A method and system for online monitoring of power transmission lines combining point cloud and video.

Abstract

This invention relates to the field of power system automation and image recognition technology, and discloses a method and system for online monitoring of transmission lines combining point cloud and video. The method includes: acquiring three-dimensional point cloud data of insulators using lidar and simultaneously acquiring multispectral video frames; denoising, registering, and reconstructing the surface of the point cloud to generate a high-fidelity geometric model; enhancing, correcting distortion, and spatiotemporally aligning the video to form a texture mapping atlas; fusing and constructing a joint pollution characterization model to calculate the pollution layer thickness, volume density, and spatial continuity indicators, and generating cleaning priority decision instructions accordingly. The system includes a three-dimensional point cloud acquisition unit, a multispectral video acquisition unit, a point cloud processing unit, a video processing unit, a joint modeling unit, a pollution quantification unit, and a decision output unit. This invention achieves accurate three-dimensional quantification of pollution through geometric-optical fusion, improving the scientific nature of cleaning decisions and operational efficiency.

Description

Technical Field

[0001] This invention belongs to the field of power system automation and image recognition technology, specifically relating to an online monitoring method and system for transmission lines that combines point cloud and video. Background Technology

[0002] With the deepening of smart grid construction, online monitoring of the operational status of key transmission line equipment, such as insulators, has become a crucial link in ensuring the safety and stability of the power system. Current mainstream monitoring methods largely rely on visible light or infrared video images for pollution assessment. The core principle is to indirectly infer the degree of pollution by analyzing two-dimensional visual features such as the color, grayscale, and coverage area of ​​polluted regions. However, such methods are inherently limited by planar imaging mechanisms and cannot obtain three-dimensional spatial distribution information of pollution on the insulator surface, especially making it difficult to quantify the actual thickness of the pollution layer and its local accumulation morphology.

[0003] Since the impact of contamination on insulation performance depends not only on the apparent area, but also on its physical thickness, density and uniformity of adhesion, relying solely on two-dimensional images can easily lead to premature or delayed cleaning decisions, which may result in a waste of maintenance resources and may also cause flashover accidents due to the failure to detect thick layers of contamination.

[0004] Point cloud-based 3D sensing technology offers a new approach to solving the aforementioned problems. LiDAR or structured light scanning can acquire high-precision 3D point cloud data of the insulator surface, which theoretically can reflect the geometric deformation caused by pollution deposition.

[0005] Existing point cloud processing methods mostly focus on geometric reconstruction or simple curvature analysis, lacking the ability to effectively characterize the complex surface topological changes induced by dirt. For example, dirt filling may cause existing micropores to disappear or form new depressions. Such changes in topological features contain key information about dirt distribution, but traditional point cloud algorithms struggle to extract discriminative quantitative indicators from them.

[0006] In existing technologies, both purely visual methods and purely point cloud methods suffer from inherent limitations in terms of limited information dimensions and one-sided feature representation. While video analysis can capture rich texture and color depth information, it lacks spatial thickness; although point clouds possess three-dimensional geometric advantages, they struggle to directly characterize the material properties and optical characteristics of dirt. Using either in isolation fails to construct a complete understanding of the state of dirt.

[0007] Especially under complex weather conditions or long-term aging scenarios, pollution exhibits complex forms such as non-uniformity, multi-layered structures, and mixed states. The limitations of single-modal data are further amplified, resulting in a severe lack of accuracy in existing monitoring systems for quantifying pollution thickness, determining risk levels, and deciding on cleaning timing. Therefore, there is an urgent need for an online monitoring method that deeply integrates video texture semantics and point cloud topology to achieve high-precision, three-dimensional, and quantifiable intelligent assessment of the pollution status of transmission line insulators. Summary of the Invention

[0008] This invention provides a method and system for online monitoring of transmission lines that combines point cloud and video data. It aims to address the technical problem that existing image-based methods can only assess the color and area of ​​contaminants, but cannot quantify the three-dimensional thickness and spatial distribution of contaminants, leading to inaccurate cleaning decisions. This invention integrates high-precision three-dimensional point cloud data with multispectral video images to construct a three-dimensional geometric model and joint characterization of material properties of contaminants on the insulator surface. This enables precise quantification of contaminant layer thickness, volume, spatial coverage density, and adhesion morphology, and based on this, generates physically interpretable cleaning priority decision instructions.

[0009] This invention provides an online monitoring method for power transmission lines that combines point cloud and video data, comprising: Three-dimensional point cloud data of insulator strings and their fittings are acquired by lidar sensors deployed on transmission towers; Multispectral video frame sequences at corresponding viewpoints are acquired by synchronously triggered visible light and ultraviolet imaging sensors; The three-dimensional point cloud data is denoised, registered, and surface reconstructed to generate a high-fidelity three-dimensional geometric model of the insulator. The multispectral video frame sequence is subjected to image enhancement, distortion correction and spatiotemporal alignment processing to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model; The texture mapping atlas is projected onto the surface of the three-dimensional geometric model to construct a joint dirt characterization model that integrates geometric structure and optical reflection properties; Based on the aforementioned pollution joint characterization model, the pollution layer thickness, volume density, and spatial continuity index of each region of the insulator are calculated. Based on the contamination layer thickness, bulk density, and spatial continuity indicators, an insulator cleaning priority score is generated, and a cleaning decision instruction is output.

[0010] Preferably, the three-dimensional point cloud data is subjected to denoising, registration, and surface reconstruction processing to generate a high-fidelity three-dimensional geometric model of the insulator, including: A statistical outlier removal algorithm is used to eliminate outliers caused by atmospheric disturbances or birds. Initial coarse registration of multi-view point cloud fragments is performed using Feature Descriptor FPFH, followed by registration using the Iterative Closest Point Algorithm. The Poisson surface reconstruction algorithm is used to generate a closed and watertight triangular mesh model, preserving the geometric details at the connection between the umbrella skirt edge and the steel foot.

[0011] Preferably, the multispectral video frame sequence is subjected to image enhancement, distortion correction, and spatiotemporal alignment processing to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model, including: Perform illumination equalization processing on visible light images based on Retinex theory to suppress local overexposure or underexposure caused by backlighting or strong glare. Background noise suppression and hotspot enhancement are performed on ultraviolet images to improve the signal-to-noise ratio of weak corona discharge signals; Radial and tangential distortion corrections are performed on all images using a pre-calibrated camera intrinsic parameter matrix and lens distortion coefficients; By using known sensor extrinsic parameters, a one-to-one mapping relationship is established between the calibrated image pixel coordinate system and the point cloud world coordinate system, achieving pixel-level spatial alignment.

[0012] Preferably, the texture mapping atlas is projected onto the surface of the three-dimensional geometric model to construct a joint dirt characterization model that integrates geometric structure and optical reflection properties, including: On each triangular facet of the three-dimensional geometric model, the corresponding visible light reflectance value and ultraviolet radiation intensity value are bound together. The area where dirt exists is defined as a surface area that simultaneously meets the following two conditions: the visible light reflectance is less than a preset threshold, and the ultraviolet radiation intensity is greater than three times the standard deviation of the mean background noise. For areas identified as contaminated, the geometric deformation characteristics caused by contamination accumulation are further identified based on the rate of change of the point cloud normal vector and the surface curvature gradient.

[0013] Preferably, based on the aforementioned joint pollution characterization model, the pollution layer thickness, bulk density, and spatial continuity index of each region of the insulator are calculated, including: The thickness of the contamination layer is obtained by comparing the Euclidean distance between the current point cloud surface and the clean state benchmark model, which is trained using point cloud data from historical periods without contamination. Bulk density is defined as the average thickness of the dirt layer per unit surface area, calculated by integral calculation according to the umbrella skirt section; The spatial continuity index is determined by calculating the number of connected components in a polluted area and the proportion of the largest connected area.

[0014] Preferably, based on the contamination layer thickness, bulk density, and spatial continuity indicators, an insulator cleaning priority score is generated, and a cleaning decision instruction is output, including: Set cleaning priority scoring function for: ; To normalize the thickness, To normalize the volume density, As an indicator of spatial continuity; When the score is greater than 0.7, the insulator is marked as "emergency cleaning"; when the score is between 0.4 and 0.7, it is marked as "planned cleaning"; when the score is less than 0.4, it is marked as "cleaning not required". The cleaning decision instruction includes the insulator number, the location of the tower to which it belongs, a heat map of the pollution distribution, the recommended cleaning method, and a forecast of the expected pollution growth trend.

[0015] Preferably, the cleanliness benchmark model is stored in a contamination benchmark database, which contains three-dimensional geometric models of insulators of the same type in their factory condition, post-rain cleanliness condition, and historical cleaning records, and is used to dynamically update the calculation benchmark for the current contamination thickness.

[0016] Preferably, the dirt distribution heatmap uses color gradations to represent the dirt layer thickness, the recommended cleaning methods include high-pressure water rinsing, mechanical scrubbing, or chemical cleaning, and the predicted dirt growth trend is generated based on an exponential growth model fitted to historical dirt data.

[0017] This invention also provides an online monitoring system for power transmission lines that combines point cloud and video, comprising: The three-dimensional point cloud acquisition unit is used to acquire three-dimensional point cloud data of insulator strings and their fittings through lidar sensors deployed on transmission towers; The multispectral video acquisition unit is used to acquire multispectral video frame sequences from corresponding viewpoints through synchronously triggered visible light and ultraviolet imaging sensors; The point cloud processing unit is used to perform denoising, registration and surface reconstruction processing on the three-dimensional point cloud data to generate a high-fidelity three-dimensional geometric model of the insulator. The video processing unit is used to perform image enhancement, distortion correction and spatiotemporal alignment processing on the multispectral video frame sequence to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model; A joint modeling unit is used to project the texture mapping atlas onto the surface of the three-dimensional geometric model to construct a joint characterization model of dirt that integrates geometric structure and optical reflection properties; The pollution quantification unit is used to calculate the pollution layer thickness, volume density, and spatial continuity index of each region of the insulator based on the pollution joint characterization model. The decision output unit is used to generate an insulator cleaning priority score based on the contamination layer thickness, volume density, and spatial continuity index, and output a cleaning decision instruction.

[0018] Preferably, when the joint modeling unit performs texture-geometry fusion, it uses the centroid coordinate interpolation method to map the image pixel attributes to the vertices of the triangular mesh, and performs surface rendering through the Phong lighting model to verify the visual consistency of the dirty area. The decision output unit uploads the cleaning decision command to the provincial power transmission monitoring center through the power dispatch data network and pushes it to the mobile terminal of the operation and maintenance personnel simultaneously. The command format conforms to the IEC61850 communication standard.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention integrates high-precision three-dimensional point cloud measurement with multispectral video perception depth, breaking through the limitations of traditional two-dimensional image methods that can only obtain the color and projected area of ​​the contaminant layer, and realizing three-dimensional quantitative characterization of the thickness, volume, spatial distribution morphology and adhesion continuity of the contaminant layer of transmission line insulators.

[0020] 2. By constructing a geometric-optical joint model, this invention can distinguish between surface-attached dirt and structural defects, avoiding misjudgment; the cleaning priority score generated based on physically interpretable quantitative indicators significantly improves the targeting of cleaning operations and the efficiency of resource utilization.

[0021] 3. The system adopts a synchronous triggering and spatiotemporal alignment mechanism to ensure strict consistency of multi-source data in time and space dimensions, providing a reliable data foundation for subsequent intelligent diagnosis.

[0022] 4. The method solves the problem of over-cleaning or under-cleaning caused by insufficient pollution state perception, extends the service life of insulators, reduces the operation and maintenance cost of power grid, and improves the operational reliability of transmission lines in heavily polluted environments. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram illustrating the core principle framework for constructing the combined pollution characterization model in this invention; Figure 3 This is a flowchart illustrating the logical process framework for 3D point cloud data processing and high-fidelity geometric model generation in this invention. Figure 4 This is a flowchart illustrating the logical flow of multispectral video frame sequence processing and texture mapping atlas generation in this invention. Figure 5 This is a flowchart illustrating the logical process of calculating the quantification index of contamination and generating the cleaning priority score in this invention. Figure 6 This is a schematic diagram of the multi-level interaction relationship and data flow between the point cloud acquisition unit, the multispectral video acquisition unit, and the edge computing platform in this invention. Detailed Implementation

[0024] refer to Figures 1 to 6 This invention provides a method and system for online monitoring of transmission lines that combines point cloud and video data. It aims to address the technical problem that existing image-based methods can only assess the color and area of ​​contaminants, but cannot quantify the three-dimensional thickness and spatial distribution of contaminants, leading to inaccurate cleaning decisions. This method integrates high-precision three-dimensional point cloud data with multispectral video images to construct a three-dimensional geometric model and joint characterization of material properties of contaminants on the insulator surface. This enables precise quantification of contaminant layer thickness, volume, spatial coverage density, and adhesion morphology, and generates physically interpretable cleaning priority decision instructions based on this.

[0025] The online monitoring method for transmission lines combining point cloud and video includes the following steps: S1 acquires three-dimensional point cloud data of insulator strings and their fittings through lidar sensors deployed on transmission towers; S2, acquires multispectral video frame sequences from the corresponding viewpoint through synchronously triggered visible light and ultraviolet imaging sensors; S3, the three-dimensional point cloud data is denoised, registered and surface reconstructed to generate a high-fidelity three-dimensional geometric model of the insulator; S4, perform image enhancement, distortion correction and spatiotemporal alignment processing on the multispectral video frame sequence to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model; S5, Project the texture mapping atlas onto the surface of the three-dimensional geometric model to construct a dirt joint characterization model that integrates geometric structure and optical reflection properties; S6. Based on the pollution joint characterization model, calculate the pollution layer thickness, volume density and spatial continuity index of each region of the insulator; S7. Based on the contamination layer thickness, volume density, and spatial continuity indicators, generate an insulator cleaning priority score and output a cleaning decision command.

[0026] In step S1, a lidar sensor deployed on the transmission tower acquires three-dimensional point cloud data of the insulator string and its fittings. This lidar uses a pulsed solid-state laser with a wavelength of 1550 nanometers, possessing eye-safe characteristics and suitable for long-term outdoor operation. Its scanning mechanism is a multi-beam rotating scan, covering a horizontal field of view of 120 degrees and a vertical field of view of 30 degrees, ensuring the complete capture of the insulator string's overall structure. The sampling frequency is set to 100,000 points per second, with each surround scan lasting 5 seconds. At least three independent scans from different azimuths are performed to eliminate point cloud gaps caused by single-view occlusion.

[0027] Each point cloud segment has a spatial resolution greater than 5 mm and a distance measurement accuracy better than ±2 mm, meeting the high-fidelity reproduction requirements for key geometric details such as the edges of insulator skirts and the connections of steel feet. The point cloud data is transmitted in real time to the edge computing platform via an industrial-grade gigabit Ethernet interface, and the transmission protocol supports the IEEE 1588 precision time protocol, providing a time reference for subsequent multi-source data synchronization.

[0028] In step S2, a multispectral video frame sequence at the corresponding viewing angle is acquired by synchronously triggered visible light and ultraviolet imaging sensors. The visible light sensor is a global shutter CMOS image sensor with a resolution of 4096×2160 pixels, a frame rate of 30 Hz, and a dynamic range greater than 120 dB, which can effectively cope with the complex lighting conditions of strong light, backlight, or alternating shadows at the power transmission line site. The ultraviolet imaging sensor uses a solar-blind ultraviolet enhanced CCD with a response band limited to 240 to 280 nanometers. This band is in the "solar-blind zone" where solar radiation is completely absorbed by the atmospheric ozone layer, which can effectively suppress background light interference and is specifically used to capture weak corona discharge signals caused by pollution.

[0029] UV sensor sensitivity greater than 10 -18 The power density per square centimeter per steradian is sufficient to detect early partial discharge phenomena. Two sensors share the same optical window, and optical paths are separated via a built-in beam splitter, ensuring a field-of-view center overlap error of less than 0.1 degrees. Hardware trigger signals are uniformly issued by the edge computing platform, with time synchronization errors controlled within less than 1 millisecond. This ensures that each video frame strictly corresponds to the same LiDAR scan moment, laying the foundation for subsequent spatiotemporal alignment.

[0030] In step S3, the 3D point cloud data undergoes denoising, registration, and surface reconstruction to generate a high-fidelity 3D geometric model of the insulator. First, a statistical outlier removal operation is performed, setting the number of neighboring points to 20 and the standard deviation factor to 1.5. Outliers introduced by atmospheric disturbances, bird flight, or electromagnetic interference are removed, retaining valid point clouds that conform to local geometric consistency. Subsequently, multi-view registration is performed on point cloud fragments from different orientations. Initial coarse registration uses the Fast Point Feature Histogram (FPFH) descriptor to extract local geometric features, and the optimal rigid body transformation matrix is ​​solved using the RANSAC algorithm to achieve approximate alignment between fragments. Based on this, the Iterative Closest Point (ICP) algorithm is used for registration, with the iteration termination condition being a root mean square error of less than 0.5 mm or an iteration count of 500.

[0031] After registration, all point clouds are merged into a single global point cloud set. Finally, a closed and watertight triangular mesh model is generated using the Poisson surface reconstruction algorithm. This algorithm recovers the implicit surface by solving the Poisson equation, with the octree depth set to 10 to balance model detail and computational efficiency. The final number of triangular mesh patches is controlled between 500,000 and 800,000, accurately preserving the umbrella skirt outline, steel foot threads, and surface micro-deformations that may be caused by dirt accumulation.

[0032] In step S4, image enhancement, distortion correction, and spatiotemporal alignment are performed on the multispectral video frame sequence to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model. For visible light images, illumination equalization processing based on Retinex theory is performed, decomposing the image into illuminance and reflectance components. Dynamic range compression is applied to the low-frequency illuminance component, while details are preserved in the high-frequency reflectance component, thereby suppressing local overexposure or underexposure areas caused by direct sunlight or cloud cover, and improving the visual contrast of dirty areas. For ultraviolet images, background noise suppression processing is performed, using temporal median filtering to eliminate random thermal noise, and adaptive threshold segmentation to enhance corona discharge hotspot areas, thereby improving the signal-to-noise ratio.

[0033] All images undergo geometric correction using pre-calibrated camera intrinsic matrix (including focal length and principal point coordinates) and lens distortion coefficients (including radial distortion k1, k2, k3 and tangential distortion p1, p2) to eliminate barrel or pincushion distortion. The corrected images are then mapped to pixel coordinates and point cloud world coordinates using known sensor extrinsic parameters (i.e., rotation matrices and translation vectors from the camera coordinate system to the world coordinate system). This mapping is achieved through backprojection: for each vertex on the surface of the 3D geometric model, its corresponding pixel position in visible and ultraviolet images is calculated, achieving pixel-level spatial alignment and forming a dual-channel texture mapping atlas that corresponds one-to-one with the model surface.

[0034] In step S5, the texture mapping atlas is projected onto the surface of the three-dimensional geometric model to construct a joint characterization model of dirt that integrates geometric structure and optical reflectance properties. Specifically, two attribute values ​​are bound to each triangular facet of the three-dimensional geometric model: one is the normalized reflectance value from the visible light image, and the other is the radiance value from the ultraviolet image. A dirty area is defined as a surface region that simultaneously meets the following two conditions: visible light reflectance is less than a preset threshold of 0.35, indicating the presence of light-absorbing dirt deposition on the surface; ultraviolet radiance is greater than three times the standard deviation of the background noise mean, indicating the presence of localized corona discharge activity caused by dirt. This dual criterion effectively distinguishes real dirt from interfering factors such as dust, water stains, or structural shadows.

[0035] For areas identified as contaminated, their geometric characteristics are further analyzed: the rate of change of the normal vector of the point cloud surface is calculated; if the angle between the normal vectors of adjacent points is greater than 15 degrees, it is marked as a region of geometric abrupt change; simultaneously, the Gaussian curvature and the average curvature gradient are calculated; if the absolute value of the gradient is greater than 0.02 mm... -1 If the surface protrusions or depressions are due to dirt accumulation, they are determined to be surface bumps or depressions caused by dirt accumulation. The above-mentioned geometric-optical joint criteria together constitute the core elements of the dirt joint characterization model.

[0036] In step S6, based on the pollution joint characterization model, the pollution layer thickness, volume density, and spatial continuity index of each region of the insulator are calculated. The pollution layer thickness is obtained by comparing the Euclidean distance between the current point cloud surface and the clean state reference model. The reference model is stored in the pollution reference database built into the pollution quantification unit, containing three-dimensional geometric models of the same type of insulator in factory condition, post-rain cleaning condition, and historical cleaning records. The system automatically selects the closest reference model as a reference based on the most recent cleaning time.

[0037] For each vertex on the model surface, calculate the shortest distance to the corresponding position on the baseline model; this distance represents the local contamination layer thickness. Bulk density is defined as the average thickness of the contamination layer per unit surface area, calculated by integrating for each insulator skirt section (e.g., upper skirt, lower skirt, and steel foot area). The formula is as follows: ; For bulk density, The area is the surface area of ​​the region. It is a filthy area. For point The thickness of the contamination layer at a given location. Spatial continuity is determined by calculating the number of connected components in the contaminated region and the area ratio of the largest connected region. The eight-neighbor connectivity criterion is used, treating the vertices of the surface triangular mesh as graph nodes; if the distance between two points is less than 10 mm and both belong to the contaminated region, an edge connection is established. All connected components are traversed using a depth-first search, and the total number is counted. and the area of ​​the largest connected region Spatial continuity index Defined as: ; For the first Area of ​​each connected region; The spatial continuity index reflects the degree of concentration of pollution distribution. The closer the value is to 1, the more continuous the pollution is and the easier it is to form a conductive channel.

[0038] In step S7, based on the contamination layer thickness, bulk density, and spatial continuity indicators, an insulator cleaning priority score is generated, and a cleaning decision instruction is output. First, the three indicators are normalized so that their values ​​are mapped to the range of 0 to 1. Normalized thickness... Obtained by dividing by the maximum permissible dirt thickness (typically 2 mm); normalized bulk density. Obtained by dividing by the historical maximum observed value; spatial continuity index Use directly. Cleaning priority scoring function. Set as: ; The weighting distribution reflects that thickness has the greatest impact on insulation performance, followed by volume density. Highly continuous contamination, although dangerous, is easily identifiable and therefore assigned a lower weight. When the score is greater than 0.7, the insulator is marked as "urgent cleaning"; when the score is between 0.4 and 0.7, it is marked as "planned cleaning"; and when the score is less than 0.4, it is marked as "no cleaning required".

[0039] The cleaning decision instruction includes the insulator number, the location of the corresponding tower, a heat map of pollution distribution (using color gradations to represent thickness), a suggested cleaning method (such as high-pressure water washing, mechanical brushing, or chemical cleaning), and a predicted pollution growth trend (based on an exponential growth model fitted to historical data). The instruction is uploaded to the provincial transmission monitoring center via the power dispatch data network and simultaneously pushed to the mobile terminals of maintenance personnel, with a format conforming to the IEC61850 communication standard.

[0040] The online monitoring system for transmission lines combining point cloud and video includes a 3D point cloud acquisition unit, a multispectral video acquisition unit, a point cloud processing unit, a video processing unit, a joint modeling unit, a pollution quantification unit, and a decision output unit. The 3D point cloud acquisition unit contains at least one solid-state lidar, installed on a crossarm support 1.5 meters directly below the insulator string. Its scanning plane is perpendicular to the insulator axis, and power is supplied by a solar energy storage device on the tower.

[0041] The multispectral video acquisition unit includes a dual-channel imaging module. The visible light channel is equipped with a fixed-focus lens with a focal length of 50mm and an aperture of F2.8, while the ultraviolet channel features a quartz lens and a bandpass filter with a center wavelength of 260nm and a bandwidth of 20nm. The two channels are designed for confocal focusing. The point cloud processing unit runs on an embedded edge computing platform, featuring a quad-core ARM Cortex-A72 processor and a dedicated point cloud acceleration coprocessor. It has 16 gigabytes of memory, uses industrial-grade solid-state drives for storage, and runs a customized Linux real-time kernel, ensuring a point cloud processing latency of less than 10 seconds.

[0042] When performing texture-geometry fusion, the joint modeling unit uses barycentric coordinate interpolation to map image pixel attributes to triangular mesh vertices and performs surface rendering using the Phong lighting model to verify the visual consistency of polluted areas under different lighting angles. The pollution quantification unit has a built-in pollution benchmark database that supports storing benchmark models categorized by insulator type, batch, and installation year, and features an automatic update mechanism. The decision output unit supports multi-level alarm tiered push notifications to ensure timely operation and maintenance response.

[0043] This embodiment achieves a precise three-dimensional assessment of the pollution status of transmission line insulators through strictly synchronized multi-source sensing, high-fidelity geometric reconstruction, pixel-level texture mapping, and physically interpretable quantization models. It fundamentally solves the technical bottleneck of traditional two-dimensional image methods being unable to quantify the thickness and spatial distribution of pollution, providing reliable data support and decision-making basis for smart grid operation and maintenance.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for online monitoring of transmission lines combining point cloud and video, characterized in that, include: Three-dimensional point cloud data of insulator strings and their fittings are acquired by lidar sensors deployed on transmission towers; Multispectral video frame sequences at corresponding viewpoints are acquired by synchronously triggered visible light and ultraviolet imaging sensors; The three-dimensional point cloud data is denoised, registered, and surface reconstructed to generate a high-fidelity three-dimensional geometric model of the insulator. The multispectral video frame sequence is subjected to image enhancement, distortion correction and spatiotemporal alignment processing to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model; The texture mapping atlas is projected onto the surface of the three-dimensional geometric model to construct a joint dirt characterization model that integrates geometric structure and optical reflection properties; Based on the aforementioned pollution joint characterization model, the pollution layer thickness, volume density, and spatial continuity index of each region of the insulator are calculated. Based on the contamination layer thickness, bulk density, and spatial continuity indicators, an insulator cleaning priority score is generated, and a cleaning decision instruction is output. Based on the aforementioned pollution joint characterization model, the pollution layer thickness, bulk density, and spatial continuity index of each region of the insulator are calculated, including: The thickness of the contamination layer is obtained by comparing the Euclidean distance between the current point cloud surface and the clean state benchmark model, which is trained using point cloud data from historical periods without contamination. Bulk density is defined as the average thickness of the dirt layer per unit surface area, calculated by integral calculation according to the umbrella skirt section; The spatial continuity index is determined by calculating the number of connected components in a polluted area and the proportion of the area of ​​the largest connected domain. Based on the contamination layer thickness, bulk density, and spatial continuity indicators, an insulator cleaning priority score is generated, and a cleaning decision instruction is output, including: Set cleaning priority scoring function for: ； To normalize the thickness, To normalize the volume density, As an indicator of spatial continuity; When the score is greater than 0.7, the insulator is marked as "emergency cleaning"; when the score is between 0.4 and 0.7, it is marked as "planned cleaning"; when the score is less than 0.4, it is marked as "cleaning not required". The cleaning decision instruction includes the insulator number, the location of the tower to which it belongs, a heat map of the pollution distribution, the recommended cleaning method, and a forecast of the expected pollution growth trend.

2. The online monitoring method for transmission lines combining point cloud and video according to claim 1, characterized in that, The three-dimensional point cloud data is subjected to denoising, registration, and surface reconstruction processing to generate a high-fidelity three-dimensional geometric model of the insulator, including: A statistical outlier removal algorithm is used to eliminate outliers caused by atmospheric disturbances or birds. Initial coarse registration of multi-view point cloud fragments is performed using Feature Descriptor FPFH, followed by registration using the Iterative Closest Point Algorithm. The Poisson surface reconstruction algorithm is used to generate a closed and watertight triangular mesh model, preserving the geometric details at the connection between the umbrella skirt edge and the steel foot.

3. The online monitoring method for transmission lines combining point cloud and video according to claim 2, characterized in that, The multispectral video frame sequence is subjected to image enhancement, distortion correction, and spatiotemporal alignment processing to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model, including: Perform illumination equalization processing on visible light images based on Retinex theory to suppress local overexposure or underexposure caused by backlighting or strong glare. Background noise suppression and hotspot enhancement are performed on ultraviolet images to improve the signal-to-noise ratio of weak corona discharge signals; Radial and tangential distortion corrections are performed on all images using a pre-calibrated camera intrinsic parameter matrix and lens distortion coefficients; By using known sensor extrinsic parameters, a one-to-one mapping relationship is established between the calibrated image pixel coordinate system and the point cloud world coordinate system, achieving pixel-level spatial alignment.

4. The online monitoring method for transmission lines combining point cloud and video according to claim 3, characterized in that, The texture mapping atlas is projected onto the surface of the three-dimensional geometric model to construct a joint dirt characterization model that integrates geometric structure and optical reflection properties, including: On each triangular facet of the three-dimensional geometric model, the corresponding visible light reflectance value and ultraviolet radiation intensity value are bound together. The area where dirt exists is defined as a surface area that simultaneously meets the following two conditions: the visible light reflectance is less than a preset threshold, and the ultraviolet radiation intensity is greater than three times the standard deviation of the mean background noise. For areas identified as contaminated, the geometric deformation characteristics caused by contamination accumulation are further identified based on the rate of change of the point cloud normal vector and the surface curvature gradient.

5. The online monitoring method for transmission lines combining point cloud and video according to claim 4, characterized in that, The cleanliness benchmark model is stored in the contamination benchmark database, which contains three-dimensional geometric models of insulators of the same type in their factory condition, post-rain cleanliness condition, and historical cleaning records, and is used to dynamically update the calculation benchmark for the current contamination thickness.

6. The online monitoring method for transmission lines combining point cloud and video according to claim 5, characterized in that, The heat map of dirt distribution uses color gradations to represent the thickness of the dirt layer. The recommended cleaning methods include high-pressure water rinsing, mechanical scrubbing, or chemical cleaning. The predicted dirt growth trend is generated based on an exponential growth model fitted to historical dirt data.

7. An online monitoring system for power transmission lines combining point cloud and video, characterized in that, The online monitoring method for transmission lines based on the point cloud and video combination according to any one of claims 1 to 6 is characterized by comprising: The three-dimensional point cloud acquisition unit is used to acquire three-dimensional point cloud data of insulator strings and their fittings through lidar sensors deployed on transmission towers; The multispectral video acquisition unit is used to acquire multispectral video frame sequences from corresponding viewpoints through synchronously triggered visible light and ultraviolet imaging sensors; The point cloud processing unit is used to perform denoising, registration and surface reconstruction processing on the three-dimensional point cloud data to generate a high-fidelity three-dimensional geometric model of the insulator. The video processing unit is used to perform image enhancement, distortion correction and spatiotemporal alignment processing on the multispectral video frame sequence to generate a texture mapping atlas that strictly corresponds to the three-dimensional geometric model; A joint modeling unit is used to project the texture mapping atlas onto the surface of the three-dimensional geometric model to construct a joint characterization model of dirt that integrates geometric structure and optical reflection properties; The pollution quantification unit is used to calculate the pollution layer thickness, volume density, and spatial continuity index of each region of the insulator based on the pollution joint characterization model. The decision output unit is used to generate an insulator cleaning priority score based on the contamination layer thickness, volume density, and spatial continuity index, and output a cleaning decision instruction.

8. The online monitoring system for transmission lines combining point cloud and video according to claim 7, characterized in that, When the joint modeling unit performs texture-geometry fusion, it uses the centroid coordinate interpolation method to map the image pixel attributes to the vertices of the triangular mesh and performs surface rendering through the Phong lighting model to verify the visual consistency of the dirty area. The decision output unit uploads the cleaning decision command to the provincial power transmission monitoring center through the power dispatch data network and pushes it to the mobile terminal of the operation and maintenance personnel simultaneously. The command format conforms to the IEC61850 communication standard.

Images