A power transmission line corridor forest fire positioning method and system based on multi-path video

Abstract

The application discloses a power transmission line corridor forest fire positioning method and system based on multiple videos, and belongs to the technical field of power transmission line forest fire detection. The method is as follows: image segmentation is performed on multiple video data collected along a power transmission line corridor to obtain a to-be-detected image containing a potential forest fire area; a five-dimensional feature vector corresponding to each camera device is extracted from the to-be-detected image to determine a first camera device that captures a real fire source point; initial spatial coordinates of the real fire source point are determined by emitting a ray according to parameter information of the first camera device and pixel coordinates of the real fire source point; error compensation and correction are performed on the initial spatial coordinates based on a GIS database of the power transmission line corridor to obtain real spatial coordinates of the real fire source point. Therefore, by implementing the application, the problems of high positioning cost, narrow positioning range and low positioning accuracy of the power transmission line corridor forest fire in the prior art can be solved.

Description

Technical Field

[0001] This application belongs to the field of power transmission line wildfire detection technology, specifically involving a method and system for locating wildfires along power transmission line corridors based on multi-channel video. Background Technology

[0002] Power transmission lines often extend into high mountains, valleys, and forested areas, where dense vegetation and complex climates make wildfires frequent and serious threats to the safe and stable operation of the power grid. Existing methods for monitoring wildfires along transmission line corridors generally use single-lens cameras to identify smoke and fire targets within the image, lacking ranging and positioning capabilities. They can only issue wildfire alarms and cannot output the actual geographical coordinates of the fire points and their relative distance to the line.

[0003] Existing technologies mostly employ binocular vision positioning, using a fixed baseline and a side-by-side deployment method to locate wildfires along power transmission lines. However, this method is only suitable for small-scale, localized monitoring and cannot adapt to the strip-like extension of power transmission lines spanning tens of kilometers. Furthermore, curves in the transmission lines and obstructions from mountains can directly create large blind spots, making accurate wildfire location difficult. Achieving high-precision wildfire location requires reliance on infrared thermal imaging, satellite remote sensing, and ground-based sensing equipment, resulting in high monitoring and location costs and hindering widespread application. Summary of the Invention

[0004] This application proposes a method and system for locating wildfires along transmission line corridors based on multi-channel video, which can solve the problems of high cost, narrow positioning range and low positioning accuracy in the existing technology for locating wildfires along transmission line corridors.

[0005] The first aspect of this application provides a method for locating wildfires along power transmission line corridors based on multi-channel video, the method comprising: Image segmentation is performed on multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas; wherein, the multiple video data are collected by various camera devices deployed along the transmission line corridor; By extracting features from the potential fire and smoke areas, the five-dimensional feature vectors corresponding to each camera device are extracted from the image to be detected, so as to determine the first camera device that captured the real fire source. Based on the parameter information of the first camera device and the pixel coordinates of the real fire source, the initial spatial coordinates of the real fire source are determined by constructing several projection rays. Based on the GIS database of the power transmission line corridor, the initial spatial coordinates are compensated and corrected for errors to obtain the true spatial coordinates of the actual fire source.

[0006] The aforementioned solution deploys multiple cameras along the power transmission line corridor. Each camera captures footage of the corridor, generating multiple video streams. This eliminates the need for additional infrared, radar, or satellite hardware, addressing the issues of numerous false alarms and inaccurate long-distance positioning in mountainous areas with fog at a physical level. It saves on wildfire location costs and expands the positioning range to the entire power transmission line corridor. The solution first roughly marks areas suspected of containing smoke or fire in the images to be detected, providing targets for subsequent feature extraction. Then, it differentiates and eliminates areas based on the inherent physical differences between real smoke and natural fog. Therefore, it extracts visual features of potential fire or fire areas from multiple dimensions to identify whether fog or smoke is present in the area, thereby determining whether a real fire source has appeared and identifying the location of the camera that captured the real fire source—the first camera. Then, by combining the location and parameters of the first camera device, the initial spatial coordinates of the real fire source point under the geographic coordinates are located through back projection. The coordinates are then verified using a GIS database to eliminate errors caused by the undulating terrain of the mountainous area and the fact that the camera device and the real fire source point are not on the same horizontal plane, so as to obtain more accurate coordinates of the real fire source point.

[0007] In one possible implementation of the first aspect, image segmentation is performed on multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas, specifically: Multiple camera devices are deployed at preset intervals along the transmission line corridor to form various camera systems; The multi-channel video data is split into several video frames, and the video frames are time-aligned and frame offset compensated to obtain standard video frames. The standard video frame is input into a preset semantic segmentation model for region identification, and the potential fireworks region in the standard video frame is marked to obtain the basic parameters of the potential fireworks region; wherein, the potential fireworks region is a semi-transparent, grayish-white and diffuse connected region; the basic parameters include the pixel coordinates, bounding rectangle, pixel area, contour point set and aspect ratio of the potential fireworks region; The image to be detected is obtained by integrating the basic parameters and the labeled standard video frames.

[0008] The above-mentioned scheme deploys multiple camera devices along the transmission line corridor. This deployment method ensures that any section of the transmission corridor, including mountains, bends, and tower sections, has two or more video feeds simultaneously, completely eliminating blind spots in the distance and ravines.

[0009] In one possible implementation of the first aspect, the standard video frame is preprocessed before being input into the semantic segmentation model. The preprocessing operations include dehazing enhancement, Gaussian noise reduction, white balance correction, and contrast stretching.

[0010] The above scheme preprocesses the images before roughly screening the smoke targets, solving common problems in field monitoring such as hazy fog, image noise, uneven lighting, and color cast, thereby improving the recognizability of faint smoke features.

[0011] In one possible implementation of the first aspect, by extracting features from the potential fire and smoke area, five-dimensional feature vectors corresponding to each camera device are extracted from the image to be detected, in order to determine the first camera device that captured the actual fire source. Specifically: Feature detection is performed on the potential smoke and fire area, and texture entropy features, smoke plume morphology features, vertical optical flow features, gray-scale light transmission gradient features, and temporal dilation rate features are extracted to obtain the five-dimensional feature vector; Based on the five-dimensional feature vector, calculate the Euclidean distance and cosine similarity between any two camera devices to obtain the comprehensive similarity between the camera devices; When the overall similarity is greater than or equal to the first threshold, the corresponding camera device is used as the first camera device to capture the real fire source.

[0012] The above scheme takes into account that clouds and fog are homogeneous water vapor with no three-dimensional structure, no dynamic growth, no uplift movement, and a simple texture, while real wildfire smoke is a combustion aerosol with five inherent characteristics: complex texture, feathery drift, thermal uplift, concentration gradient, and continuous expansion. Therefore, based on the above visual characteristics, five different dimensions of visual features are extracted from the potential smoke and fire areas in the image to accurately determine whether the image is clouds and fog or smoke and fire, thereby identifying the first camera device that captured the real fire source, providing a basis for subsequent fire source location.

[0013] In one possible implementation of the first aspect, the texture entropy feature, smoke plume morphology feature, vertical optical flow feature, grayscale transmission gradient feature, and temporal dilation rate feature are extracted respectively to obtain the five-dimensional feature vector, specifically: Based on the grayscale image of the potential fireworks region, a corresponding grayscale co-occurrence matrix is ​​constructed. The texture complexity of the potential fireworks region is calculated using the grayscale co-occurrence matrix to obtain the texture entropy feature. The complete contour of the potential smoke area is extracted by edge detection technology. The average curvature of the contour is calculated and the number of effective feather bifurcation points is counted based on the complete contour to obtain the smoke feather morphology features. The vertical optical flow characteristics are obtained by calculating the mean vertical optical flow of the potential fireworks region by tracking the inter-frame pixel motion; Gray-scale sampling is performed radially from the center point of the potential pyrotechnic area to calculate the average gray-scale gradient amplitude of the pyrotechnics, thereby obtaining the gray-scale light transmission gradient characteristics. The area of ​​the potential pyrotechnic region is continuously collected at fixed sampling intervals to calculate the expansion rate of the potential pyrotechnic region and obtain the temporal expansion rate characteristics.

[0014] In one possible implementation of the first aspect, the initial spatial coordinates of the real fire source are determined by constructing several projection rays based on the parameter information of the first camera device and the pixel coordinates of the real fire source. Specifically: The pixel coordinates of the actual fire source in the image to be detected are obtained, and the pixel coordinates are converted into image coordinates in a local coordinate system using the built-in parameter matrix of the first camera device; wherein, the local coordinate system is the coordinate system adopted by the first camera device; Taking the focus center point of the first camera device as the starting point of the ray, a ray is emitted toward the image coordinates to obtain the direction vector of the ray; By using the peripheral parameter matrix of the first camera device, the direction vector is transformed from the local coordinate system to the geographic coordinate system to obtain the projection ray of the first camera device in the geographic coordinate system; For the same real fire source, the three-dimensional spatial coordinates where all the projected rays intersect are obtained by using the spatial line parametric equations corresponding to the projected rays. These three-dimensional spatial coordinates are the initial spatial coordinates.

[0015] The above method first identifies the first camera device that captures the actual fire source. Then, based on the external and internal parameters of the first camera device, it performs coordinate system transformation on the pixel coordinates of the actual fire source to determine its spatial coordinates in the geographic coordinate system, thereby determining the actual location of the fire source. Compared to deploying or adding hardware such as infrared, radar, and satellite, this wildfire location method only requires the deployment of multiple camera devices, making it more cost-effective, efficient, and economical.

[0016] In one possible implementation of the first aspect, for the same real fire source point, the three-dimensional spatial coordinates where all the projected rays intersect are obtained through the spatial line parameter equations corresponding to the projected rays, specifically: For the first camera device that captures the same real fire source, the spatial line parameter equation is generated for the corresponding projection ray. A system of equations is constructed by simultaneously solving all the parametric equations of the spatial lines. The optimization objective is to minimize the sum of the squares of the spatial distances from the three-dimensional spatial coordinates to all the projected rays. The system of equations is then solved to obtain the three-dimensional spatial coordinates.

[0017] In one possible implementation of the first aspect, based on a GIS database of the transmission line corridor, error compensation and correction are performed on the initial spatial coordinates to obtain the true spatial coordinates of the actual fire source point, specifically as follows: The initial spatial coordinates include planar coordinates and elevation components; From the GIS database, find the standard elevation of the land surface corresponding to the plane coordinates, and calculate the elevation deviation between the standard elevation of the land surface and the elevation components; By combining the preset offset coefficient and the elevation deviation, the planar coordinates are corrected to obtain the true spatial coordinates.

[0018] The above scheme takes into account factors such as mountainous terrain undulations, long-distance perspective effects, and lens imaging scaling, which can lead to systematic errors in the calculated coordinates. Therefore, the GIS database is introduced to compensate and correct the initial spatial coordinates. By performing terrain matching compensation on the plane coordinates, the positional offset caused by elevation shift is eliminated.

[0019] One possible implementation of the first aspect also includes: Based on the channel protection red line range provided by the GIS database, the actual spatial coordinates are subjected to limit detection, and coordinate points falling outside the channel protection red line range are eliminated; Identify the pole closest to the actual fire source and calculate its coordinates based on the GIS database; Based on the coordinates of the tower, the actual spatial coordinates, and the coordinates of the attachment point with the transmission line, the horizontal and vertical safe distances from the actual fire source to the transmission line are calculated.

[0020] The aforementioned scheme also distinguishes whether the actual fire source is located within the transmission line area, thereby eliminating fire sources unrelated to the transmission line and outside the scope of power grid monitoring and control, thus eliminating invalid location results. Furthermore, considering that the further the fire source is from the camera equipment, the more pronounced the perspective scaling effect of image pixels becomes; tiny pixel shifts are amplified into spatial coordinate deviations (pixel scaling deviations), leading to distortion in the distance calculation of the fire point relative to the towers and conductors. Therefore, the scheme finds the tower closest to the fire source and performs pixel scaling to achieve scale compensation, obtaining a true and reliable location of the actual fire source.

[0021] The second aspect of this application provides a multi-channel video-based system for locating wildfires along a power transmission line corridor. The system includes: an image acquisition module, an image feature extraction module, a fire source coordinate calculation module, and an error correction module. The image acquisition module is used to segment multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas; wherein, the multiple video data are collected by various camera devices deployed along the transmission line corridor; The image feature extraction module is used to extract the five-dimensional feature vectors corresponding to each camera device from the image to be detected by performing feature extraction on the potential fire area, so as to determine the first camera device that captured the real fire source. The fire source coordinate calculation module is used to determine the initial spatial coordinates of the real fire source by constructing several projection rays based on the parameter information of the first camera device and the pixel coordinates of the real fire source. The error correction module is used to perform error compensation and correction on the initial spatial coordinates based on the GIS database of the power transmission line corridor, so as to obtain the true spatial coordinates of the actual fire source. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the specific process of a method for locating wildfires in a power transmission line corridor based on multiple video feeds, provided in one embodiment of this application. Figure 2 This is a structural diagram of a power transmission line corridor wildfire location system based on multi-channel video provided in one embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0026] First Embodiment To address the blind spot problem in strip-shaped power transmission lines, this application's embodiment pioneers a strip-shaped multi-channel overlapping network mode specifically for power transmission line corridors. This includes: for special scenarios where power transmission lines are long, strip-shaped, have many curves, and are obstructed by mountains, multiple camera devices are deployed along the transmission line corridor at preset intervals (generally one camera device every 2-3 km), and the effective field-of-view overlap rate between adjacent camera devices is maintained at no less than 35% throughout the entire process. This equipment deployment method ensures that any section of the transmission line corridor, including mountains, curves, and towers, has simultaneous coverage by two or more video feeds, completely eliminating remote blind spots and ravine blind spots. Furthermore, it breaks through the limitations of traditional binocular fixed-point monitoring, achieving continuous monitoring of ultra-long lines without blind spots, perfectly adapting to the narrow strip-shaped power line maintenance scenario.

[0027] The embodiments of this application are based on the above-mentioned strip-shaped multi-path overlapping networking mode to locate wildfires in the transmission line corridor.

[0028] like Figure 1 As shown, to address the problems of high cost, narrow positioning range, and low positioning accuracy in locating wildfires along transmission line corridors in the prior art, the first embodiment of this application provides a detailed flowchart of a method for locating wildfires along transmission line corridors based on multi-channel video. This embodiment's method includes steps S1 to S4, detailed below: Step S1: Perform image segmentation on the multiple video data collected along the transmission line corridor to obtain the image to be detected, which contains potential smoke and fire areas.

[0029] Based on the aforementioned strip-shaped multi-path overlapping network mode, multiple video data streams are captured on the transmission line corridor and its surrounding area. Simultaneously with video capture, a globally unified timestamp calibration is performed on the multiple video data streams. Specifically, starting from the first camera position detecting a suspected fire source, the existence duration of valid targets (i.e., suspected fire sources) in consecutive frames is continuously counted. If a valid target disappears beyond a set interval, the timing is interrupted and the count is restarted. The total continuous detection duration of the suspected fire source is finally obtained. Based on this total continuous detection duration, the acquired multiple video data streams are time-aligned.

[0030] After timing alignment, frame offset compensation is performed on the multi-channel video data to control the frame timing deviation of the video data captured by multiple cameras within 20ms, thereby achieving strict frame-level synchronization.

[0031] After the above steps, the multi-channel video data is split into corresponding standard video frames to obtain standard video frames output by multiple camera devices that are spatially fully covered and temporally synchronized. This eliminates the spatiotemporal asynchrony error for subsequent cross-camera feature comparison and multi-ray intersection positioning.

[0032] The standard video frames are input into a preset semantic segmentation model for region identification, marking the potential smoke and fire regions in the standard video frames and obtaining the basic parameters of the potential smoke and fire regions. The semantic segmentation model is specifically trained for the gray-white smoke pattern of power transmission corridors and can automatically identify and segment semi-transparent, gray-white, and diffuse connected regions in the image. These connected regions carry significant smoke / fire features and can therefore be called potential smoke and fire regions.

[0033] Before region identification, all standard video frames undergo preprocessing to address common issues in field monitoring such as fog, image noise, uneven lighting, and color cast, thereby improving the recognizability of faint smoke and fire features. The preprocessing operations include defogging enhancement, Gaussian noise reduction, white balance correction, and contrast stretching.

[0034] In the process of labeling potential fireworks regions, the semantic segmentation model also records the basic parameters of each potential fireworks region, including the pixel coordinates, bounding rectangle, pixel area, contour point set, and aspect ratio of the potential fireworks region.

[0035] Finally, the semantic segmentation model outputs the image to be detected with the potential fireworks region labeled, as well as the basic parameters of the potential fireworks region.

[0036] The above process only roughly filters out fire sources and does not distinguish between real smoke / fire (related to the fire source) and clouds / fog (unrelated to the fire source). The identification of real fire sources will be carried out in the following steps.

[0037] Step S2: By extracting features from the potential fire and smoke area, the five-dimensional feature vectors corresponding to each camera device are extracted from the image to be detected, so as to determine the first camera device that captured the real fire source.

[0038] Although clouds and smoke are similar, they differ visually. Clouds and smoke are homogeneous water vapor, lacking three-dimensional structure, dynamic growth, upward movement, and simple texture; real wildfire smoke, on the other hand, is combustion aerosol, possessing five inherent characteristics: complex texture, feathery drift, thermal rise, concentration gradient, and continuous expansion. Therefore, this application embodiment extracts corresponding five-dimensional feature vectors from multiple camera angles based on these five inherent characteristics to determine whether smoke exists in the image to be detected, thereby identifying the first camera device that captured the real fire source.

[0039] First, feature detection is performed on the potential fireworks region, extracting texture entropy features, smoke plume morphology features, vertical optical flow features, gray-level translucency gradient features, and temporal expansion rate features to obtain the five-dimensional feature vector. The extraction process of the five-dimensional feature vector is as follows: A gray-level co-occurrence matrix is ​​constructed based on the grayscale image of the potential fireworks region; the texture complexity of the potential fireworks region is calculated using the gray-level co-occurrence matrix to obtain the texture entropy feature; the complete contour of the potential fireworks region is extracted using edge detection technology; the average curvature of the contour is calculated and the number of effective plume bifurcation points is counted based on the complete contour to obtain the smoke plume morphology feature; the mean vertical optical flow of the potential fireworks region is calculated by tracking inter-frame pixel motion to obtain the vertical optical flow feature; grayscale sampling is performed radially from the center point of the potential fireworks region to calculate the average grayscale gradient amplitude of the fireworks to obtain the gray-level translucency gradient feature; the area of ​​the potential fireworks region is continuously sampled at fixed sampling intervals to calculate the expansion rate of the fireworks to obtain the temporal expansion rate feature.

[0040] 1. Determination and extraction of the texture entropy features.

[0041] Based on the grayscale image of the potential fireworks region, a corresponding grayscale co-occurrence matrix is ​​constructed. The texture complexity of the potential fireworks region is calculated using the Shannon information entropy formula through the grayscale co-occurrence matrix. The specific calculation formula is as follows: ; in, H The complexity of the texture, i , j The pixel coordinates in the grayscale image. p ( i , j ) represents the normalized probability of pairs of gray levels appearing in the neighborhood.

[0042] if H <Base entropy and If the frame fluctuation threshold is less than 1, it is determined that the texture of the potential fireworks area is uniform and static, and that it has cloud and fog features. if H≥ Baseline entropy or ≥ If the frame fluctuation threshold is used, it is determined that the texture of the potential fireworks area is messy and variable and has smoke features. Therefore, the corresponding texture entropy features are extracted from the potential fireworks area.

[0043] Optionally, in this embodiment of the application, the base entropy is set to 3.2 and the frame fluctuation threshold is set to 0.4.

[0044] 2. Determination and extraction of the plume-like morphological features of the smoke.

[0045] The complete contour of the potential pyrotechnic area is extracted using the Canny edge detection method, and the average curvature of the contour is calculated based on the complete contour. K And count the number of valid feather bifurcation points N .

[0046] If K < 0.025 and N < 3, then the edge of the potential fireworks area is determined to be smooth and without bifurcation, belonging to fog; If K ≥ 0.025 or N ≥ 3, then the edge of the potential pyrotechnic area is determined to be wavy and bifurcated, possessing a smoke feather structure, and thus belongs to smoke.

[0047] 3. Determination and extraction of the vertical optical flow features.

[0048] The dense optical flow algorithm is used to track the inter-frame pixel motion of the potential pyrotechnic region and calculate the mean vertical optical flow of the potential pyrotechnic region. .

[0049] if If the value is less than 1.2, it is determined that there is no lifting motion in the potential smoke and fire area and that static clouds and fog exist. if If the value is greater than or equal to 1.2, it is determined that there is smoke in the potential pyrotechnic area and that the smoke has the characteristic of rising upwards.

[0050] 4. Determination and extraction of the grayscale transmission gradient features.

[0051] Gray-level sampling is performed radially outward from the center point of the potential pyrotechnic area to calculate the average gray-level gradient amplitude. .

[0052] if If the value is less than 8, the potential pyrotechnic area is determined to have uniform grayness and consistent light transmission, and is therefore classified as homogeneous cloud / fog. if If the value is greater than 8, the potential pyrotechnic area is determined to have a denser smoke at the center and a lighter smoke at the edges, exhibiting the characteristic of stratified light transmission due to smoke concentration.

[0053] 5. Determination and extraction of the temporal expansion rate characteristics.

[0054] The area of ​​the potential pyrotechnic zone is continuously sampled at fixed intervals, and the expansion rate of the potential pyrotechnic zone is calculated. The calculation formula is: ; in, The area of ​​the potential pyrotechnic zone. The sampling interval is typically 1 second.

[0055] if If the value is less than 45, the potential pyrotechnic area is determined to be stable and without expansion, and is therefore considered to be fog. if If the value is greater than or equal to 45, the potential pyrotechnic area is determined to be continuously spreading and constitutes smoke.

[0056] Based on the data collected above, the five-dimensional feature vector of each camera device is determined, and its expression is: .

[0057] Based on the five-dimensional feature vector, the Euclidean distance and cosine similarity between any two camera devices are calculated to obtain the comprehensive similarity between the camera devices, so as to determine which camera positions have the same shooting target.

[0058] To characterize the degree of numerical difference between different camera positions, the Euclidean distance between the two camera devices is calculated: ; in, d The Euclidean distance between the two camera devices. F A For camera equipment A The five-dimensional eigenvectors, F B For camera equipment B The five-dimensional eigenvectors.

[0059] To characterize the consistency of feature change trends across different camera positions, the cosine similarity between the two camera devices is calculated: ; in, For camera equipment A With camera equipment B Cosine similarity between them.

[0060] The comprehensive similarity is calculated using the following formula: ; Among them, S sim The overall similarity, d max This represents the maximum value of the Euclidean distance.

[0061] When the overall similarity is greater than or equal to the first threshold, it will be determined that the two corresponding camera devices have captured the actual fire source, and these two camera devices will be designated as the first camera devices. Other camera devices that have not been determined to have captured the actual fire source will not be considered in the following steps.

[0062] Optionally, in this embodiment of the application, the first threshold is set to 0.75.

[0063] Step S3: Based on the parameter information of the first camera device and the pixel coordinates of the real fire source, the initial spatial coordinates of the real fire source are determined by constructing several projection rays.

[0064] The built-in and peripheral parameters of the first camera device are collected. The built-in parameters include the lens's inherent optical parameters such as focal length, principal point coordinates, and distortion coefficient; the peripheral parameters include the three-dimensional coordinates corresponding to the GPS device, mounting height, pitch angle, and yaw angle, which describe the device's spatial position and attitude in the geographic coordinate system.

[0065] Based on the built-in parameters, construct the corresponding built-in parameter matrix; based on the peripheral parameters, construct the corresponding peripheral parameter matrix.

[0066] This application employs a spatial back-projection method, transforming the pixel coordinates of the actual fire source point in the image to be detected using the built-in parameter matrix to obtain the image coordinates of the pixel coordinates in a local coordinate system. This local coordinate system is the coordinate system used by the first camera device. Simultaneously, distortion correction is applied to the image coordinates using the lens distortion coefficient of the first camera device to eliminate positional deviations caused by lens deformation.

[0067] Using the focus center point of the first camera device as the starting point of the ray, a ray is emitted towards the image coordinates after distortion correction to obtain the direction vector of the ray.

[0068] Then, using the peripheral parameter matrix, the direction vector is transformed from the local coordinate system to the geographic coordinate system to obtain the projection ray of the first camera device in the geographic coordinate system.

[0069] The projected rays from each of the first cameras capturing the same real fire source are obtained, and a corresponding spatial line parametric equation is constructed for each projected ray, with the specific expression as follows: ; Where X, Y, and Z are the three-dimensional spatial coordinates of the actual fire source; X ci Y ci Z ci Let be the coordinates of the focus center point of the i-th first camera device. t Let be the extension step size of the ray. a i , b i , c i For the first i The direction vector of the ray.

[0070] The above spatial line parametric equations are grouped into a system of equations. Ideally, all rays would intersect precisely at the three-dimensional spatial point of the actual fire source. However, in actual engineering, due to installation deviations, minor vibrations, and residual distortions, the rays will not perfectly intersect at a single point. Therefore, a least-squares optimal solution is introduced to calculate the spatial line parametric equations to locate the initial spatial coordinates of the actual fire source in the geographic coordinate system.

[0071] The optimization objective is to minimize the sum of the squared spatial distances from the three-dimensional spatial coordinates to all the projected rays. The least squares method is used to solve the system of equations to obtain the three-dimensional spatial coordinates. During the calculation, equations introduced by other first-camera devices can be used to offset errors caused by device installation offsets, residual lens distortion, and minor posture jitter, suppressing positioning divergence and obtaining more accurate initial spatial coordinates of the true fire source.

[0072] Finally, the initial spatial coordinates are converted into the latitude, longitude, and altitude format commonly used by the power grid.

[0073] Step S4: Based on the GIS database of the power transmission line corridor, perform error compensation and correction on the initial spatial coordinates to obtain the true spatial coordinates of the actual fire source.

[0074] In step S3, coordinate calculation is performed solely based on the camera equipment, failing to consider that factors such as mountainous terrain undulations, long-distance perspective effects, and lens imaging scaling can all lead to systematic deviations in the calculated coordinates. Therefore, this embodiment introduces a GIS database of the transmission line corridor, performing error compensation, boundary filtering, and scale calibration on the initial spatial coordinates in three layers to obtain more accurate and reliable spatial coordinates of the actual fire source points that meet power grid operation and maintenance standards. The GIS database contains a large amount of real-measured power grid spatial data, such as the terrain where the transmission line is located, the corridor route, and the precise coordinates of the towers / hanging points.

[0075] In this embodiment, the initial spatial coordinates are first corrected to planar coordinates, then the coordinate points are checked to see if they are within the line protection zone, and finally the long-distance pixel scaling deviation is corrected. The specific steps S4.1 to S4.3 are as follows: (1) Correct the planar coordinates. Considering the undulating terrain of the mountainous area, the camera equipment and the fire source are not on the same horizontal plane. The oblique perspective projection will cause the elevation value of the initial spatial coordinates calculated by geometry to shift (perspective distortion), and the planar coordinates will also shift with the elevation deviation. Therefore, it is necessary to correct the planar coordinates of the initial spatial coordinates. X 0, Y 0) Perform terrain matching compensation.

[0076] In the GIS database, find the surface standard elevation corresponding to the plane coordinates (i.e., the plane coordinates in the database).X 0, Y 0) Z corresponding to the entire transmission line corridor dem ), calculate the elevation deviation between the standard elevation of the land surface and the elevation components.

[0077] An offset coefficient is constructed based on the camera's installation height, pitch angle, and terrain slope. This offset coefficient, combined with the elevation deviation, is used to correct the planar coordinates, yielding the true spatial coordinates. The specific expression is as follows: ; Where (X1,Y1,Z1) are the corrected initial spatial coordinates, that is, the true spatial coordinates of the actual fire source. k This is the offset coefficient. The elevation deviation is mentioned.

[0078] (2) Check whether the coordinate point is located within the line protection zone. Step S3 only performs spatial geometric calculations and cannot distinguish whether the actual fire source is located within the transmission line protection zone. For example, although external targets such as distant mountain fires and roadside farmland burning are also considered as actual fire sources, they are not closely related to the transmission line and do not fall within the scope of power grid monitoring and control. They are invalid positioning results and need to be screened out and removed.

[0079] Vector data of the protection red line of the transmission line corridor is extracted from the GIS database, including the coordinates of the closed geographic polygons of the line centerline and the legal protection boundaries on both sides, in order to construct the range of the corridor protection red line.

[0080] The system checks whether the actual spatial coordinates fall within the protection red line of the channel. If the coordinates are within the protection red line, the actual fire source is determined to be an irrelevant target such as a distant mountain fire or farmland burning. The corresponding positioning data is then directly removed, and the subsequent process is terminated.

[0081] If the coordinates are located outside the protection red line of the passage, the real fire source is determined to be a valid target, and its real spatial coordinates are retained.

[0082] (3) Correcting pixel scaling deviation at long distances. Because the farther the real fire source is from the camera device, the more obvious the perspective scaling effect of the image pixels becomes; small pixel offsets will be amplified into spatial coordinate deviations (pixel scaling deviations), causing the distance calculation of the real fire source relative to the pole and conductor to be distorted. Therefore, it is necessary to offset the real spatial coordinates and calibrate the relative distance with the pole and conductor.

[0083] If the relative distance between the camera and the actual fire source exceeds 1000m, it is considered a long-distance observation and requires pixel scaling deviation correction.

[0084] First, obtain the tower closest to the actual fire source and its standard coordinates, as well as the coordinates of the conductor suspension point.

[0085] Based on the standard coordinates of the tower, calculate the initial line-of-sight distance between the tower and the actual spatial coordinates, as well as the image pixel scaling factor (this factor is calibrated by the camera focal length and the actual observation distance; the greater the distance, the larger the factor).

[0086] The real spatial coordinates are compensated based on the initial viewing distance, and the specific calculation formula is as follows: ; Where (X2,Y2,Z2) are the compensated real space coordinates. s This is the scaling factor for image pixels.

[0087] Based on the compensated real spatial coordinates and the coordinates of the hanging point, calculate the horizontal and vertical safe distances from the real fire source to the power transmission line.

[0088] Furthermore, this application embodiment also sets four levels of early warning: blue, yellow, orange, and red. The alarm level of the real fire source is determined by combining three indicators: continuous detection duration, feature matching confidence, and fireball expansion rate. The level increases with each level, and the corresponding danger gradually worsens.

[0089] The continuous detection duration refers to the duration during which potential smoke and fire areas persist in multiple video streams. If a potential smoke and fire area disappears beyond a set interval, the timing is interrupted and the accumulation restarts. The feature matching confidence score is determined by normalization and weighted summation; the closer the value is to 1, the higher the matching degree with real smoke features. The fireball expansion rate is the area increase of the potential smoke and fire area per unit time, directly representing the speed of fire spread.

[0090] The alarm levels for real fire sources provided in this application embodiment are shown in Table 1 below.

[0091] Table 1 Alarm Levels for Actual Ignition Sources

[0092] Implementing the embodiments of this application has the following beneficial effects: This application embodiment deploys multiple camera devices along the power transmission line corridor. Each camera captures images of the corridor, obtaining multiple video feeds. This eliminates the need for additional infrared, radar, or satellite hardware, addressing the issues of numerous false alarms and inaccurate long-distance positioning in mountainous foggy conditions at a physical level. It saves on wildfire location costs and expands the positioning range to the entire power transmission line corridor. In the images to be detected, areas suspected of being fire hazards are first roughly marked to provide targets for subsequent feature extraction. Then, based on the inherent physical differences between real smoke and natural fog, these areas are distinguished and eliminated. Therefore, visual features of potential fire hazard areas are extracted from multiple dimensions to identify whether fog or smoke exists in the area, thereby determining whether a real fire source has appeared and identifying the location of the camera that captured the real fire source—the first camera device. Then, by combining the location and parameters of the first camera device, the initial spatial coordinates of the real fire source point under the geographic coordinates are located through back projection. The coordinates are then verified using a GIS database to eliminate errors caused by the undulating terrain of the mountainous area and the fact that the camera device and the real fire source point are not on the same horizontal plane, so as to obtain more accurate coordinates of the real fire source point.

[0093] Second Embodiment Furthermore, in order to implement the multi-channel video-based power transmission line corridor wildfire positioning system corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, Figure 2 A detailed structural diagram of a multi-channel video-based wildfire location system for power transmission line corridors is provided. For ease of explanation, only the parts relevant to this embodiment are shown. The multi-channel video-based wildfire location system for power transmission line corridors provided in this application embodiment includes: The image acquisition module 201 is used to perform image segmentation on multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas; wherein, the multiple video data are acquired by various camera devices deployed along the transmission line corridor.

[0094] Based on the aforementioned strip-shaped multi-path overlapping network mode, multiple video data streams are captured on the transmission line corridor and its surrounding area. Simultaneously with video capture, a globally unified timestamp calibration is performed on the multiple video data streams. Specifically, starting from the first camera position detecting a suspected fire source, the existence duration of valid targets (i.e., suspected fire sources) in consecutive frames is continuously counted. If a valid target disappears beyond a set interval, the timing is interrupted and the count is restarted. The total continuous detection duration of the suspected fire source is finally obtained. Based on this total continuous detection duration, the acquired multiple video data streams are time-aligned.

[0095] After timing alignment, frame offset compensation is performed on the multi-channel video data to control the frame timing deviation of the video data captured by multiple cameras within 20ms, thereby achieving strict frame-level synchronization.

[0096] After the above steps, the multi-channel video data is split into corresponding standard video frames to obtain standard video frames output by multiple camera devices that are spatially fully covered and temporally synchronized. This eliminates the spatiotemporal asynchrony error for subsequent cross-camera feature comparison and multi-ray intersection positioning.

[0097] The standard video frames are input into a preset semantic segmentation model for region identification, marking the potential smoke and fire regions in the standard video frames and obtaining the basic parameters of the potential smoke and fire regions. The semantic segmentation model is specifically trained for the gray-white smoke pattern of power transmission corridors and can automatically identify and segment semi-transparent, gray-white, and diffuse connected regions in the image. These connected regions carry significant smoke / fire features and can therefore be called potential smoke and fire regions.

[0098] Before region identification, all standard video frames undergo preprocessing to address common issues in field monitoring such as fog, image noise, uneven lighting, and color cast, thereby improving the recognizability of faint smoke and fire features. The preprocessing operations include defogging enhancement, Gaussian noise reduction, white balance correction, and contrast stretching.

[0099] In the process of labeling potential fireworks regions, the semantic segmentation model also records the basic parameters of each potential fireworks region, including the pixel coordinates, bounding rectangle, pixel area, contour point set, and aspect ratio of the potential fireworks region.

[0100] Finally, the semantic segmentation model outputs the image to be detected with the potential fireworks region labeled, as well as the basic parameters of the potential fireworks region.

[0101] The above process only roughly filters out fire sources and does not distinguish between real smoke / fire (related to the fire source) and clouds / fog (unrelated to the fire source). The identification of real fire sources will be carried out in the following steps.

[0102] The image feature extraction module 202 is used to extract the five-dimensional feature vectors corresponding to each camera device from the image to be detected by performing feature extraction on the potential fire area, so as to determine the first camera device that captured the real fire source.

[0103] Although clouds and smoke are similar, they differ visually. Clouds and smoke are homogeneous water vapor, lacking three-dimensional structure, dynamic growth, upward movement, and simple texture; real wildfire smoke, on the other hand, is combustion aerosol, possessing five inherent characteristics: complex texture, feathery drift, thermal rise, concentration gradient, and continuous expansion. Therefore, this application embodiment extracts corresponding five-dimensional feature vectors from multiple camera angles based on these five inherent characteristics to determine whether smoke exists in the image to be detected, thereby identifying the first camera device that captured the real fire source.

[0104] First, feature detection is performed on the potential fireworks region, extracting texture entropy features, smoke plume morphology features, vertical optical flow features, gray-level translucency gradient features, and temporal expansion rate features to obtain the five-dimensional feature vector. The extraction process of the five-dimensional feature vector is as follows: A gray-level co-occurrence matrix is ​​constructed based on the grayscale image of the potential fireworks region; the texture complexity of the potential fireworks region is calculated using the gray-level co-occurrence matrix to obtain the texture entropy feature; the complete contour of the potential fireworks region is extracted using edge detection technology; the average curvature of the contour is calculated and the number of effective plume bifurcation points is counted based on the complete contour to obtain the smoke plume morphology feature; the mean vertical optical flow of the potential fireworks region is calculated by tracking inter-frame pixel motion to obtain the vertical optical flow feature; grayscale sampling is performed radially from the center point of the potential fireworks region to calculate the average grayscale gradient amplitude of the fireworks to obtain the gray-level translucency gradient feature; the area of ​​the potential fireworks region is continuously sampled at fixed sampling intervals to calculate the expansion rate of the fireworks to obtain the temporal expansion rate feature.

[0105] 1. Determination and extraction of the texture entropy features.

[0106] Based on the grayscale image of the potential fireworks region, a corresponding grayscale co-occurrence matrix is ​​constructed. The texture complexity of the potential fireworks region is calculated using the Shannon information entropy formula through the grayscale co-occurrence matrix. The specific calculation formula is as follows: ; in, H The complexity of the texture, i , j The pixel coordinates in the grayscale image. p ( i , j ) represents the normalized probability of pairs of gray levels appearing in the neighborhood.

[0107] if H <Base entropy and If the frame fluctuation threshold is less than 1, it is determined that the texture of the potential fireworks area is uniform and static, and that it has cloud and fog features. if H≥ Baseline entropy or ≥ If the frame fluctuation threshold is used, it is determined that the texture of the potential fireworks area is messy and variable and has smoke features. Therefore, the corresponding texture entropy features are extracted from the potential fireworks area.

[0108] Optionally, in this embodiment of the application, the base entropy is set to 3.2 and the frame fluctuation threshold is set to 0.4.

[0109] 2. Determination and extraction of the plume-like morphological features of the smoke.

[0110] The complete contour of the potential pyrotechnic area is extracted using the Canny edge detection method, and the average curvature of the contour is calculated based on the complete contour. K And count the number of valid feather bifurcation points N .

[0111] If K < 0.025 and N < 3, then the edge of the potential fireworks area is determined to be smooth and without bifurcation, belonging to fog; If K ≥ 0.025 or N ≥ 3, then the edge of the potential pyrotechnic area is determined to be wavy and bifurcated, possessing a smoke feather structure, and thus belongs to smoke.

[0112] 3. Determination and extraction of the vertical optical flow features.

[0113] The dense optical flow algorithm is used to track the inter-frame pixel motion of the potential pyrotechnic region and calculate the mean vertical optical flow of the potential pyrotechnic region. .

[0114] if If the value is less than 1.2, it is determined that there is no lifting motion in the potential smoke and fire area and that static clouds and fog exist. if If the value is greater than or equal to 1.2, it is determined that there is smoke in the potential pyrotechnic area and that the smoke has the characteristic of rising upwards.

[0115] 4. Determination and extraction of the grayscale transmission gradient features.

[0116] Gray-level sampling is performed radially outward from the center point of the potential pyrotechnic area to calculate the average gray-level gradient amplitude. .

[0117] if If the value is less than 8, the potential pyrotechnic area is determined to have uniform grayness and consistent light transmission, and is therefore classified as homogeneous cloud / fog. if If the value is greater than 8, the potential pyrotechnic area is determined to have a denser smoke at the center and a lighter smoke at the edges, exhibiting the characteristic of stratified light transmission due to smoke concentration.

[0118] 5. Determination and extraction of the temporal expansion rate characteristics.

[0119] The area of ​​the potential pyrotechnic zone is continuously sampled at fixed intervals, and the expansion rate of the potential pyrotechnic zone is calculated. The calculation formula is: ; in, The area of ​​the potential pyrotechnic zone. The sampling interval is typically 1 second.

[0120] if If the value is less than 45, the potential pyrotechnic area is determined to be stable and without expansion, and is therefore considered to be fog. if If the value is greater than or equal to 45, the potential pyrotechnic area is determined to be continuously spreading and constitutes smoke.

[0121] Based on the data collected above, the five-dimensional feature vector of each camera device is determined, and its expression is: .

[0122] Based on the five-dimensional feature vector, the Euclidean distance and cosine similarity between any two camera devices are calculated to obtain the comprehensive similarity between the camera devices, so as to determine which camera positions have the same shooting target.

[0123] To characterize the degree of numerical difference between different camera positions, the Euclidean distance between the two camera devices is calculated: ; in, d The Euclidean distance between the two camera devices. F A For camera equipment A The five-dimensional eigenvectors, F B For camera equipment B The five-dimensional eigenvectors.

[0124] To characterize the consistency of feature change trends across different camera positions, the cosine similarity between the two camera devices is calculated: ; in, For camera equipment A With camera equipment B Cosine similarity between them.

[0125] The comprehensive similarity is calculated using the following formula: ; Among them, S sim The overall similarity, d max This represents the maximum value of the Euclidean distance.

[0126] When the overall similarity is greater than or equal to the first threshold, it will be determined that the two corresponding camera devices have captured the actual fire source, and these two camera devices will be designated as the first camera devices. Other camera devices that have not been determined to have captured the actual fire source will not be considered in the following steps.

[0127] Optionally, in this embodiment of the application, the first threshold is set to 0.75.

[0128] The fire source coordinate calculation module 203 is used to determine the initial spatial coordinates of the real fire source by constructing several projection rays based on the parameter information of the first camera device and the pixel coordinates of the real fire source.

[0129] The built-in and peripheral parameters of the first camera device are collected. The built-in parameters include the lens's inherent optical parameters such as focal length, principal point coordinates, and distortion coefficient; the peripheral parameters include the three-dimensional coordinates corresponding to the GPS device, mounting height, pitch angle, and yaw angle, which describe the device's spatial position and attitude in the geographic coordinate system.

[0130] Based on the built-in parameters, construct the corresponding built-in parameter matrix; based on the peripheral parameters, construct the corresponding peripheral parameter matrix.

[0131] This application employs a spatial back-projection method, transforming the pixel coordinates of the actual fire source point in the image to be detected using the built-in parameter matrix to obtain the image coordinates of the pixel coordinates in a local coordinate system. This local coordinate system is the coordinate system used by the first camera device. Simultaneously, distortion correction is applied to the image coordinates using the lens distortion coefficient of the first camera device to eliminate positional deviations caused by lens deformation.

[0132] Using the focus center point of the first camera device as the starting point of the ray, a ray is emitted towards the image coordinates after distortion correction to obtain the direction vector of the ray.

[0133] Then, using the peripheral parameter matrix, the direction vector is transformed from the local coordinate system to the geographic coordinate system to obtain the projection ray of the first camera device in the geographic coordinate system.

[0134] The projected rays from each of the first cameras capturing the same real fire source are obtained, and a corresponding spatial line parametric equation is constructed for each projected ray, with the specific expression as follows: ; Where X, Y, and Z are the three-dimensional spatial coordinates of the actual fire source; X ci Y ci Z ci Let be the coordinates of the focus center point of the i-th first camera device. t Let be the extension step size of the ray. a i , b i , c i For the first i The direction vector of the ray.

[0135] The above spatial line parametric equations are grouped into a system of equations. Ideally, all rays would intersect precisely at the three-dimensional spatial point of the actual fire source. However, in actual engineering, due to installation deviations, minor vibrations, and residual distortions, the rays will not perfectly intersect at a single point. Therefore, a least-squares optimal solution is introduced to calculate the spatial line parametric equations to locate the initial spatial coordinates of the actual fire source in the geographic coordinate system.

[0136] The optimization objective is to minimize the sum of the squared spatial distances from the three-dimensional spatial coordinates to all the projected rays. The least squares method is used to solve the system of equations to obtain the three-dimensional spatial coordinates. During the calculation, equations introduced by other first-camera devices can be used to offset errors caused by device installation offsets, residual lens distortion, and minor posture jitter, suppressing positioning divergence and obtaining more accurate initial spatial coordinates of the true fire source.

[0137] Finally, the initial spatial coordinates are converted into the latitude, longitude, and altitude format commonly used by the power grid.

[0138] Error correction module 204 is used to perform error compensation and correction on the initial spatial coordinates based on the GIS database of the power transmission line corridor to obtain the true spatial coordinates of the actual fire source.

[0139] The above calculations are based solely on coordinate solving using camera equipment, failing to consider the systematic biases caused by mountainous terrain undulations, long-distance perspective effects, and lens zoom. Therefore, this embodiment introduces a GIS database of the transmission line corridor, performing error compensation, boundary filtering, and scale calibration on the initial spatial coordinates in three layers to obtain more accurate and reliable spatial coordinates of the actual fire source points that meet power grid operation and maintenance standards. The GIS database contains a large amount of real-measured power grid spatial data, such as the terrain where the transmission line is located, the corridor route, and the precise coordinates of the towers / hanging points.

[0140] In this embodiment, the initial spatial coordinates are first corrected to planar coordinates, then the coordinate points are checked to see if they are within the line protection zone, and finally the long-distance pixel scaling deviation is corrected. The specific steps S4.1 to S4.3 are as follows: (1) Correct the planar coordinates. Considering the undulating terrain of the mountainous area, the camera equipment and the fire source are not on the same horizontal plane. The oblique perspective projection will cause the elevation value of the initial spatial coordinates calculated by geometry to shift (perspective distortion), and the planar coordinates will also shift with the elevation deviation. Therefore, it is necessary to correct the planar coordinates of the initial spatial coordinates. X 0, Y 0) Perform terrain matching compensation.

[0141] In the GIS database, find the surface standard elevation corresponding to the plane coordinates (i.e., the plane coordinates in the database).X 0, Y 0) Z corresponding to the entire transmission line corridor dem ), calculate the elevation deviation between the standard elevation of the land surface and the elevation components.

[0142] An offset coefficient is constructed based on the camera's installation height, pitch angle, and terrain slope. This offset coefficient, combined with the elevation deviation, is used to correct the planar coordinates, yielding the true spatial coordinates. The specific expression is as follows: ; Where (X1,Y1,Z1) are the corrected initial spatial coordinates, that is, the true spatial coordinates of the actual fire source. k This is the offset coefficient. The elevation deviation is mentioned.

[0143] (2) Check whether the coordinate point is located within the line protection zone. Step S3 only performs spatial geometric calculations and cannot distinguish whether the actual fire source is located within the transmission line protection zone. For example, although external targets such as distant mountain fires and roadside farmland burning are also considered as actual fire sources, they are not closely related to the transmission line and do not fall within the scope of power grid monitoring and control. They are invalid positioning results and need to be screened out and removed.

[0144] Vector data of the protection red line of the transmission line corridor is extracted from the GIS database, including the coordinates of the closed geographic polygons of the line centerline and the legal protection boundaries on both sides, in order to construct the range of the corridor protection red line.

[0145] The system checks whether the actual spatial coordinates fall within the protection red line of the channel. If the coordinates are within the protection red line, the actual fire source is determined to be an irrelevant target such as a distant mountain fire or farmland burning. The corresponding positioning data is then directly removed, and the subsequent process is terminated.

[0146] If the coordinates are located outside the protection red line of the passage, the real fire source is determined to be a valid target, and its real spatial coordinates are retained.

[0147] (3) Correcting pixel scaling deviation at long distances. Because the farther the real fire source is from the camera device, the more obvious the perspective scaling effect of the image pixels becomes; small pixel offsets will be amplified into spatial coordinate deviations (pixel scaling deviations), causing the distance calculation of the real fire source relative to the pole and conductor to be distorted. Therefore, it is necessary to offset the real spatial coordinates and calibrate the relative distance with the pole and conductor.

[0148] If the relative distance between the camera and the actual fire source exceeds 1000m, it is considered a long-distance observation and requires pixel scaling deviation correction.

[0149] First, obtain the tower closest to the actual fire source and its standard coordinates, as well as the coordinates of the conductor suspension point.

[0150] Based on the standard coordinates of the tower, calculate the initial line-of-sight distance between the tower and the actual spatial coordinates, as well as the image pixel scaling factor (this factor is calibrated by the camera focal length and the actual observation distance; the greater the distance, the larger the factor).

[0151] The real spatial coordinates are compensated based on the initial viewing distance, and the specific calculation formula is as follows: ; Where (X2,Y2,Z2) are the compensated real space coordinates. s This is the scaling factor for image pixels.

[0152] Based on the compensated real spatial coordinates and the coordinates of the hanging point, calculate the horizontal and vertical safe distances from the real fire source to the power transmission line.

[0153] Furthermore, this application embodiment also sets four levels of early warning: blue, yellow, orange, and red. The alarm level of the real fire source is determined by combining three indicators: continuous detection duration, feature matching confidence, and fireball expansion rate. The level increases with each level, and the corresponding danger gradually worsens.

[0154] The continuous detection duration refers to the duration during which potential smoke and fire areas persist in multiple video streams. If a potential smoke and fire area disappears beyond a set interval, the timing is interrupted and the accumulation restarts. The feature matching confidence score is determined by normalization and weighted summation; the closer the value is to 1, the higher the matching degree with real smoke features. The fireball expansion rate is the area increase of the potential smoke and fire area per unit time, directly representing the speed of fire spread.

[0155] Implementing the embodiments of this application has the following beneficial effects: This application embodiment deploys multiple camera devices along the power transmission line corridor. Each camera captures images of the corridor, obtaining multiple video feeds. This eliminates the need for additional infrared, radar, or satellite hardware, addressing the issues of numerous false alarms and inaccurate long-distance positioning in mountainous foggy conditions at a physical level. It saves on wildfire location costs and expands the positioning range to the entire power transmission line corridor. In the images to be detected, areas suspected of being fire hazards are first roughly marked to provide targets for subsequent feature extraction. Then, based on the inherent physical differences between real smoke and natural fog, these areas are distinguished and eliminated. Therefore, visual features of potential fire hazard areas are extracted from multiple dimensions to identify whether fog or smoke exists in the area, thereby determining whether a real fire source has appeared and identifying the location of the camera that captured the real fire source—the first camera device. Then, by combining the location and parameters of the first camera device, the initial spatial coordinates of the real fire source point under the geographic coordinates are located through back projection. The coordinates are then verified using a GIS database to eliminate errors caused by the undulating terrain of the mountainous area and the fact that the camera device and the real fire source point are not on the same horizontal plane, so as to obtain more accurate coordinates of the real fire source point.

[0156] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, or improvements made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for locating wildfires along power transmission line corridors based on multi-channel video, characterized in that, include: Image segmentation is performed on multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas; wherein, the multiple video data are collected by various camera devices deployed along the transmission line corridor; By extracting features from the potential fire and smoke areas, the five-dimensional feature vectors corresponding to each camera device are extracted from the image to be detected, so as to determine the first camera device that captured the real fire source. Based on the parameter information of the first camera device and the pixel coordinates of the real fire source, the initial spatial coordinates of the real fire source are determined by constructing several projection rays. Based on the GIS database of the power transmission line corridor, the initial spatial coordinates are compensated and corrected for errors to obtain the true spatial coordinates of the actual fire source.

2. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 1, characterized in that, Image segmentation was performed on multiple video data collected along the transmission line corridor to obtain the images to be detected, which contain potential smoke and fire areas. Specifically: Multiple camera devices are deployed at preset intervals along the transmission line corridor to form various camera systems; The multi-channel video data is split into several video frames, and the video frames are time-aligned and frame offset compensated to obtain standard video frames. The standard video frame is input into a preset semantic segmentation model for region identification, and the potential fireworks region in the standard video frame is marked to obtain the basic parameters of the potential fireworks region; wherein, the potential fireworks region is a semi-transparent, grayish-white and diffuse connected region; the basic parameters include the pixel coordinates, bounding rectangle, pixel area, contour point set and aspect ratio of the potential fireworks region; The image to be detected is obtained by integrating the basic parameters and the labeled standard video frames.

3. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 2, characterized in that, Before inputting the standard video frame into the semantic segmentation model, the standard video frame is preprocessed, including dehazing enhancement, Gaussian noise reduction, white balance correction, and contrast stretching.

4. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 1, characterized in that, By extracting features from the potential pyrotechnic area, five-dimensional feature vectors corresponding to each camera device are extracted from the image to be detected, in order to determine the first camera device that captured the actual fire source. Specifically: Feature detection is performed on the potential smoke and fire area, and texture entropy features, smoke plume morphology features, vertical optical flow features, gray-scale light transmission gradient features, and temporal dilation rate features are extracted to obtain the five-dimensional feature vector; Based on the five-dimensional feature vector, calculate the Euclidean distance and cosine similarity between any two camera devices to obtain the comprehensive similarity between the camera devices; When the overall similarity is greater than or equal to the first threshold, the corresponding camera device is used as the first camera device that captured the real fire source.

5. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 4, characterized in that, The texture entropy feature, smoke plume morphology feature, vertical optical flow feature, gray-level transmittance gradient feature, and temporal dilation rate feature are extracted respectively to obtain the five-dimensional feature vector, specifically: Based on the grayscale image of the potential fireworks region, a corresponding grayscale co-occurrence matrix is ​​constructed. The texture complexity of the potential fireworks region is calculated using the grayscale co-occurrence matrix to obtain the texture entropy feature. The complete contour of the potential smoke area is extracted by edge detection technology. The average curvature of the contour is calculated and the number of effective feather bifurcation points is counted based on the complete contour to obtain the smoke feather morphology features. The vertical optical flow characteristics are obtained by calculating the mean vertical optical flow of the potential fireworks region by tracking the inter-frame pixel motion; Gray-scale sampling is performed radially from the center point of the potential pyrotechnic area to calculate the average gray-scale gradient amplitude of the pyrotechnics, thereby obtaining the gray-scale light transmission gradient characteristics. The area of ​​the potential pyrotechnic region is continuously collected at fixed sampling intervals to calculate the expansion rate of the potential pyrotechnic region and obtain the temporal expansion rate characteristics.

6. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 1, characterized in that, Based on the parameter information of the first camera device and the pixel coordinates of the actual fire source, the initial spatial coordinates of the actual fire source are determined by constructing several projection rays, specifically as follows: The pixel coordinates of the actual fire source in the image to be detected are obtained, and the pixel coordinates are converted into image coordinates in a local coordinate system using the built-in parameter matrix of the first camera device; wherein, the local coordinate system is the coordinate system adopted by the first camera device; Taking the focus center point of the first camera device as the starting point of the ray, a ray is emitted toward the image coordinates to obtain the direction vector of the ray; By using the peripheral parameter matrix of the first camera device, the direction vector is transformed from the local coordinate system to the geographic coordinate system to obtain the projection ray of the first camera device in the geographic coordinate system; For the same real fire source, the three-dimensional spatial coordinates where all the projected rays intersect are obtained by using the spatial line parametric equations corresponding to the projected rays. These three-dimensional spatial coordinates are the initial spatial coordinates.

7. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 6, characterized in that, For the same real fire source point, the three-dimensional spatial coordinates where all the projected rays intersect are obtained through the spatial line parameter equations corresponding to the projected rays, specifically: For the first camera device that captures the same real fire source, the spatial line parameter equation is generated for the corresponding projection ray. A system of equations is constructed by simultaneously solving all the parametric equations of the spatial lines. The optimization objective is to minimize the sum of the squares of the spatial distances from the three-dimensional spatial coordinates to all the projected rays. The system of equations is then solved to obtain the three-dimensional spatial coordinates.

8. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 1, characterized in that, Based on the GIS database of the power transmission line corridor, the initial spatial coordinates are error-compensated and corrected to obtain the true spatial coordinates of the actual fire source point, specifically: The initial spatial coordinates include planar coordinates and elevation components; From the GIS database, find the standard elevation of the land surface corresponding to the plane coordinates, and calculate the elevation deviation between the standard elevation of the land surface and the elevation components; By combining the preset offset coefficient and the elevation deviation, the planar coordinates are corrected to obtain the true spatial coordinates.

9. The method for locating wildfires along power transmission line corridors based on multi-channel video as described in claim 8, characterized in that, Also includes: Based on the channel protection red line range provided by the GIS database, the actual spatial coordinates are subjected to limit detection, and coordinate points falling outside the channel protection red line range are eliminated; Identify the pole closest to the actual fire source and calculate its coordinates based on the GIS database; Based on the coordinates of the tower, the actual spatial coordinates, and the coordinates of the attachment point with the transmission line, the horizontal and vertical safe distances from the actual fire source to the transmission line are calculated.

10. A wildfire location system for power transmission line corridors based on multi-channel video, characterized in that, include: Image acquisition module, image feature extraction module, fire source coordinate calculation module, and error correction module; The image acquisition module is used to segment multiple video data collected along the transmission line corridor to obtain a detection image containing potential smoke and fire areas; wherein, the multiple video data are collected by various camera devices deployed along the transmission line corridor; The image feature extraction module is used to extract the five-dimensional feature vectors corresponding to each camera device from the image to be detected by performing feature extraction on the potential fire area, so as to determine the first camera device that captured the real fire source. The fire source coordinate calculation module is used to determine the initial spatial coordinates of the real fire source by constructing several projection rays based on the parameter information of the first camera device and the pixel coordinates of the real fire source. The error correction module is used to perform error compensation and correction on the initial spatial coordinates based on the GIS database of the power transmission line corridor to obtain the true spatial coordinates of the actual fire source.

Images