A power transmission line icing thickness self-calibration measurement method, device and system

Abstract

The application discloses a kind of transmission line icing thickness self-calibration measurement method, device and system, the method establishes bare conductor three-dimensional reference axis under no icing state, and multiple frames of images containing to-be-measured conductor and calibration reference are continuously collected when measuring, current camera external parameter is solved frame by frame using calibration reference feature point, and current three-dimensional reference axis and multiple local measurement sections are determined in combination with conductor spatial geometric relationship;With bare conductor reference axis as measurement center, extract two sides icing outer boundary candidate points at each local measurement section, convert pixel scale into spatial physical scale on corresponding local measurement section, and cross-frame time sequence persistence screening is carried out to multiple frames of candidate boundary to eliminate transient false boundary;According to the spatial measurement result of effective icing outer boundary, the composite diameter of icing conductor is calculated, and then apparent radial icing thickness and eccentric icing two-side thickness are determined.The application effectively overcomes the influence of camera external parameter drift, calibration reference and conductor not coplanar, eccentric icing axis offset and transient interference in ice and snow environment on the accuracy of thickness measurement.

Description

Technical Field

[0001] This invention relates to the field of power transmission line condition monitoring technology, specifically to a method, device, and system for self-calibrating measurement of icing thickness on power transmission lines. Background Technology

[0002] Icing on transmission lines increases the mechanical load on conductors and the risk of wind deflection, which can lead to faults such as line breakage, tower collapse, and decreased insulation performance in severe cases. To reduce the risks of manual inspection, existing monitoring solutions typically use cameras to acquire images of the conductors, identify icing areas using edge detection, image segmentation, or deep learning methods, and estimate the icing thickness based on the pixel width of the iced conductor in the image. For example, patent application CN112365475A discloses a transmission line icing monitoring method and system based on image processing, which obtains the icing boundary and calculates the icing thickness based on geometric relationships; patent CN113569804B discloses a transmission line icing monitoring system and risk alarm method based on image recognition, which provides dimensional indications in multiple thickness measurement directions through a measurement indicator device. These technologies provide a foundation for icing image measurement.

[0003] However, transmission line monitoring devices are installed outdoors for extended periods. The image acquisition unit is affected by wind vibration, micro-displacement of the support, temperature difference deformation, and maintenance adjustments, and its external parameters may change relative to the initial calibration state. At the same time, the calibration reference object and the conductor under test are often at different spatial depths. If the diameter of the conductor is directly converted using the pixel scale on the calibration target plane, a systematic scale error will be generated that is amplified with the shooting distance. On the other hand, under eccentric icing conditions, the geometric center or skeleton of the icing outer contour is not the same as the true axis of the bare conductor. If the measurement center is directly established based on the icing contour skeleton, it may cause distortion of the icing thickness on both sides, or even negative thickness or misjudgment of the eccentric direction.

[0004] Furthermore, rapidly moving snowflakes during snowfall, scattered ice crystals, specular reflections caused by nighttime lighting, fog, and localized obstructions can all create transient false boundaries. Relying solely on a single frame image for boundary determination and thickness calculation makes it difficult to guarantee the stability and evaluability of measurement results under unattended conditions. Summary of the Invention

[0005] The purpose of this invention is to provide a self-calibration measurement method, device, and system for icing thickness of transmission lines, in order to solve the problems of scale drift, boundary misdetection, and inability to automatically identify low-quality results caused by camera extrinsic drift, non-coplanarity of calibration reference and conductor, offset of eccentric icing axis, and transient interference from snowflakes and reflections in existing image thickness measurement.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A self-calibration measurement method for ice thickness on transmission lines includes the following steps: S1: Establishing the bare conductor reference: When the conductor to be measured is in an ice-free state, acquire the bare conductor reference image, extract the two sides of the bare conductor in the bare conductor reference image, and determine the two-dimensional reference axis of the bare conductor based on the two sides of the bare conductor; combine the pre-established spatial geometric relationship of the conductor, the camera intrinsic parameters, and the initialized camera extrinsic parameters determined based on the calibration reference feature points in the bare conductor reference image, restore the two-dimensional reference axis of the bare conductor to the calibration reference coordinate system, obtain the three-dimensional reference axis of the bare conductor, and store the nominal diameter of the bare conductor; S2: Acquire the current image and determine the current camera extrinsic parameters: The image acquisition unit continuously acquires multiple frames of current images that simultaneously contain the area of ​​the conductor to be measured and the calibration reference area. The calibration reference unit is set separately from the image acquisition unit and is fixed on the stable component of the transmission tower. The calibration reference unit is equipped with multiple reference feature points with known coordinates in the calibration reference coordinate system. For each frame of the current image, the current camera extrinsic parameters corresponding to the current image are determined based on the image coordinates of the reference feature points in the current image, the known coordinates of the reference feature points in the calibration reference coordinate system, and the camera intrinsic parameters corresponding to the current imaging state. S3: Determine the current 3D reference axis and local measurement cross-section: Based on the current camera extrinsic parameters, the 3D reference axis of the bare conductor, the spatial geometric relationship of the conductor, the projection position of the visible conductor segment in the current image, and the preset conductor sag range and swing range, determine the 3D reference axis of the conductor to be measured in the current state, and project the 3D reference axis onto the current image according to the current camera extrinsic parameters to obtain the bare conductor reference axis of the current image; determine multiple local measurement positions along the 3D reference axis, and determine the corresponding local measurement cross-section based on the 3D axis point and 3D local tangent vector at each local measurement position; S4: Determine the outer boundary of the ice accretion and perform spatial scale compensation: At each local measurement location, establish a two-dimensional boundary search line corresponding to the corresponding local measurement section based on the reference axis of the bare conductor in the current image. Determine candidate points of the outer boundary of the ice accretion on both sides of the reference axis of the bare conductor in the current image. Based on the current camera extrinsic parameters and the spatial geometric information of the corresponding local measurement section, convert the pixel scale of the candidate points of the outer boundary of the ice accretion in the current image into the physical scale on the corresponding local measurement section to obtain the spatial measurement results of the candidate outer boundary of the ice accretion. S5: Perform multi-frame temporal persistence screening: perform cross-frame association on candidate points of the outer ice boundary in multiple frames of the current image, determine the temporal persistence of the candidate boundary based on the number of frames of persistence of the candidate point of the outer ice boundary relative to the reference axis of the bare conductor in the current image, the stability of the normal distance, and the inter-frame motion characteristics, and remove transient candidate boundaries that do not meet the preset temporal persistence conditions to obtain the effective outer ice boundary. S6: Determine the icing thickness: Determine the composite diameter of the icing conductor at each effective local measurement position based on the spatial measurement results corresponding to the effective icing outer boundary, and determine the apparent radial icing thickness based on the composite diameter of the icing conductor and the nominal diameter of the bare conductor; For eccentric icing, use the three-dimensional reference axis obtained after the current state correction of the three-dimensional reference axis of the bare conductor as the reference axis of the bare conductor. Based on the spatial measurement results corresponding to the effective icing outer boundaries on both sides, determine the spatial radial distance from the three-dimensional axis point to the effective icing outer boundaries on both sides at each effective local measurement position. Determine the icing thickness on both sides based on the spatial radial distance on both sides and the nominal diameter of the bare conductor, and determine the eccentricity direction and degree based on the difference in icing thickness on both sides. S7: Perform measurement confidence evaluation and targeted re-acquisition: Determine the measurement confidence based on calibration quality, boundary positioning quality, temporal consistency, occlusion quality, and image quality; when the measurement confidence reaches a preset confidence threshold, output the icing thickness measurement result; when the measurement confidence is lower than the preset confidence threshold, determine the dominant quality parameter that caused the decrease in measurement confidence, generate the corresponding re-acquisition control signal based on the dominant quality parameter, and re-execute the corresponding image acquisition or measurement processing based on the re-acquisition control signal.

[0007] Furthermore, in step S2, the calibration reference unit is a planar calibration target, which is provided with at least four non-collinear reference feature points, orientation codes, and scale markers with known actual lengths. After distortion correction of the current image, the correspondence between the image coordinates of the reference feature points and their known coordinates in the calibration reference coordinate system is established. The random sampling consistency method is used to eliminate erroneous corresponding points, and the PnP algorithm is used to determine the current camera extrinsic parameters corresponding to each frame of the current image. When the image acquisition unit uses a variable focal length lens, the camera intrinsic parameters and lens distortion parameters corresponding to each focal length state are pre-stored, and the corresponding camera intrinsic parameters and lens distortion parameters are called according to the focal length state corresponding to the current image.

[0008] Further, in step S1, edge points on both sides of the bare conductor are extracted from multiple measurement positions in the bare conductor reference image. The midpoint of the edge points on both sides at the same measurement position is determined as the center position of the bare conductor. Curve fitting is performed on multiple bare conductor center positions to obtain the two-dimensional reference axis of the bare conductor. The spatial linear relationship or catenary relationship of the conductor relative to the calibration reference coordinate system is established using the line design parameters, conductor suspension point positions, or field measurement results. Combined with the initial camera extrinsic parameters, the center positions of multiple bare conductors are restored to the calibration reference coordinate system to obtain the three-dimensional reference axis of the bare conductor. In step S3, based on the projection positions of the visible continuous points of the conductor in the current image, the three-dimensional reference axis of the bare conductor is corrected in the current state within the sag range and swing range of the conductor to obtain the three-dimensional reference axis in the current state.

[0009] Further, in step S3, for each local measurement position, a three-dimensional axis point on the three-dimensional reference axis and a three-dimensional local tangent vector at that three-dimensional axis point are determined, and a spatial plane passing through the three-dimensional axis point and perpendicular to the three-dimensional local tangent vector is determined as the corresponding local measurement section; in step S4, an image measurement position corresponding to each local measurement position is determined on the current image bare conductor reference axis, a two-dimensional normal vector is determined based on the two-dimensional local tangent vector of the current image bare conductor reference axis at the image measurement position, and a two-dimensional boundary search line passing through the current image bare conductor reference axis is constructed along the two-dimensional normal vector; on the two-dimensional boundary search line, the comprehensive cost of the candidate boundary is determined based on the image gradient, the probability of icing foreground, the boundary continuity of adjacent image measurement positions, and the distance from the candidate boundary to the current image bare conductor reference axis, and a pair of candidate points with the smallest comprehensive cost of the candidate boundary are selected as the icing outer boundary candidate points located on both sides of the current image bare conductor reference axis.

[0010] Furthermore, the three-dimensional axis point at the i-th local measurement location is denoted as... A i , three-dimensional axis points A i The three-dimensional local tangent vector at point is denoted as t i The corresponding local measurement section is denoted as P i The local measurement section P i satisfy: (X- A i ) T t i =0 Where X represents any spatial point on the local measurement section, in step S4, the candidate points of the outer boundary of the ice covering on both sides of the i-th local measurement position are back-projected into spatial rays according to the camera intrinsic parameters and the current camera extrinsic parameters, and the spatial rays are aligned with the local measurement section. P i The two intersecting points are denoted as follows: B i + and B i - In step S6, the Euclidean distance between the two three-dimensional boundary points is denoted as the composite diameter of the icing conductor. d i ,according to d i =‖ B i + - Bi - || Determine the composite diameter of the icing conductor d i The nominal diameter of the bare conductor is denoted as D 0, and according to h i =( d i - D 0) / 2 Determine the apparent radial icing thickness at the i-th local measurement location. h i .

[0011] Furthermore, in step S4, under the conditions that the distortion correction of the current image has been completed, the angle between the normal vector of the calibration reference plane and the normal vector of the i-th local measurement section is not greater than a preset angle threshold, and the i-th local measurement position is within the pre-calibrated effective field of view, the spatial scale compensation is performed using a simplified depth ratio calculation method, and the known actual length on the calibration reference is denoted as... L ref The pixel length of the known actual length in the corrected current image is denoted as... l ref The depth of the calibration reference plane relative to the image acquisition unit is denoted as... Z t Let the depth of the i-th local measurement section relative to the image acquisition unit be denoted as . Z w,i Let the pixel distance between the outer ice-covered boundary points on both sides of the i-th local measurement position be denoted as . d p,i and in accordance with d i =( Z w,i / Z t )×( L ref / l ref )× d p,i Determine the composite diameter of the icing conductor at the i-th local measurement location. d i .

[0012] Furthermore, in step S5, the number of consecutively acquired frames of the current image is denoted as M, and the threshold for the continuous frame count of the candidate boundary points is denoted as... T cont , where 1≤ Tcont ≤M; Perform cross-frame association on candidate points of the outer boundary of ice in the current image of M frames, determine the normal distance variance of the candidate boundary points relative to the reference axis of the bare conductor in the current image and the inter-frame motion vector of the candidate boundary points, and determine the candidate boundary points with a continuous number of frames less than the continuous frame number threshold, a normal distance variance greater than a preset distance variance threshold, or inter-frame motion characteristics that meet the preset transient target conditions as transient candidate boundaries and remove them. The transient candidate boundaries include pseudo boundaries formed by snowflakes, scattered ice crystals or supplementary light reflection.

[0013] Furthermore, in step S7, the calibration quality is determined based on the reprojection error of the current reference feature points and the number of effective reference feature points, and the calibration quality is denoted as... q ref The boundary positioning quality is determined based on the boundary gradient intensity of the outer icing boundary, the icing foreground probability, and the boundary continuity of adjacent measurement locations. This boundary positioning quality is denoted as... q edge Timing consistency is determined based on the multi-frame thickness variance and the number of boundary persistence frames, and the timing consistency is denoted as... q time The occlusion mass is determined based on the proportion of the effective local measurement section, and the occlusion mass is denoted as... q occ Image quality is determined based on high-frequency energy, motion blur index, and saturation pixel ratio, and this image quality is denoted as... q img Normalize each quality parameter to 0 to 1, and follow the... Q= β 1· q ref + β 2· q edge + β 3· q time + β 4. q occ + β 5. q img Determine the measurement confidence level Q, where, β 1. β 2. β 3. β 4 and β All 5 are non-negative weights, and satisfy the following conditions: β 1+ β 2+ β 3+ β 4+ β 5 = 1; When the measurement confidence level Q is lower than the preset confidence threshold, according to βk·(1- q k Calculate the weighted quality missing amount corresponding to the k-th quality parameter, and determine the quality parameter with the largest weighted quality missing amount as the dominant quality parameter; when the dominant quality parameter is calibration quality, expand the calibration benchmark search range and redetermine the current camera extrinsic parameters; when the dominant quality parameter is boundary positioning quality or image quality, adjust the exposure time, gain, focus position, or fill light state; when the dominant quality parameter is temporal consistency, increase the number of consecutive acquisition frames or shorten the frame interval; when the dominant quality parameter is occlusion quality, delay re-acquisition; when the measurement confidence Q is still lower than the preset confidence threshold after reaching the preset re-acquisition limit, output a measurement invalidation flag and the corresponding low quality reason information.

[0014] Another objective of this invention is to provide a self-calibrating field measurement device for icing thickness of transmission lines. This device includes an image acquisition unit, a processing unit, a memory, a control unit, and a communication unit. The image acquisition unit acquires a current image that simultaneously includes the area of ​​the conductor to be measured and a calibration reference area. The calibration reference area is provided by a calibration reference unit, which is separately installed from the image acquisition unit and fixed to a stabilizing component of the transmission tower. The calibration reference unit has reference feature points with known coordinates in the calibration reference coordinate system. The processing unit is connected to the image acquisition unit and receives the current image output by the image acquisition unit. It then determines the current camera extrinsic parameters based on the image coordinates of the reference feature points in the current image and the known coordinates in the calibration reference coordinate system. The memory stores camera intrinsic parameters, lens distortion parameters, and calibration parameters. The system includes the known coordinates of reference feature points in the calibration reference coordinate system, the nominal diameter of the bare conductor, the two-dimensional reference axis of the bare conductor, the three-dimensional reference axis of the bare conductor, and the spatial geometric relationship of the conductor. The processing unit is also used to determine the three-dimensional reference axis and local measurement section of the conductor under test in the current state based on the current camera extrinsic parameters, the three-dimensional reference axis of the bare conductor, the spatial geometric relationship of the conductor, and the projection position of the visible conductor segment in the current image. The three-dimensional reference axis is used as the measurement spatial reference for icing outer boundary positioning, spatial scale compensation, multi-frame temporal continuity screening, icing thickness calculation, and measurement confidence evaluation. The control unit is used to output the corresponding re-acquisition control signal according to the dominant quality parameter that causes the measurement confidence to decrease when the measurement confidence is lower than the preset confidence threshold. The communication unit is used to interact with the remote processing platform.

[0015] Another objective of this invention is to provide a self-calibration measurement system for icing thickness of transmission lines. This system includes a field measurement device, a calibration reference unit, and a remote processing platform. The field measurement device includes an image acquisition unit, a processing unit, a memory, a control unit, and a communication unit. The image acquisition unit acquires current images of the area to be measured and the calibration reference area. The processing unit, connected to the image acquisition unit, receives the current images output by the image acquisition unit and determines the current camera extrinsic parameters based on reference feature points in the calibration reference area. It also determines the three-dimensional reference axis of the conductor under test in the current state based on the current camera extrinsic parameters, a pre-established three-dimensional reference axis of the bare conductor, and the spatial geometric relationship of the conductor. Using the three-dimensional reference axis as the measurement spatial reference, it performs local measurement section determination, icing outer boundary positioning, spatial scale compensation, multi-frame temporal continuity filtering, icing thickness calculation, and measurement confidence evaluation. The control unit outputs a re-acquisition control signal based on the measurement confidence level. The calibration reference unit... The calibration reference unit is fixedly installed on a stabilizing component of the transmission tower and is separated from the image acquisition unit. The calibration reference unit is a passive optical calibration target. The calibration reference unit and the conductor under test are located within the common effective field of view of the image acquisition unit. The reference feature points set on the calibration reference unit are imaged by the image acquisition unit and input to the processing unit. The processing unit determines the current camera extrinsic parameters based on the image coordinates of the reference feature points and the known coordinates in the calibration reference coordinate system. The calibration reference unit is equipped with multiple reference feature points with known coordinates in the calibration reference coordinate system. The remote processing platform is connected to the field measurement device through a communication unit. The remote processing platform includes a result storage module and a verification service module. The result storage module is used to store the ice thickness measurement results according to the device identification and acquisition time. The verification service module is used to verify the ice thickness measurement results based on the ice thickness distribution changes at adjacent measurement times and the measurement confidence level, and to issue a re-collection command to the field measurement device when the preset re-collection conditions are met.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention separates the calibration reference from the image acquisition unit and uses the feature points of the calibration reference to solve the current camera extrinsic parameters frame by frame. This enables the geometric correction to reflect the extrinsic parameter drift caused by wind vibration, temperature difference deformation and maintenance adjustment of the camera support during long-term outdoor operation in real time, overcoming the problem of scale drift over time introduced by the traditional solution that relies on one-time calibration parameters. 2. This invention does not directly use the pixel scale of the calibration target plane for conductor thickness measurement. Instead, it estimates the spatial depth of the conductor cross-section and compensates for the pixel scale based on the depth ratio. This effectively eliminates the systematic scale error caused by the difference in shooting distance, which is magnified with distance. It avoids the measurement deviation caused by directly applying the pixel scale of the nearby calibration target to the distant conductor.

[0017] 3. This invention uses the three-dimensional reference axis of the bare conductor established in the un-iced state as the spatial reference of the measurement center, and projects it onto the current image in combination with the current external parameters and conductor state, so as to ensure that the position of the bare conductor axis can still be accurately located when the icing profile grows unevenly, thereby eliminating the thickness distortion on both sides and the misjudgment of the eccentric direction caused by replacing the conductor center with the icing profile skeleton or centroid.

[0018] 4. This invention introduces multi-frame temporal persistence constraints, performs cross-frame association on candidate boundaries in continuously acquired multi-frame images, and utilizes the difference between the positional stability of the real icing boundary relative to the reference axis of the conductor and the randomness of the motion of transient targets such as snowflakes, scattered ice crystals and supplementary light reflections to effectively eliminate transient false boundaries and significantly improve the stability and anti-interference ability of boundary positioning.

[0019] 5. This invention establishes a multi-dimensional measurement confidence evaluation system covering calibration quality, boundary positioning quality, temporal consistency, occlusion quality, and image quality. When the confidence is insufficient, it performs targeted re-acquisition based on the dominant low-quality factors, realizing automatic evaluation of measurement result quality and autonomous re-acquisition under low-quality conditions. This avoids directly outputting low-quality results as valid measurement values, providing a reliable guarantee for long-term stable operation under unattended conditions. Attached Figure Description

[0020] Figure 1 This is a flowchart of the self-calibration measurement method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the self-calibration measurement system according to an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the spatial scale recovery when the calibration reference and the conductor are not coplanar in an embodiment of the present invention. Figure 4 This is a schematic diagram of the current state correction and projection of the three-dimensional reference axis of the bare conductor, the reference axis of the bare conductor, and multiple local measurement sections in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the removal of transient snowflakes and reflections using multi-frame persistent constraints in an embodiment of the present invention. Figure 6 This is a schematic diagram of the measurement confidence evaluation and closed-loop re-sampling in an embodiment of the present invention; Figure 7 This is a schematic diagram of the on-site installation relationship according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figure 2 and Figure 7As shown, this invention provides a self-calibration measurement system for icing thickness of transmission lines, including a field measurement device 100. The field measurement device 100 includes an image acquisition unit 110, a processing unit 130, a memory 140, a control unit 150, and a communication unit 160. The system also includes a calibration reference unit 120 separated from the image acquisition unit 110 and a remote processing platform 200. The calibration reference unit 120 is a passive optical calibration target, fixedly installed on a stable component of the transmission tower crossarm or tower body, spatially separated from the image acquisition unit 110. The calibration reference unit 120 and the conductor under test are located within the common effective area of ​​the image acquisition unit 110. Within the field of view, the image of the calibration reference unit 120 is acquired by the image acquisition unit 110 and input to the processing unit 130. It serves as the content of the calibration reference area in the current image and participates in the solution of the current camera extrinsic parameters. The remote processing platform 200 is connected to the communication unit 160 via a wireless communication network such as 4G / 5G, and is used to receive measurement data uploaded by the field measurement device 100 and issue control commands. The remote processing platform 200 further includes a result storage module 210 and a verification service module 220. The image acquisition unit 110 is installed on the main structure of the tower or a dedicated installation platform. The image acquisition unit 110 is used to acquire images that simultaneously include the area of ​​the conductor to be measured. The current image of the domain and calibration reference area, wherein the calibration reference unit 120 is installed on a stable component of the crossarm or tower body of the transmission tower, and reference feature points with known coordinates in the calibration reference coordinate system are set on it. The calibration reference unit 120 is set separately from the image acquisition unit 110, so that when the image acquisition unit 110 undergoes support deflection or attitude change due to wind vibration, temperature difference deformation or maintenance adjustment, the projection of the calibration reference unit 120 in the current image can truly reflect the current pose of the camera relative to the tower reference. The effective field of view of the image acquisition unit 110 should simultaneously cover the calibration reference unit 120 and the conductor to be measured. The processing unit 130 is connected to the image acquisition unit 110. 0. The memory 140 and the control unit 150 are connected and used to perform image processing, camera extrinsic parameter solving, spatial scale compensation, icing boundary positioning, thickness calculation, and confidence evaluation. The memory 140 is used to store camera intrinsic parameters, lens distortion parameters, known coordinates of calibration reference feature points in the calibration reference coordinate system, nominal diameter of the conductor, two-dimensional reference axis of the bare conductor, three-dimensional reference axis of the bare conductor, and spatial geometric relationship of the conductor. The control unit 150 is used to output corresponding re-sampling control signals according to the dominant low-quality factors when the measurement confidence is lower than a preset threshold. The communication unit 160 is used to realize data interaction between the field measurement device 100 and the remote processing platform 200.

[0023] In a preferred embodiment, the calibration reference unit 120 is a planar calibration target, on which at least four non-collinear reference patterns, direction codes, and scale marks with known actual lengths are provided. The image acquisition unit 110 can be a visible light camera and is equipped with a supplementary lighting component and an imaging control circuit for adjusting exposure time, gain, supplementary lighting state, and focus position. A fixed-focus lens is preferred. When a variable-focus lens is used, the memory 140 pre-stores the camera intrinsic parameters and lens distortion parameters corresponding to each focal length state and calls the corresponding parameters according to the current focal length state. The processing unit 130 can be an embedded processor, a graphics processor, or a neural network processor.

[0024] During the initialization phase, a baseline image was first acquired in an ice-free state. I 0. Extract the edges of the bare conductor on both sides of the image, take the midpoint of the edge points on both sides at the same measurement section, and perform spline curve fitting on the midpoint of each section to obtain the two-dimensional reference axis of the bare conductor parameterized by the arc length s. C 0( s Simultaneously, using line design parameters, conductor suspension point locations, or field measurement results, the spatial relationship between the conductor and the calibration reference coordinate system is established as a straight line or catenary. L w Further based on the reference image I The calibration reference feature points in 0 are used to solve for the camera extrinsic parameters during initialization, and combined with the camera intrinsic parameters. C 0( s )and L w The center positions of each bare conductor are restored to the calibration reference coordinate system, resulting in the three-dimensional reference axis of the bare conductor parameterized according to the arc length s. C w ( s ), and will C 0( s ), C w ( s )and L w Stored in memory 140.

[0025] At measurement time t, the image acquisition unit 110 continuously acquires M frames of the current image. I t,mWhere M is preferably not less than 5, each frame of image records the corresponding exposure time, gain, focal length, fill light status and acquisition timestamp, etc. In order to adapt to complex lighting environments such as nighttime fill light and snowfall, different exposure parameters can be used for continuous acquisition, or the exposure and fill light status can be adjusted during subsequent low confidence re-acquisition. If a zoom lens is used, the corresponding camera intrinsic parameters and lens distortion parameters are selected according to the focal length status recorded in each frame. For ease of description, the frame subscript m will be omitted in the following text without causing ambiguity, and the corresponding quantity of the current frame will be used for description.

[0026] like Figure 3 As shown, for the m-th frame of the current image, the pixel coordinates of the current reference feature points are extracted from the calibration reference region of the current image. p j,t,m =( u j,t,m , v j,t,m After distortion correction of the image, a model is established based on image encoding, template matching, or local descriptors. p j,t,m The known coordinates of the calibration reference feature points in the calibration reference coordinate system P j The correspondence between (Xj, Yj, 0) is preferably determined by using the Random Sampling Consensus (RANSAC) method to eliminate erroneous correspondences, and then the camera extrinsic parameters for the current frame are solved based on the Perspective-n-Point (PnP) algorithm. R t,m | t t,m ].

[0027] Let the average reprojection error of the effective corresponding points be... e t The number of valid corresponding points is K t Then the calibrated quality q ref According to e t and K t As an optional implementation method, it can be adopted. q ref =exp(- e t / s e )·min(1, K t / K 0) Calculate the calibration mass, where s e The reprojection error scale parameter. K0 represents the expected number of effective feature points, which can also be determined using a piecewise function obtained from calibration experiments. e t and K t Normalization is performed, and the camera extrinsic parameters are used to establish the correction relationship between the current image coordinates and the reference coordinates. Specifically, when distinguishing frames... e t , K t and q ref Corresponding to the m-th frame respectively e t,m , K t,m and q ref,t,m .

[0028] To determine the three-dimensional reference axis in the current state, the spatial geometric relationship Lw of the conductor can be represented as a parameterized spatial curve determined by the reference geometric parameters, sag correction, and lateral sway. The continuous center points of the visible conductor are extracted from the current image. The candidate parameterized spatial curve is projected onto the current image based on the current camera intrinsic and extrinsic parameters. The sum of squared reprojection distances between the projected curve and the continuous center points of the visible conductor is used as the objective function. The current curve parameters are solved using the preset conductor sag range and sway range as parameter constraints to obtain the three-dimensional reference axis Cw,t(s) in the current state. When part of the conductor is occluded, only the visible conductor segment is used in the objective function calculation, and the occluded segment is constrained by the conductor hanging point position and the spatial straight line relationship or the continuity of the catenary.

[0029] This invention does not directly use the pixel scale after calibration of the target plane for guideline measurement. Instead, it further performs spatial scale compensation for the depth difference between the calibration target and the guideline. For the i-th local measurement section, the spatial geometric relationship of the guideline established in the initialization phase is used. L w Based on the projected position of the conductor in the current image and the allowable sag and swing range, estimate the spatial depth of the cross section in the camera coordinate system. Z w,i This depth can be jointly estimated based on the coordinates of the traverse suspension point, the catenary parameters, the projected positions of the continuous points of the traverse in the current image, and the current camera extrinsic parameters. Simultaneously, it can be estimated based on the current camera extrinsic parameters and the three-dimensional reference axis of the bare traverse. C w ( s ), Spatial geometric relationship of conductors L w And the projected positions of the continuous points of the visible conductor in the current image, within the allowable sag and swing range, determine the three-dimensional reference axis in the current state. Cw,t ( s For the i-th measurement position, determine the three-dimensional axis point on the three-dimensional reference axis. A i and three-dimensional local tangent vector t i and will go through A i And perpendicular to t i spatial plane P i Defined as the corresponding local measurement section, that is, the local measurement section satisfies the following formula: (X- A i ) T t i =0 Where X represents any spatial point on the local measurement section.

[0030] In a more precise implementation, the two outer boundary pixels detected on the measurement section are compared using the camera intrinsic parameter K and the current extrinsic parameter [...]. R t | t t The back projection is a spatial ray in the camera coordinate system, and the two spatial rays are respectively aligned with the corresponding local measurement sections. P i Intersection yields three-dimensional boundary points. B i + and B i - The composite diameter of the i-th cross section d i This is the Euclidean distance between two 3D boundary points, calculated as follows: d i =‖ B i + - B i - || When the distortion correction of the current image has been completed, the angle between the normal vector of the calibration reference plane and the normal vector of the i-th local measurement section of the traverse is not greater than a preset angle threshold, and the i-th local measurement position is within the pre-calibrated effective field of view, a simplified depth ratio calculation method can be used; the preset angle threshold and the effective field of view are predetermined based on the allowable scale error and the camera calibration results, assuming the known actual length on the calibration target is... L ref The pixel length in the corrected image isl ref The depth of the calibrated target plane is Z t The depth of the local measurement section of the i-th conductor is Z w,i The pixel distance between the two outer boundary points is d p,i The composite diameter can be calculated using the following formula: d i =( Z w,i / Z t )×( L ref / l ref )× d p,i Where Lref is the actual distance between the two endpoints P1 and P2 of the scale marker on the calibration target. l ref The pixel distance between the projection points p1 and p2 of P1 and P2 in the current calibrated image; d p,i The outer boundary points of the ice cover on both sides of the i-th local measurement location. B i + and B i - Projection point b in the current corrected image i + and b i - The pixel distance between them.

[0031] Under a set of typical installation parameters, the depth ratio of the calibration target plane to the guide wire is approximately 3.2:11.5. Experiments with a guide wire depth variation of ±15% show that the dimensional error generated by the above simplified formula is less than 0.3%, which can meet the accuracy requirements of engineering measurement.

[0032] Preprocessing operations such as brightness normalization, local contrast enhancement, and noise suppression are performed on the conductor area. For local overexposure areas caused by nighttime lighting, weighting can be performed based on the saturation pixel ratio and gradient direction. For low contrast issues caused by fog and snow, local histogram equalization or multi-scale illumination separation methods can be used to enhance the visibility of the conductor and ice-covered boundaries.

[0033] Identification of conductors and icing areas can be achieved through traditional edge detection combined with region growing algorithms, or through semantic segmentation models. In one embodiment, the semantic segmentation model adopts UN... et The training dataset contains 12,000 images of icing on transmission lines, covering various icing types such as rime, hoarfrost, mixed rime, and dry snow. It also includes different lighting and weather conditions, such as daytime lighting, nighttime supplementary lighting, and heavy snow background. Pixel-level annotations are divided into three categories: bare conductors, iced areas, and background.

[0034] To improve the accuracy of icing outer boundary localization, which is directly related to thickness measurement, the boundary neighborhood in the training loss is weighted. For pixel p, let d be the Euclidean distance to the nearest icing outer boundary pixel, then the weights can be set as follows: w(p) = 1 + 9·exp(- d 2 / 8) In the training and evaluation data of this embodiment, after adding the aforementioned boundary loss weighting, the average distance error of the model boundary decreased from 1.82 pixels to 0.97 pixels, and the average absolute error of icing thickness was further reduced by approximately 12%. It is understood that the above weighting form and network structure are only one optional implementation method and can also be replaced by other models that can output the probability of the outer contour of icing.

[0035] like Figure 4 As shown, after geometric correction is completed in the current frame, the three-dimensional reference axis of the bare conductor is determined based on the current camera extrinsic parameters and the data stored during the initialization phase. C w ( s And the projected position of the currently visible conductor segment, within the allowable sag and oscillation range, determine the three-dimensional reference axis in the current state. C w,t ( s ), and through the current camera projection relationship C w,t ( s Projecting this onto the current corrected image yields the reference axis of the bare conductor. C ref ( s For situations where the reference axis is partially invisible due to localized snow accumulation, frost, or component obstruction, the reference axis can be determined based on the adjacent visible sections. C ref ( s The curve is extended, and the reference axis position of the shaded section is obtained by combining the catenary constraint.

[0036] According to the preset arc length interval Δs exist C ref ( s Select N measurement locations on the screen. s i Fit the local tangent vector τ of the reference axis in the neighborhood of each location. s i), and determine the corresponding normal vector n( s i ), along the normal vector n( s i ) Direction construction through C ref ( s i Two-dimensional boundary search line l i And the boundary search line is compared with the three-dimensional local measurement section. P i Establish a correspondence. If the curvature at a certain measurement location exceeds a preset threshold, overlaps with the hardware area, or the occlusion ratio exceeds a threshold, then the local measurement section corresponding to that measurement location is marked as an invalid section and will not participate in subsequent thickness calculations.

[0037] Two-dimensional boundary search lines corresponding to each valid measurement location l i Extracting grayscale gradient g i ( l ) Icing Prospect Probability p i ( l The predicted boundary locations of adjacent sections are calculated, and candidate boundary sets are formed on both sides of the bare conductor reference axis. The comprehensive cost of the candidate boundaries can be expressed as: J i = w 1· J g + w 2· J p + w 3· J c + w 4. J d in, J g Cost for gradient direction and intensity J p As a cost of changes in foreground probability, J c To maintain continuity with adjacent cross-sectional boundaries, J d The reasonable cost of the distance from the candidate boundary to the reference axis of the bare conductor. w 1 to w 4 represents a non-negative weight, chosen to satisfy both directional and continuity constraints while minimizing the overall cost. J i The pair of candidate points that yield the minimum value are used as the outer boundary points. bi + and b i - .

[0038] The composite diameter is obtained for the i-th measurement section. d i Subsequently, apparent radial icing thickness h i Calculate according to the following formula: h i =( d i - D 0) / 2 The apparent radial icing thickness described in this invention is a geometric thickness calculated based on the composite diameter of the iced conductor and the nominal diameter of the bare conductor, determined by spatial measurements of the outer boundary of the iced conductor. It does not involve the conversion of icing density, mass, or shape factor.

[0039] For the case of eccentric icing, the three-dimensional axis point at the i-th local measurement location is used. A i As the center of the bare conductor axis, calculate the three-dimensional axis points respectively. A i To the two three-dimensional boundary points B i + and B i - Spatial radial distance r i + and r i - The thickness of the eccentric ice layer on both sides was obtained. h i + = r i + - D 0 / 2 and h i - = r i - - D 0 / 2, and with h i + and h i -The difference determines the direction and degree of eccentricity. Since the measurement center is obtained from the un-iced reference axis, and the centroid or skeleton of the eccentric icy profile is not used as the bare guide center, systematic errors caused by the skeleton shifting towards the thick ice side can be effectively avoided. When the icy thickness on either side calculated directly according to the above relationship is less than 0, the corresponding measurement section is marked as a boundary positioning anomaly, and the boundary positioning quality is reduced. q edge This section is not output as an effective thickness value and can trigger subsequent re-mining.

[0040] like Figure 5 As shown, after solving the current extrinsic parameters and geometric registration for the continuously acquired M frames of images, the candidate boundaries of each frame are correlated across frames. The actual icing boundary is fixed to the conductor as a whole and maintains a relatively stable position relative to the reference axis of the bare conductor in the continuous frames. The appearance time, movement direction and brightness changes of snowflakes, scattered ice crystals and reflective points have obvious transient characteristics.

[0041] In one embodiment, M is preferably not less than 5 frames, and the number of consecutive frames for candidate boundary points is a threshold. T cont Satisfying 1≤ T cont ≤M, preferred T cont No less than 5 frames; a continuous ratio threshold can also be set. or and demand T cont / M≥ or Candidate boundary points to C ref ( s i normal distance variance s d 2 Preferably, the distance is no greater than (1.5 pixels)². The inter-frame displacement vector v of the candidate boundary points is calculated using the KLT optical flow algorithm and decomposed into components along the conductor tangential direction τ and normal direction n. v τ , v n ,when‖ v n ||>2.0 pixels / frame and|| v τ ‖ / (‖ v n ‖+ e When the value is less than 0.3, it is determined to be a transient target that crosses laterally; when v n When the symbol alternately reverses within three consecutive frames, it is also identified as a transient target. Candidate points with local grayscale peaks and contrasts on both sides of the boundary that significantly exceed the typical distribution of ice-covered edges are also eliminated.

[0042] For the multiple retained section thickness values, the median can be calculated first. h m And the median absolute deviation MAD will satisfy | h i - h m |≤ α MAD+ e The measured cross section is included in the effective set V, and then the median, truncated mean or the mean weighted by the boundary positioning quality of each thickness value in the effective set is output. At the same time, the thickness sequence arranged along the arc length of the conductor is also output.

[0043] like Figure 6 As shown, the measurement confidence level Q is used to evaluate whether the current measurement result can be output as a valid value. Specifically, the calibration quality... q ref The boundary positioning quality is determined by the reprojection error of the current reference feature points and the number of effective feature points. q edge Determined by boundary gradient strength, foreground probability, and continuity of adjacent sections; temporal consistency. q time Determined by multi-frame thickness variance and boundary duration frames; occlusion quality q occ It can be determined by the effective cross-sectional ratio; image quality q img It can be determined by high-frequency energy, motion blur index, and saturation pixel ratio.

[0044] After normalizing each quality parameter to the 0-1 range, a weighted fusion method can be used to obtain the overall confidence level: Q= β 1· q ref + β 2· q edge + β 3· q time + β 4. q occ + β 5. q img ,in β 1 to β All 5 are non-negative weights and β 1+ β 2+ β 3+ β 4+ β 5=1 In an embodiment obtained by fitting measured data, the weights are taken as follows: β 1 = 0.25 β 2 = 0.30 β 3 = 0.20 β 4 = 0.15 β 5 = 0.10. It is understandable that the above weights can be retrained or calibrated according to different routes, cameras and working environments.

[0045] When Q is lower than the preset confidence threshold T Q At that time, targeted re-sampling is carried out based on the dominant quality parameter that caused the decrease in Q. The dominant quality parameter can be determined as the weighted quality missing quantity. β k ·(1- q k Maximum mass parameter: If q ref If the value is too low, expand the calibration benchmark search range and resolve the current extrinsic parameters; if q edge or q img If the light is too low, adjust the exposure, gain, focus position, or fill light settings; if... q time If the value is too low, increase the number of consecutive frames or shorten the frame interval; if... q occ If the value is too low, the measurement will be delayed until the obstruction or the conductor swing weakens and the data is collected again. If the threshold requirement is still not met after the preset maximum number of re-sampling attempts is reached, an invalid measurement flag and the corresponding reason information will be output.

[0046] In one practical installation embodiment, the calibration reference unit 120 is fixed to the stabilizing component of the transmission tower crossarm and is installed separately from the image acquisition unit 110. The image acquisition unit 110 is installed on the main structure of the tower, approximately 28m above the ground. The optical center of the camera is approximately 3.2m from the center of the calibration target plane. The calibration target is a 300mm × 300mm aluminum substrate. The actual distance between the centers of adjacent reference circles is... L ref =150mm, the conductor to be tested is LGJ-300 / 40 steel-cored aluminum stranded wire, nominal diameter D 0=23.94mm, the traverse measurement section is about 11.5m from the camera, the horizontal offset is about 2.8m, the camera pitch angle is about -9.5°, the yaw angle is about 0°, the camera focal length is 12mm, the pixel size is 3.45μm, and the image resolution is 1920×1080 pixels.

[0047] Under the above installation conditions, the calibration target and the traverse are significantly non-coplanar, with a depth ratio of approximately 3.2:11.5. By solving the current camera extrinsic parameters through the feature points of the calibration target in the current image, and performing spatial back projection or depth ratio compensation on the traverse measurement cross section, the physical scale of the traverse cross section can be re-established after the camera attitude changes.

[0048] To verify the accuracy of the icing thickness measurement method of the present invention, a custom-made polyethylene sleeve calibrated with a micrometer was set on a simulated transmission line experimental platform to simulate icing. The actual thicknesses were set to 0, 2.07, 5.12, 10.21 and 20.35 mm, respectively. The test conditions covered shooting distances of 8 m, 12 m and 16 m, pitch angles of -15°, -10°, 0° and +10°, yaw angles of -8°, 0° and +8°, ​​conductor swing of 0 and 30 mm (0.5 Hz), and meteorological conditions of no snow, moderate snow and heavy snow, as well as nighttime supplementary lighting on or off. The number of samples under each condition was no less than 50 frames. Some measurement results are shown in Table 1 below.

[0049] Table 1 Measurement results of icing thickness under complex working conditions

[0050] The test results above show that, under the listed complex ice and snow, conductor swing, and supplementary lighting conditions, the mean absolute error (MAE) for each thickness grade is less than 1 mm.

[0051] To further verify the impact of bare conductor reference axis transfer on measurement accuracy under eccentric icing conditions, five types of significantly eccentric icing conditions were selected, and multiple effective measurement position samples were obtained under each condition for comparative experiments. The measured eccentricity ratio, i.e., the ratio of the thickness of the thick side to the thickness of the thin side, ranged from 1.2 to 2.8, and the icing thickness ranged from 5 mm to 18 mm. The methods of directly using the icing contour skeleton as the centerline and the present invention were used respectively. C w ( s The result is obtained by correcting and projecting the current state. C ref ( s The baseline axis method was used to process the data, and the bare conductor axis marked manually was used as the true value for comparison. The results are shown in Table 2.

[0052] Table 2 Comparison Results of Bare Conductor Reference Axis Restoration

[0053] The results in Table 2 show that eccentric icing causes the centroid of the icing profile to shift towards the thicker ice side. When the icing profile skeleton is used as the center of the bare conductor, the axis positioning error and the thickness error on both sides are significantly increased. After adopting the current state correction and projection method of the three-dimensional reference axis of the present invention, the average axis deviation is reduced to 0.62 pixels and the accuracy of eccentricity direction judgment is increased to 96%.

[0054] In the single-factor comparison experiment with depth compensation, the apparent radial icing thickness was fixed at 10.21 mm, and the shooting distances were set to 8 m, 12 m, and 16 m, respectively. Other conditions remained unchanged: direct view, no snow, and no swaying. The target surface size conversion without depth compensation and the Zw / of the present invention were respectively used. Z t Depth compensation was calculated, and the results are shown in Table 3.

[0055] Table 3. Comparison of Depth Compensation Results at Different Shooting Distances

[0056] As shown in Table 3, without depth compensation, the error generated by directly using the target surface scale coefficient increases significantly as the depth difference between the guide wire and the calibration target increases; however, the error generated by using the Zw / of the present invention... Z t After depth compensation, the average absolute error within the range of 8m to 16m was controlled within 0.5mm, indicating that scale compensation can effectively suppress the systematic scale drift caused by the non-coplanarity of the calibration target and the guide wire.

[0057] In the dynamic camera attitude change experiment, the same conductor and an apparent radial icing thickness of 12.15 mm were used. The pitch and yaw angles of the camera were changed by a precision gimbal. The processing method of using only the initial fixed extrinsic parameters was compared with the processing method of resolving the current extrinsic parameters for each frame. The results are shown in Table 4.

[0058] Table 4 Comparison of dynamic camera pose changes

[0059] As shown in Table 4, the measurement error under fixed extrinsic parameters increases significantly with the increase of the camera pitch or yaw angle. However, after resolving the current camera extrinsic parameters using the field calibration reference for each frame, the average absolute error under the listed attitude change conditions is no greater than 0.61 mm, which effectively reduces the impact of support deflection and attitude changes on the measurement results during long-term camera operation.

[0060] To evaluate the contribution of three core technologies—non-coplanar scale compensation, bare conductor reference axis transfer, and multi-frame continuous screening—to the overall measurement performance, ablation experiments were conducted on the same test set. The test set consisted of a continuous acquisition sequence under complex ice and snow and swing conditions, containing a total of 1000 frames of images. Four configurations were set up: complete scheme, removal of depth compensation, removal of axis transfer, and removal of multi-frame screening. The results are shown in Table 5.

[0061] Table 5. Results of ablation experiments using core technologies

[0062] Table 5 shows that removing any core component will lead to an increase in measurement error or false detection rate. Specifically, after removing depth compensation, MAE increased from 0.52 mm to 1.68 mm; after removing the bare conductor reference axis transfer, the measurement error under eccentric icing conditions increased significantly; after removing multi-frame continuous screening, transient candidate boundaries formed by snowflakes, ice crystals and reflections are more likely to be mistakenly retained as true boundaries.

[0063] On the same test set, the complete scheme of this invention was further compared with a fixed pixel ratio scheme, a traditional camera calibration scheme that only initializes once, a single-frame processing scheme, and a scheme that does not perform confidence-based closed-loop resampling. The evaluation metrics included mean absolute error (MAE) and 95th percentile absolute error. E 95 The false detection rate of abnormal boundaries with a single cross-sectional thickness error greater than 5 mm is shown in Table 6.

[0064] Table 6. Comprehensive Comparison Results of Different Measurement Schemes

[0065] The comprehensive comparison results show that the present invention significantly reduces the mean absolute error, 95th percentile absolute error, and false detection rate of abnormal boundaries by comprehensively applying current camera extrinsic parameter updates, non-coplanar scale compensation, bare conductor reference axis transfer, multi-frame transient interference screening, and measurement confidence closed-loop control.

[0066] The measurement confidence level Q was calculated and correlated with the absolute error of thickness measurement for over 2000 test samples under different working conditions. Statistical results show that: when Q > 0.75, the proportion of samples with errors exceeding 1.5 mm is less than 1.8%; when Q is between 0.5 and 0.75, this proportion is approximately 9.5%; and when Q < 0.5, this proportion rises to over 36%. In one embodiment, the confidence threshold is... T Q Set to 0.65 to strike a balance between the efficiency of measurement results and the proportion of low-quality results being re-collected.

[0067] Using each sub-item quality parameter as the independent variable, and 1 / (e+ e e Let be the dependent variable, where e is the true measurement absolute error. e e To prevent the denominator from being zero, a positive number is used for multiple linear regression and normalization, which yields the following result: β 1 to β A set of weights of 5: β 1 = 0.25 β 2 = 0.30 β 3 = 0.20 β 4 = 0.15 β 5 = 0.10. When using this set of weights, the Spearman correlation coefficient between the measurement confidence level Q and the actual error is -0.83, indicating that there is a strong negative correlation between the confidence level and the measurement error, and the confidence level can effectively reflect the reliability of the measurement results.

[0068] In this embodiment, the threshold, weight, frame rate, cross-sectional interval, and network structure parameters can be calibrated or adjusted according to the field measurement device, conductor type, and line environment; among them, the number of consecutive acquisition frames, normal distance variance threshold, inter-frame motion criterion, and confidence threshold can all be determined based on actual measurement data.

[0069] After the measurement data is uploaded, the remote processing platform 200 saves the measurement results according to the device identifier and acquisition time through its internal result storage module 210 for historical trend analysis and cross-time period comparison. The verification service module 220 is used to compare the thickness distribution at adjacent times and issues a re-sampling instruction to the field measurement device 100 through the communication unit 160 when the following situations occur: the measurement result changes abruptly and is significantly inconsistent with the weather changes; the field device is in a state of low confidence for a continuous period of time; or there is an abnormal measurement section.

[0070] To reduce communication load, under normal conditions, the field measurement device 100 can only upload structured measurement results and thumbnails; when the confidence level is lower than the platform's review threshold, the thickness exceeds the value of interest, or the thickness growth rate is abnormal, the original image will be uploaded for review.

[0071] The method of the present invention can be implemented by software, hardware or a combination of software and hardware. When the computer program is executed by the processing unit 130, the field measurement device 100 completes all functions such as image acquisition, current camera extrinsic parameter solution, current state correction and projection of the three-dimensional reference axis of the bare conductor, local section construction, spatial scale compensation, boundary positioning, thickness calculation, multi-frame anomaly removal, confidence evaluation and re-sampling control.

[0072] The above description is only a preferred embodiment of the present invention. For those skilled in the art, any equivalent substitutions or conventional adjustments made to the calibration reference graphic form, camera model, boundary detection algorithm, depth estimation method, quality fusion method, and re-sampling control parameters without departing from the technical concept of the present invention shall fall within the protection scope of the present invention.

Claims

1. A self-calibration measurement method for icing thickness of transmission lines, characterized in that: Includes the following steps: S1: Establishing the bare conductor reference: When the conductor to be measured is in an ice-free state, acquire the bare conductor reference image, extract the two sides of the bare conductor in the bare conductor reference image, and determine the two-dimensional reference axis of the bare conductor based on the two sides of the bare conductor; combine the pre-established spatial geometric relationship of the conductor, the camera intrinsic parameters, and the initialized camera extrinsic parameters determined based on the calibration reference feature points in the bare conductor reference image, restore the two-dimensional reference axis of the bare conductor to the calibration reference coordinate system, obtain the three-dimensional reference axis of the bare conductor, and store the nominal diameter of the bare conductor; S2: Acquire the current image and determine the current camera extrinsic parameters: The image acquisition unit continuously acquires multiple frames of current images that simultaneously contain the area of ​​the conductor to be measured and the calibration reference area. The calibration reference unit is set separately from the image acquisition unit and is fixed on the stable component of the transmission tower. The calibration reference unit is equipped with multiple reference feature points with known coordinates in the calibration reference coordinate system. For each frame of the current image, the current camera extrinsic parameters corresponding to the current image are determined based on the image coordinates of the reference feature points in the current image, the known coordinates of the reference feature points in the calibration reference coordinate system, and the camera intrinsic parameters corresponding to the current imaging state. S3: Determine the current 3D reference axis and local measurement cross-section: Based on the current camera extrinsic parameters, the 3D reference axis of the bare conductor, the spatial geometric relationship of the conductor, the projection position of the visible conductor segment in the current image, and the preset conductor sag range and swing range, determine the 3D reference axis of the conductor to be measured in the current state, and project the 3D reference axis onto the current image according to the current camera extrinsic parameters to obtain the bare conductor reference axis of the current image; determine multiple local measurement positions along the 3D reference axis, and determine the corresponding local measurement cross-section based on the 3D axis point and 3D local tangent vector at each local measurement position; S4: Determine the outer boundary of the ice accretion and perform spatial scale compensation: At each local measurement location, establish a two-dimensional boundary search line corresponding to the corresponding local measurement section based on the reference axis of the bare conductor in the current image. Determine candidate points of the outer boundary of the ice accretion on both sides of the reference axis of the bare conductor in the current image. Based on the current camera extrinsic parameters and the spatial geometric information of the corresponding local measurement section, convert the pixel scale of the candidate points of the outer boundary of the ice accretion in the current image into the physical scale on the corresponding local measurement section to obtain the spatial measurement results of the candidate outer boundary of the ice accretion. S5: Perform multi-frame temporal persistence screening: perform cross-frame association on candidate points of the outer ice boundary in multiple frames of the current image, determine the temporal persistence of the candidate boundary based on the number of frames of persistence of the candidate point of the outer ice boundary relative to the reference axis of the bare conductor in the current image, the stability of the normal distance, and the inter-frame motion characteristics, and remove transient candidate boundaries that do not meet the preset temporal persistence conditions to obtain the effective outer ice boundary. S6: Determine the icing thickness: Determine the composite diameter of the icing conductor at each effective local measurement position based on the spatial measurement results corresponding to the effective icing outer boundary, and determine the apparent radial icing thickness based on the composite diameter of the icing conductor and the nominal diameter of the bare conductor; For eccentric icing, use the three-dimensional reference axis obtained after the current state correction of the three-dimensional reference axis of the bare conductor as the reference axis of the bare conductor. Based on the spatial measurement results corresponding to the effective icing outer boundaries on both sides, determine the spatial radial distance from the three-dimensional axis point to the effective icing outer boundaries on both sides at each effective local measurement position. Determine the icing thickness on both sides based on the spatial radial distance on both sides and the nominal diameter of the bare conductor, and determine the eccentricity direction and degree based on the difference in icing thickness on both sides. S7: Perform measurement confidence evaluation and targeted re-acquisition: Determine the measurement confidence based on calibration quality, boundary positioning quality, temporal consistency, occlusion quality, and image quality; when the measurement confidence reaches a preset confidence threshold, output the icing thickness measurement result; when the measurement confidence is lower than the preset confidence threshold, determine the dominant quality parameter that caused the decrease in measurement confidence, generate the corresponding re-acquisition control signal based on the dominant quality parameter, and re-execute the corresponding image acquisition or measurement processing based on the re-acquisition control signal.

2. The self-calibration measurement method for icing thickness of transmission lines according to claim 1, characterized in that: In step S2, the calibration reference unit is a planar calibration target. The planar calibration target is equipped with at least four non-collinear reference feature points, orientation codes, and scale markers with known actual lengths. After distortion correction of the current image, the correspondence between the image coordinates of the reference feature points and their known coordinates in the calibration reference coordinate system is established. The random sampling consistency method is used to eliminate erroneous corresponding points, and the PnP algorithm is used to determine the current camera extrinsic parameters corresponding to each frame of the current image. When the image acquisition unit uses a zoom lens, the camera intrinsic parameters and lens distortion parameters corresponding to each focal length state are pre-stored, and the corresponding camera intrinsic parameters and lens distortion parameters are called according to the focal length state corresponding to the current image.

3. The self-calibration measurement method for icing thickness of transmission lines according to claim 1, characterized in that: In step S1, edge points on both sides of the bare conductor are extracted from multiple measurement locations in the bare conductor reference image. The midpoint of the edge points on both sides at the same measurement location is determined as the center position of the bare conductor. Curve fitting is performed on multiple bare conductor center positions to obtain the two-dimensional reference axis of the bare conductor. The spatial linear relationship or catenary relationship of the conductor relative to the calibration reference coordinate system is established using the line design parameters, conductor suspension point positions or field measurement results. Combined with the initialized camera extrinsic parameters, the center positions of multiple bare conductors are restored to the calibration reference coordinate system to obtain the three-dimensional reference axis of the bare conductor. In step S3, based on the projection positions of the continuous points of the visible conductor in the current image, the three-dimensional reference axis of the bare conductor is corrected in the current state within the sag range and swing range of the conductor to obtain the three-dimensional reference axis in the current state.

4. The self-calibration measurement method for icing thickness of transmission lines according to claim 1, characterized in that: In step S3, for each local measurement position, a three-dimensional axis point on the three-dimensional reference axis and a three-dimensional local tangent vector at that three-dimensional axis point are determined, and a spatial plane passing through the three-dimensional axis point and perpendicular to the three-dimensional local tangent vector is determined as the corresponding local measurement section. In step S4, an image measurement position corresponding to each local measurement position is determined on the current image bare conductor reference axis. A two-dimensional normal vector is determined based on the two-dimensional local tangent vector of the current image bare conductor reference axis at the image measurement position, and a two-dimensional boundary search line passing through the current image bare conductor reference axis is constructed along the two-dimensional normal vector. On the two-dimensional boundary search line, the comprehensive cost of the candidate boundary is determined based on the image gradient, the probability of icing foreground, the boundary continuity of adjacent image measurement positions, and the distance from the candidate boundary to the current image bare conductor reference axis. The pair of candidate points with the smallest comprehensive cost of the candidate boundary is selected as the icing outer boundary candidate points located on both sides of the current image bare conductor reference axis.

5. The self-calibration measurement method for icing thickness of transmission lines according to claim 4, characterized in that: Let the three-dimensional axis point at the i-th local measurement location be denoted as . A i , three-dimensional axis points A i The three-dimensional local tangent vector at point is denoted as τ i The corresponding local measurement section is denoted as Π i The local measurement section Π i satisfy: (X- A i ) T τ i =0 Where X represents any spatial point on the local measurement section, in step S4, the candidate points of the outer boundary of the ice covering on both sides of the i-th local measurement position are back-projected into spatial rays according to the camera intrinsic parameters and the current camera extrinsic parameters, and the spatial rays are aligned with the local measurement section. Π i The two intersecting points are denoted as follows: B i + and B i - In step S6, the Euclidean distance between the two three-dimensional boundary points is denoted as the composite diameter of the icing conductor. d i ,according to d i =‖ B i + - B i - ‖ Determine the composite diameter of the icing conductor d i The nominal diameter of the bare conductor is denoted as D 0, and according to h i =( d i - D 0) / 2 Determine the apparent radial icing thickness at the i-th local measurement location. h i .

6. The self-calibration measurement method for icing thickness of transmission lines according to claim 4, characterized in that: In step S4, under the conditions that the distortion correction of the current image has been completed, the angle between the normal vector of the calibration reference plane and the normal vector of the i-th local measurement section is not greater than a preset angle threshold, and the i-th local measurement position is within the pre-calibrated effective field of view, the spatial scale compensation is performed using a simplified depth ratio calculation method, and the known actual length on the calibration reference is denoted as... L ref The pixel length of the known actual length in the corrected current image is denoted as... l ref The depth of the calibration reference plane relative to the image acquisition unit is denoted as... Z t Let the depth of the i-th local measurement section relative to the image acquisition unit be denoted as . Z w,i Let the pixel distance between the outer ice-covered boundary points on both sides of the i-th local measurement position be denoted as . d p,i and in accordance with d i =( Z w,i / Z t )×( L ref / l ref )× d p,i Determine the composite diameter of the icing conductor at the i-th local measurement location. d i .

7. The self-calibration measurement method for icing thickness of transmission lines according to claim 1, characterized in that: In step S5, the number of consecutively acquired frames of the current image is denoted as M, and the threshold for the number of consecutive frames of the candidate boundary points is denoted as... T cont , where 1≤ T cont ≤M; Perform cross-frame association on candidate points of the outer boundary of ice in the current image of M frames, determine the normal distance variance of the candidate boundary points relative to the reference axis of the bare conductor in the current image and the inter-frame motion vector of the candidate boundary points, and determine the candidate boundary points with a continuous number of frames less than the continuous frame number threshold, a normal distance variance greater than a preset distance variance threshold, or inter-frame motion characteristics that meet the preset transient target conditions as transient candidate boundaries and remove them. The transient candidate boundaries include pseudo boundaries formed by snowflakes, scattered ice crystals or supplementary light reflection.

8. The self-calibration measurement method for icing thickness of transmission lines according to claim 1, characterized in that: In step S7, the calibration quality is determined based on the reprojection error of the current reference feature points and the number of effective reference feature points, and the calibration quality is denoted as... q ref The boundary positioning quality is determined based on the boundary gradient intensity of the outer icing boundary, the icing foreground probability, and the boundary continuity of adjacent measurement locations. This boundary positioning quality is denoted as... q edge Timing consistency is determined based on the multi-frame thickness variance and the number of boundary persistence frames, and the timing consistency is denoted as... q time The occlusion mass is determined based on the proportion of the effective local measurement section, and the occlusion mass is denoted as... q occ Image quality is determined based on high-frequency energy, motion blur index, and saturation pixel ratio, and this image quality is denoted as... q img Normalize each quality parameter to 0 to 1, and follow the... Q= β 1· q ref + β 2· q edge + β 3· q time + β 4· q occ + β 5· q img Determine the measurement confidence level Q, where, β 1. β 2. β 3. β 4 and β All 5 are non-negative weights, and satisfy the following conditions: β 1+ β 2+ β 3+ β 4+ β 5=1； When the measurement confidence level Q is lower than the preset confidence threshold, according to βk·(1- q k Calculate the weighted quality missing amount corresponding to the k-th quality parameter, and determine the quality parameter with the largest weighted quality missing amount as the dominant quality parameter; when the dominant quality parameter is calibration quality, expand the calibration benchmark search range and redetermine the current camera extrinsic parameters; when the dominant quality parameter is boundary positioning quality or image quality, adjust the exposure time, gain, focus position, or fill light state; when the dominant quality parameter is temporal consistency, increase the number of consecutive acquisition frames or shorten the frame interval; when the dominant quality parameter is occlusion quality, delay re-acquisition; when the measurement confidence Q is still lower than the preset confidence threshold after reaching the preset re-acquisition limit, output a measurement invalidation flag and the corresponding low quality reason information.

9. A field measurement device for self-calibrating icing thickness of transmission lines, characterized in that: The device includes an image acquisition unit, a processing unit, a memory, a control unit, and a communication unit. The image acquisition unit is used to acquire a current image that simultaneously includes the area of ​​the conductor to be measured and the calibration reference area. The calibration reference area is provided by a calibration reference unit that is separately set from the image acquisition unit and fixed on the stabilizing component of the transmission tower. The calibration reference unit is provided with reference feature points that have known coordinates in the calibration reference coordinate system. The processing unit is connected to the image acquisition unit and is used to receive the current image output by the image acquisition unit and determine the current camera extrinsic parameters based on the image coordinates of the reference feature points in the current image and the known coordinates in the calibration reference coordinate system. The memory is used to store camera intrinsic parameters, lens distortion parameters, known coordinates of calibration reference feature points in the calibration reference coordinate system, nominal diameter of bare conductor, two-dimensional reference axis of bare conductor, three-dimensional reference axis of bare conductor, and spatial geometric relationship of conductor; the processing unit is also used to determine the three-dimensional reference axis and local measurement section of the conductor under test in the current state based on the current camera extrinsic parameters, three-dimensional reference axis of bare conductor, spatial geometric relationship of conductor, and projection position of visible conductor segment in the current image, and to use the three-dimensional reference axis as the measurement spatial reference for icing outer boundary positioning, spatial scale compensation, multi-frame temporal continuity screening, icing thickness calculation, and measurement confidence evaluation; The control unit is used to output a corresponding resampling control signal based on the dominant quality parameter that causes the measurement confidence to decrease when the measurement confidence is lower than a preset confidence threshold; the communication unit is used to interact with the remote processing platform.

10. A self-calibration measurement system for icing thickness of transmission lines, characterized in that: The system includes an on-site measurement device, a calibration reference unit, and a remote processing platform. The on-site measurement device includes an image acquisition unit, a processing unit, a memory, a control unit, and a communication unit. The image acquisition unit acquires current images of the area to be measured and the calibration reference area. The processing unit, connected to the image acquisition unit, receives the current images output by the image acquisition unit, determines the current camera extrinsic parameters based on reference feature points in the calibration reference area, and determines the three-dimensional reference axis of the traverse under the current state based on the current camera extrinsic parameters, a pre-established three-dimensional reference axis of the bare traverse, and the spatial geometric relationship of the traverse. Using the three-dimensional reference axis as the measurement spatial reference, it performs local measurement section determination, icing outer boundary positioning, and spatial scale determination. The system includes compensation, multi-frame temporal continuity filtering, icing thickness calculation, and measurement confidence evaluation. The control unit outputs a re-acquisition control signal based on the measurement confidence. The calibration reference unit is fixedly installed on the stable component of the transmission tower and is separated from the image acquisition unit. The calibration reference unit is a passive optical calibration target. The calibration reference unit and the conductor under test are located within the common effective field of view of the image acquisition unit. The reference feature points set on the calibration reference unit are imaged by the image acquisition unit and input to the processing unit. The processing unit determines the current camera extrinsic parameters based on the image coordinates of the reference feature points and the known coordinates in the calibration reference coordinate system. The calibration reference unit has multiple reference feature points with known coordinates in the calibration reference coordinate system. The remote processing platform is connected to the field measurement device via a communication unit. The remote processing platform includes a result storage module and a verification service module. The result storage module is used to store the ice thickness measurement results according to the device identifier and the acquisition time. The verification service module is used to verify the ice thickness measurement results based on the changes in ice thickness distribution at adjacent measurement times and the measurement confidence level, and to issue a re-mining instruction to the field measurement device when the preset re-mining conditions are met.

Images