame images in the video stream.

[0099] Specifically, for the shadow area of ​​the real-time image, the method for identifying the fire point is:

[0100] Extract shadow area images of multiple consecutive frames of real-time images;

[0101] Filter out pixels whose R component of RGB value is greater than a preset component threshold value from each frame of shadow area image;

[0102] Starting from the second frame of shadow area image, the position change of the pixels selected in each frame of shadow area image compared with the pixels selected in the first frame of shadow area image is calculated respectively, and the changed frame is selected according to the position change; assuming that a total of 10 consecutive frames of shadow area images are obtained, the positions of the pixels selected in each frame of the following 9 frames of shadow area images are compared with the pixels selected in the first frame. If the position change exceeds 10%, the frame is defined as a changed frame;

[0103] If the number of changed frames exceeds a preset number threshold (eg, 5 frames), it is determined that there is a potential fire point, otherwise it is determined that there is no fire point.

[0104] After a normal image is blocked by a shadow, the RGB components will decrease. Therefore, the R value that increases in the shadow state can be judged as a potential fire point. On the other hand, the R component (red) is used as the fire point because the changes in the flame color and the R component of the substation area are the most obvious. Based on this principle, the present invention designs a fire point detection strategy for the shadow area, which can achieve rapid identification and ensure the accuracy of identification. The reason why the last 9 frames are compared with the first frame instead of directly comparing adjacent frame images is that the interval between the two frames is short and the shape of the fire point may not have changed significantly. Therefore, comparing them with the first frame is more effective.

[0105] S43. For nighttime observation, extract the bright spot data of the normal captured image and compare it with the light spot library. After eliminating the bright spot data in the light spot library, determine whether there is a potential fire hazard; identify other violations through the infrared image after image enhancement processing.

[0106] First, obtain multiple normal images observed by multi-angle cameras in real time under night observation conditions, and extract the bright spot data (highlighted pixels, which may be light spots or fire spots) in each normal image. The RGB value of light is usually constant, or changes slowly and periodically. Considering universality, it is necessary to observe multiple frames of continuous normal images to determine whether the RGB value of the same bright spot continues to change rapidly. If so, it may be a fire spot, and the corresponding bright spot data is retained.

[0107] The acquired bright spot data is compared with the light data in the constructed light spot library, and the bright spots existing in the lamp database are removed from the normal shot image. If there are still remaining bright spots in the image after removal, it is determined to be a fire hazard.

[0108] In the case of nighttime observation, the present invention eliminates the influence of nighttime lights on fire point judgment by constructing a light point library, thereby reducing the misjudgment rate.

[0109] For infrared images observed at night, step S2 has obtained an artificial intelligence model that can convert the input night infrared images into normal daytime images. Based on the normal daytime images obtained by the artificial intelligence model, other illegal behaviors other than fire points are identified. The specific behavior identification method can also use artificial intelligence models constructed by machine learning, deep learning, etc. to identify other illegal behaviors.

[0110] The present invention establishes a three-dimensional scene model of the substation area and combines the camera calibration of different observation angles to quickly extract scene images with the same observation angle as the real-time image from the three-dimensional scene model. Based on the real-time image and scene image at each observation angle, multi-directional observation and collaborative analysis are performed from different angles to achieve accurate recognition of the observed target, thereby identifying and sensing the fire point and improving the recognition accuracy. The present invention can more accurately identify and preliminarily assess the fire situation through multi-angle fire point observation.

[0111] Corresponding to the above method embodiment, the present invention also proposes a substation violation behavior identification system based on three-dimensional scene auxiliary perception, the system comprising:

[0112] Scene modeling module: used to perform 3D modeling of the substation area, obtain a 3D scene model, and calibrate the positions and observation angles of different cameras in the 3D scene model;

[0113] Image acquisition module: used to obtain real-time images of the same area taken by cameras with different observation angles at the same time;

[0114] Scene extraction module: used to extract scene images with the same observation angle as the real-time image from the three-dimensional scene model;

[0115] Fire point identification module: used for multi-directional fire point identification based on real-time images and scene images from different observation angles.

[0116] The above system embodiments and method embodiments correspond one to one, and for a brief description of the system embodiments, please refer to the method embodiments.

[0117] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory, a communication interface and a bus; wherein the processor, memory, and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the aforementioned method of the present invention.

[0118] The present invention also discloses a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, wherein the computer instructions enable the computer to implement all or part of the steps of the method described in the embodiment of the present invention. The storage medium includes: a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk, and other media that can store program codes.

[0119] The system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be distributed to multiple network units. A person skilled in the art may select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment without creative effort.

[0120] 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, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying fire points in a substation based on dynamic morphological perception and light point library, characterized in that: The method comprises: Carry out 3D modeling of the substation area to obtain a 3D scene model, and calibrate the positions and observation angles of different cameras in the 3D scene model; Real-time images of the same area taken by cameras with different observation angles at the same time are obtained respectively; Extracting scene images having the same observation angle as the real-time image from the three-dimensional scene model respectively; Multi-directional fire point identification based on real-time images and scene images from different observation angles; The real-time images captured by cameras with different observation angles at the same time on the same area specifically include: The observation environment is divided into three situations according to the sunshine conditions: normal observation, shadow observation and night observation; For normal observation situations, obtain video stream data from multiple cameras that can observe the same area, and extract real-time images at the same time under different observation angles; For shadow observation, obtain the video stream data of multiple cameras that can observe the same area, extract the real-time images at the same time under different observation angles, and divide each real-time image into a normal area and a shadow area; For nighttime observation, the camera installed on the top of the substation continuously records the position and coordinates of the lights in the non-observation area, and builds a light point library based on the continuously recorded position and coordinates of the lights in the non-observation area; obtains video stream data of multiple cameras that can observe the same area, and extracts normal images and infrared images at the same time under different observation angles as real-time images; performs image enhancement processing on the infrared image; The multi-directional fire point identification based on real-time images and scene images at different observation angles specifically includes: For normal observation situations, the real-time image and scene image of each observation angle are compared, the structural similarity between the images is calculated, and the fire point is identified based on the structural similarity; For shadow observation, for normal areas of real-time images, the same fire point identification method as that for normal observation is used to determine whether there are hidden dangers of fire points; for shadow areas of real-time images, the changes in the R component of the RGB values ​​of the pixels in the shadow areas are observed to determine whether there are hidden dangers of fire points; For nighttime observation, extract the bright spot data of normal shooting images and compare them with the light spot library. After removing the bright spot data in the light spot library, determine whether there is a potential fire hazard. Other illegal behaviors are identified through infrared images after image enhancement processing. For normal observation situations, the real-time image and the scene image at each observation angle are compared respectively, and the structural similarity between the images is calculated. The fire point identification based on the structural similarity specifically includes: If, in the calculation results of the structural similarity under all observation angles, there is a situation where the structural similarity is greater than the preset similarity threshold and less than or equal to the preset similarity threshold, the real-time image is refreshed, and the structural similarity between the real-time image and the scene image under each observation angle is recalculated; If the structural similarity under all observation angles is greater than the preset similarity threshold, it is determined that there is no fire point; If the structural similarities under all observation angles are less than or equal to the preset similarity threshold, the real-time image and the scene image are divided into grids, and the structural similarity corresponding to each grid is calculated respectively; Grid areas whose structural similarity under all observation angles is less than or equal to the preset similarity threshold are screened out, the screened grid areas are continuously observed and the edge information change rate is calculated through edge detection. If the edge information change rate is greater than the preset change rate threshold, it is determined that there is a fire hazard.

2. The method for identifying fire points in substations based on morphological dynamic perception and light point library according to claim 1 is characterized in that: The three-dimensional modeling of the substation area to obtain a three-dimensional scene model and calibrating the positions and observation angles of different cameras in the three-dimensional scene model specifically includes: The targets in the substation area are divided into constant targets and non-constant targets, where constant targets include buildings, fences, and substation equipment in the substation area; others are non-constant targets; Establish a 3D model of a constant target and perform texture mapping to construct a 3D scene model; The positions and observation angles of different cameras are calibrated so that each location in the substation area can be observed by at least two cameras.

3. The method for identifying fire points in substations based on morphological dynamic perception and light point library according to claim 2 is characterized in that: The division of each real-time image into a normal area and a shadow area specifically includes: Calculate the azimuth of the sun relative to the substation area based on the current time and date; According to the position and coordinates of the constant target and the azimuth of the sun relative to the substation area, the real-time shadow line in the current observation area is calculated, and each real-time image is divided into a normal area and a shadow area according to the real-time shadow line.

4. The method for identifying fire points in substations based on morphological dynamic perception and light point library according to claim 2 is characterized in that: For the shadow area of ​​the real-time image, judging whether there is a fire point hazard by observing the change of the R component of the RGB value of the pixel point in the shadow area specifically includes: Acquire shadow area images of multiple consecutive frames of real-time images; Filter out pixels whose R component of RGB value is greater than a preset component threshold value from each frame of shadow area image; Starting from the second frame of the shadow area image, respectively calculate the position change of the pixel points selected in each frame of the shadow area image compared with the pixel points selected in the first frame of the shadow area image, and select the changed frame according to the position change; If the number of changed frames exceeds a preset number threshold, it is determined that there is a potential fire point, otherwise it is determined that there is no fire point.

5. A substation violation behavior identification system based on three-dimensional scene assisted perception using the method described in any one of claims 1 to 4, characterized in that: The system comprises: Scene modeling module: used to perform 3D modeling of the substation area, obtain a 3D scene model, and calibrate the positions and observation angles of different cameras in the 3D scene model; Image acquisition module: used to obtain real-time images of the same area taken by cameras with different observation angles at the same time; Scene extraction module: used to extract scene images with the same observation angle as the real-time image from the three-dimensional scene model; Fire point identification module: used for multi-directional fire point identification based on real-time images and scene images from different observation angles.

6. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; Wherein, the processor, memory, and communication interface communicate with each other via the bus; The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 4.

Images