A power distribution network construction operation behavior specification detection method

Abstract

The application provides a kind of based on power distribution network construction operation behavior specification detection method, the application relates to power system construction operation safety management and intelligent monitoring technical field, the specification detection method of the application includes: identifying the needle insulator in the video stream collected by first worker and second worker as operation anchor point, generates the steady video stream with insulator as the image physical center;Steady video stream is mapped to cylindrical surface and weighted fusion is carried out, generate panorama, and panorama is logically divided into quadrant field of view containing main visual angle, opposite visual angle and virtual field of view, the compliance area boundary is mapped back to main visual angle by inverse cylindrical-plane projection, and depth and safety distance detection is carried out;Generate dynamic occlusion mask under panoramic space, when detecting that main visual angle is blocked, lock opposite visual angle as complementary window, determine the binding quality under the blocked area by calculating the spiral texture energy and periodic variance in the complementary window.

Description

Technical Field

[0001] This invention relates to the field of safety management and intelligent monitoring technology for power system construction operations, and more specifically, to a detection method based on the behavioral norms of power distribution network construction operations. Background Technology

[0002] As the "last mile" connecting users to the power system, the construction quality and operation and maintenance level of the distribution network directly affect the reliability of power supply and public safety. In the construction of 10kV and below medium and low voltage distribution networks, due to the complex line structure, numerous nodes, and limited working space, most key processes still require manual, high-altitude, precision operations. Especially for typical scenarios such as double-crossarm pin insulator conductor binding, current line splicing, and tension pole tightening, due to the process requirements involving conductor tension balance control, hardware assembly torque verification, and insulation distance maintenance, it is usually necessary for two or more workers to work face-to-face in coordination on the tower or inside an insulated bucket truck.

[0003] Currently, the standardized supervision of such high-altitude operations in power distribution networks mainly relies on on-site visual supervision by ground monitoring personnel and regular on-site inspections. However, this traditional manual supervision model has significant physical limitations. Taking the conductor binding operation of double crossarm pin insulators on 10kV overhead lines as an example, the work point is usually located at the top of the tower more than 10 meters above the ground. Ground monitoring personnel are limited by the upward viewing angle and line of sight, making it difficult to clearly observe the details of conductor winding on the top and back of the insulator (such as the number of binding loops, the tightness, and the direction of aluminum sheathing tape laying). In addition, the ground view is often obstructed by the tower hardware, the crossarm itself, or the body of the workers, resulting in a large number of blind spots in supervision, making it difficult to accurately judge the standardization of minor actions at high altitudes.

[0004] With the widespread adoption of smart wearable devices, some construction companies have begun equipping their workers with monocular head-mounted cameras or law enforcement recorders for monitoring through post-event playback or real-time image transmission. However, in practical applications, existing video monitoring technologies based on single-person, single-viewpoint operation have revealed serious technical deficiencies in multi-person collaborative work scenarios. First, severe self-occlusion cannot be overcome. During close-range operations such as wire binding or bolt tightening, workers' arms, hands, and tools frequently obstruct the camera's view, resulting in critical work points being in long-term blind spots, preventing the algorithm from extracting effective visual features. Second, monocular vision lacks spatial depth information and panoramic perception capabilities. Existing detection methods typically only perform static recognition of a single person's attire (such as whether a safety helmet is worn), lacking effective detection methods for the crucial spatiotemporal collaboration logic in two-person operations. For example, it is impossible to accurately determine whether the two workers are synchronized in their force application sequence, to construct a complete verification of the binding quality on the back side of the insulator, or to perceive the three-dimensional spatial distance between the workers and live conductors. When workers' heads shake or their viewing angles change, the video footage from a single perspective will experience severe shaking and target loss, resulting in high false alarm and false negative rates for detection algorithms. Therefore, there is an urgent need for a method to detect the behavior of power distribution network construction workers that can overcome blind spots caused by single-view occlusion, possess all-around spatial perception capabilities, and effectively verify the logic of multi-person collaborative operations. Summary of the Invention

[0005] This invention proposes a method for detecting the behavior norms of power distribution network construction operations, which includes the following steps:

[0006] The pin insulators in the video streams collected by the first and second operators are identified as operation anchor points. Based on the inverse image stabilization transformation, a stabilized video stream with the insulator as the physical center of the image is generated.

[0007] Construct a virtual cylindrical projection model coaxial with the work object, map the stabilized video stream onto the cylindrical surface and perform weighted fusion to generate a panoramic image, and logically divide the panoramic image into quadrant fields of view that include the main viewpoint, the opposite viewpoint and the virtual field of view.

[0008] Key hand points are extracted in the virtual field of view, and the boundaries of the compliant area are mapped back to the main view through inverse cylindrical-planar projection to perform depth and safe distance detection.

[0009] A dynamic occlusion mask is generated in the panoramic space. When occlusion of the main viewpoint is detected, the opposite viewpoint is locked as a complementary viewpoint. The binding quality in the occluded area is determined by calculating the spiral texture energy and periodic variance in the complementary viewpoint.

[0010] Performing an inverse image stabilization transformation based on anchor point centering includes: using a convolutional neural network to detect the bounding box of the region of interest of the pin insulator and extracting feature points within the region; and solving the inter-frame relative motion homography matrix by minimizing the reprojection error. :

[0011]

[0012] Where, time t and The first extracted within the insulator region The homogeneous coordinates of the feature points are as follows: and , Let be the set of interior points. Let be the motion matrix to be solved; Represent the third component of the vector; construct the inverse stable image matrix. Transform the original image:

[0013]

[0014] in, For the width and height of the image resolution, The coordinates of the detected insulator center. The angle of rotation of the camera relative to the direction of gravity.

[0015] Constructing a virtual cylindrical projection model coaxial with the work object involves mapping any pixel in the stabilized video stream. Mapped to the unfolded plane coordinates of the cylinder surface Its mapping model is:

[0016] in, This represents the horizontal azimuth angle relative to the optical axis. This represents the vertical height on the developed surface of the cylinder. This is the effective focal length of the camera.

[0017] The stabilized video stream is mapped onto the cylindrical surface and then weighted and fused to generate a panoramic image, including: calculating the panoramic image using a linear weighted fusion algorithm. upper phase angle is Height is pixel value at :

[0018]

[0019] in, and These are the pixel values ​​from the perspectives of the first and second workers, respectively. The weighting function for the overlapping region;

[0020] The panoramic image is logically divided into quadrant fields of view, including the main viewpoint, the opposing viewpoint, and the virtual field of view. Specifically: Defined as channel one That is, the first-person perspective; Defined as channel two That is, the perspective; and Defined as channel three respectively and Channel Four Channel 3 and Channel Four This is a virtual field of view.

[0021] Extract key hand points in the virtual field of view and perform depth and safe distance detection, including: in channel three Detection of key pixel coordinates of hand Restored to the azimuth angle in panoramic space and vertical height :

[0022]

[0023] in This represents the total width of the panoramic image;

[0024] Constructing the radial depth deviation operator To quantify the degree of contact between the hand and the insulator surface:

[0025]

[0026] in From the perspective of the target operational surface theory; if Greater than the preset floating threshold If so, it is determined to be a false swab.

[0027] Mapping the boundary of the compliant area back to the main viewpoint via inverse cylindrical-plane projection utilizes the following formula to determine the boundary points in the virtual field of view. Mapping back to the main perspective pixel coordinates :

[0028]

[0029] Draw depth benchmark lines and offside penalty area lines on the main view video stream using projected coordinates.

[0030] Generate a dynamic occlusion mask in panoramic space and lock the viewpoint as a complementary viewpoint, including: generating a binarized occlusion mask. When the main view work area The area of ​​the mask inside exceeds the threshold. Time-triggered detection:

[0031]

[0032] Locking and blocking center azimuth angle Phase difference is The area serves as the center of the complementary window. :

[0033]

[0034] And in Channel 2 Extracting Area of ​​interest centered .

[0035] Calculating the energy of the spiral texture within the complementary viewport includes: using the Gabor kernel function. With area of ​​interest Perform convolution operation to obtain the response And calculate the average texture energy. :

[0036]

[0037] in A set of pixels for complementary viewports;

[0038] like Below the preset threshold If the binding is loose, it is determined that the binding is loose.

[0039] Calculate the periodic variance and determine the binding quality in the occluded area, including:

[0040] Convolutional response The waveform is obtained by projecting along the height axis. Calculate the periodic variance of the wave crest spacing. And constructing a dead zone quality discrimination function :

[0041]

[0042]

[0043] in The threshold for permissible uniformity of arrangement.

[0044] This invention proposes a detection system for construction operation behavior standards in power distribution networks, the system comprising:

[0045] Video stabilization module: Identifies pin insulators in the video streams captured by the first and second operators as operation anchor points, and generates a stabilized video stream with the insulator as the physical center of the image based on inverse image stabilization transformation;

[0046] Field of view segmentation module: Constructs a virtual cylindrical projection model coaxial with the work object, maps the stabilized video stream onto the cylindrical surface and performs weighted fusion to generate a panoramic image, and logically segments the panoramic image into quadrant fields of view that include the main viewpoint, the opposite viewpoint and the virtual field of view;

[0047] Single-point detection module: Extracts key hand points in the virtual field of view, and maps the boundary of the compliant area back to the main view through inverse cylindrical-planar projection to perform depth and safe distance detection;

[0048] Complementary detection module: Generates a dynamic occlusion mask in panoramic space. When occlusion is detected in the main viewpoint, it locks the opposite viewpoint as a complementary viewpoint. By calculating the spiral texture energy and periodic variance within the complementary viewpoint, it determines the binding quality in the occluded area.

[0049] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned detection method based on the behavior norms of power distribution network construction.

[0050] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for detecting the behavior norms of power distribution network construction operations.

[0051] This invention effectively solves the detection challenges caused by worker swaying and limited field of view during high-altitude construction operations in power distribution networks by constructing a multi-source video stream stabilization and panoramic reconstruction mechanism based on visual anchor points. In typical scenarios such as conductor binding of double crossarm pin insulators on 10kV overhead lines, this invention utilizes rigid insulators in the working environment as an absolute physical reference. Through reverse motion compensation technology, it eliminates the irregular and significant shaking unique to high-altitude operations, achieving pixel-level image stabilization without relying on expensive and electromagnetically interference-prone external positioning equipment. This not only provides a stable spatial reference for subsequent image processing but also eliminates perspective distortion at the edges of wide-angle lenses by mapping the video streams from dual-person perspectives to a unified cylindrical coordinate system. This reconstructs the side blind spots that were originally at the edge of a single person's vision into a high-fidelity virtual frontal field of view, achieving 360-degree digital reproduction of the work object.

[0052] Furthermore, this invention overcomes the key technical bottleneck of monocular vision's lack of depth information by utilizing a multi-view geometric constraint mechanism. By dividing the panoramic space into standardized quadrant fields of view, this invention can accurately resolve the depth distance of the main viewpoint using the horizontal pixel distribution of the side viewpoint, thereby constructing a virtual depth sensor. This mechanism enables the detection system to accurately distinguish in two-dimensional video footage whether the operator's hand is effectively operating close to the insulator, making a false movement suspended in front of the conductor, or excessively extending beyond the safety red line of the insulator's top slot. This geometric measurement method based on complementary fields of view achieves real-time monitoring of the operator's contact with the conductor and safe distance without increasing hardware costs or wearing weight, significantly improving the standardization of refined operations in the power distribution network.

[0053] Furthermore, this invention resolves the physical contradiction of inevitably obstructed vision during intensive manual operations by establishing a quality inspection logic based on complementary de-obstruction. Addressing the issue of hand obstruction of critical crimping points during conductor binding, this invention utilizes the mechanical conductivity and texture continuity characteristics of the wire. When obstruction occurs in the main viewpoint, it automatically invokes a complementary viewpoint from the opposing viewpoint for texture energy analysis. By quantifying the tightness, regularity, and periodic variance of the back-side binding wires of the insulator, the system can infer the applied force quality from the physical state of the back side, thus logically achieving perspective detection. This method effectively prevents the escape of hidden defects such as incomplete binding, loose strands, or overlapping, ensuring that each loop of binding wire meets process standards. Attached Figure Description

[0054] Figure 1 This is a flowchart of the detection of the behavior norms for power distribution network construction operations based on the present invention. Detailed Implementation

[0055] This embodiment provides an application scenario and system architecture for detecting the behavior norms of power distribution network construction operations. Specifically, the application scenario is set as the conductor binding operation environment for double crossarm pin insulators on a 10kV power distribution network overhead line. In this environment, the construction operation is completed collaboratively by at least two qualified personnel.

[0056] The distribution network support structure consists of concrete poles or steel pipe poles, with double-layered or double-sided iron crossarms (referred to as double crossarms) installed on top. A preset working distance is maintained between the double crossarms, allowing workers to stand or suspend themselves on the sides of the crossarms. The work involves pin insulators fixedly installed on the double crossarms and bare or insulated conductors installed in the top or neck slots of the insulators. Pin insulators are ceramic or composite material components with a rotating structure; their physical characteristics are rigid and their positions are relatively fixed. The conductor binding operation involves using aluminum tape or binding wire to tightly wrap and fix the conductor to the neck of the pin insulator. This operation requires the binding wire to form a continuous, uniform, and tight loop around the neck of the insulator, without any loose strands, overlapping, or skipped turns.

[0057] The work involves a primary operator (the main operator) and a secondary operator (the assistant operator or monitoring personnel). During the operation, the primary and secondary operators are positioned on opposite sides of the pin insulator, forming a face-to-face collaborative relationship. During construction, the vertical center axis of the pin insulator is typically used as the origin, with the conductor's direction as the X-axis. The primary operator is positioned directly in front of the insulator (0-degree viewing angle), primarily responsible for threading and tightening the binding wire in the main line of sight. The secondary operator is positioned directly behind the insulator (180-degree viewing angle), primarily responsible for assisting with tool transfer, tightening the binding wire, and monitoring the binding quality on the back side of the insulator. The two operators' lines of sight intersect in the core work area, the insulator neck, but each operator has blind spots due to obstructions from the insulator itself or their own arms.

[0058] To acquire multi-source video data for behavioral compliance detection, both the first and second workers wore smart safety helmets. A high-definition wide-angle camera was fixedly integrated at the forehead of each helmet, with the camera's optical axis aligned with the worker's line of sight to simulate a first-person perspective. The camera featured image stabilization and low-light compensation, supporting at least 1080P resolution and a high frame rate of 60fps. Furthermore, the smart helmet also integrated an inertial measurement unit (IMU) and a wireless communication module. The IMU collected head posture data, including pitch, roll, and yaw angles, to help determine the camera's motion status; the wireless communication module transmitted the two heterogeneous video streams to an edge computing node in real time.

[0059] The system in this embodiment also includes a portable edge computing terminal deployed at the work site. The first video stream collected by the first operator and the second video stream collected by the second operator are concurrently transmitted to the edge computing terminal via a wireless network. To ensure the accuracy of subsequent logical verification, the system employs a strict time synchronization mechanism. The edge computing terminal acts as the master clock source, aligning the two video streams with millisecond-level timestamps using the Network Time Protocol (NTP). Simultaneously, based on the calibration images collected before the start of the operation, the system establishes a unified three-dimensional spatial coordinate system, spatially mapping and associating the 0-degree viewpoint coordinate system defined by the first video stream with the 180-degree viewpoint coordinate system defined by the second video stream.

[0060] Next, this embodiment details the stabilization of two heterogeneous video streams based on visual anchor points in the detection method for construction operation behavior norms in power distribution networks.

[0061] During the binding of conductors for double crossarm pin insulators on a 10kV overhead line, both the first and second workers wore smart safety helmets for close-range operations. Because the workers were standing at height with foot straps or inside an insulated bucket truck, their heads inevitably swayed irregularly due to breathing, muscle exertion, and high-altitude wind loads. This self-motion interference caused the core work object to shift drastically and become blurred in the acquired raw video stream. Directly performing panoramic stitching or motion analysis on this raw video would result in severe ghosting and misjudgments due to misalignment of the coordinate systems.

[0062] The specific implementation process of stabilizing two heterogeneous video streams based on visual anchor points includes the following three sub-steps:

[0063] S1.1: Identification of work anchor points and area locking;

[0064] Receive the first video stream collected by the first operator. The second video stream collected by the second operator Considering that during conductor binding operations, workers' hands frequently wrap around the insulator neck, and that hands are a high-dynamic interference item, while the background sky or ground is invalid information, this step must accurately locate the body area of ​​the pin insulator.

[0065] For the input time The original image The process involves processing and employing a pre-trained convolutional neural network optimized for power fitting textures for target detection. During detection, the unique rotating ceramic skirt structure and fixed physical properties of the pin-type insulator are utilized as prior knowledge to output a bounding box of the region of interest surrounding the insulator body. .

[0066] Let the coordinates of the detected bounding box center be... To avoid detection frame jitter caused by hand obstruction, detection confidence should only be increased when the detection confidence level is high. Updated only at that time. The position is determined by the position of the anchor point; otherwise, the predicted position of the previous frame is used, thus ensuring the stability of the anchor point selection at the first level.

[0067] S1.2: Optical flow tracing and motion matrix calculation based on local rigidity features;

[0068] In lock Subsequently, in order to obtain sub-pixel level camera motion parameters, the system only... Internal feature points (glaze reflection points). The feature points inside the insulator are rigidly attached to the pole, while... External feature points may be moving clouds, the ground, or the swaying arm of a worker; introducing external points can cause image stabilization calculations to fail.

[0069] Set time The first extracted within the insulator region The homogeneous coordinates of the feature points are: ,in and They represent the first The system tracks the horizontal and vertical coordinates of a feature point in the image pixel coordinate system at time t-1. The system uses the pyramid optical flow method to trace this point at time t-1. The homogeneous coordinates of feature points .

[0070] Considering that camera movement with the head includes translation, rotation, and scaling due to the worker's forward and backward movement, the inter-frame motion relationship is determined by a... homography matrix To describe it, we construct the following least squares optimization model to solve it. :

[0071]

[0072] in, This represents the set of interior points filtered by the RANSAC algorithm. Let be the motion matrix to be solved. The third component of the vector is used for normalization. This invention seeks an optimal transformation matrix such that the positional error of feature points on the insulator after transformation is minimized compared to the position observed in the current frame. This process eliminates interference from abnormal feature points caused by operator hand obstruction.

[0073] S1.3: Inverse image stabilization transformation based on anchor point centralization;

[0074] Unlike conventional image stabilization that aims for smooth image transitions, this application requires that the insulator be anchored at the geometric center of the image, defining the physical center of the image, i.e., the preset anchor point target position. ,in The width and height represent the image resolution.

[0075] Calculate from the initial time up to the current moment cumulative motion matrix To avoid cumulative error drift, this application does not directly use... Instead, it calculates in real time the insulator center detected in the current frame. Map back Furthermore, it can correct the camera rotation angle using an inverse image stabilization matrix. .

[0076] The structure combines rotational correction and translational compensation, and its mathematical expression is:

[0077]

[0078] Among them, parameters Indicates by The decomposed rotation angle of the camera in the current frame relative to the direction of gravity. The first term is expressed by the translation factor. Forcefully pull the insulator back to the center of the image; second item This is to counteract random jitter between frames.

[0079] use For the original image Each pixel coordinate Perform inverse mapping to generate a stabilized image. After this transformation, the output video stream In the image, the insulator remains centered and vertical, while the pole crossarm, the sky, and the worker's arm in the background move relative to the insulator.

[0080] This embodiment solves the technical problem that makes it impossible to perform detailed inspections due to viewpoint movement during operation.

[0081] Specifically, the feature extraction in the algorithm is limited to the insulator region, avoiding interference from hand movements during the binding operation on motion estimation; the construction of the inverse image stabilization matrix not only achieves image stabilization but also spatial normalization of the coordinate system. The final video stream is visually equivalent to shooting with the camera rigidly fixed to the crossarm.

[0082] The following embodiment details the panoramic cylindrical reconstruction and four-quadrant field-of-view generation in the method of this invention. After completing the image stabilization step, two stabilized video streams are obtained, each with the pin-type insulator as the absolute center and vertically erected: the stabilized video stream of the first operator. Video stream of the incubator for the second operator In double-crossarm conductor binding operations, the specifications require that the binding wire be tightly wrapped 360 degrees around the insulator neck. However, the first worker, positioned directly in front of the insulator, cannot see through the insulator body to view the binding wire on the back; the second worker, positioned directly behind, also cannot observe the details on the front. More importantly, the side area where the conductor contacts the insulator is a crucial angle for determining whether the binding wire has skipped a turn or is suspended, and this area is at the edge of both workers' perspectives, resulting in severe distortion and incomplete information in a single image.

[0083] Simple planar image stitching cannot intuitively demonstrate the continuous movement characteristics surrounding the insulator. Therefore, this embodiment constructs a virtual cylindrical projection model coaxial with the work object based on the cylindrical geometry of the pin insulator. By... and The image is mapped onto the cylindrical surface and pixel-blended to generate an unfolded panoramic view, which is then logically divided into four standardized quadrant fields of view to eliminate visual blind spots.

[0084] S2.1: Cylindrical projection transformation based on coaxial insulators;

[0085] A cylindrical coordinate system is established with the central axis of the pin insulator as the rotation axis (Z-axis). In the inverse image stabilization transformation based on anchor point centering in S1.3, the insulator has already been locked at the physical center of the image through inverse image stabilization. By default, the optical center of the input image coincides with the origin of the cylindrical coordinate system in this step.

[0086] Let the resolution of the stabilized image be... The effective focal length of the camera is For the input image any pixel on , express or Map it to the unfolded plane coordinates of the cylindrical surface. Since the insulator neck is the core area for conductor binding, the projected radius... This is set as the estimated distance from the camera's optical center to the insulator surface. Based on perspective projection geometry, the following planar-cylinder forward mapping model is constructed:

[0087]

[0088] in, This indicates the horizontal azimuth angle relative to the principal optical axis. This represents the vertical height on the cylindrical unfolded surface. The invention uses a wide-angle lens, and the image at the edges, i.e., the sides of the insulator, typically exhibits stretching distortion. By using a nonlinear transformation of the arctangent function, planar pixels are pasted onto the cylindrical surface, thereby correcting the perspective stretching at the edges of the insulator and ensuring that the tie wire maintains its true geometric proportions in the side view.

[0089] S2.2: Dual-field spatial registration and weighted fusion;

[0090] After obtaining the cylindrical projection, it is necessary to convert the projected image from the first operator into a digital image. And projected images from the second operator merged into the same panoramic image superior.

[0091] Panoramic view The horizontal axis corresponds to the following angle range: According to the location of the first worker (Directly in front), its field of view The coverage angle range is approximately The corresponding range mapped to the panoramic image is The second worker is located at (Directly behind), its field of view The coverage angle range is approximately .

[0092] in, This refers to the horizontal field of view of the camera; for a wide-angle lens, Therefore, their field of vision is and There are overlapping areas nearby. .

[0093] In the overlapping area, i.e., the side of the insulator, to eliminate the seam caused by differences in lighting or slight angular deviations between the two video streams, let the phase angle of a point on the panoramic image be . Its pixel value The calculation is as follows:

[0094] Weighting function and Defined as:

[0095]

[0096] in, and These are the starting and ending angles of the overlapping region, respectively. and To obtain from the projected image and projected images The angle of acquisition is Height is The pixel values ​​at the location. Through the above splicing, the present invention not only achieves a visually smooth transition, but also integrates the observation information of two people on the side of the insulator at the physical level, so that the texture details of the side tie wire, such as whether there are burrs and whether the crimping is tight, can be double-confirmed.

[0097] S2.3: Logical segmentation of the four-quadrant standard field of view;

[0098] To facilitate independent testing of standards from different orientations, the system will generate panoramic images. Logically divided into four independent standard field-of-view channels ,in .

[0099] Channel 1 (First-person perspective): Excerpt The region. This field of view is entirely derived from... The central area has the highest resolution and is used to detect the details of the first operator's hand movements.

[0100] Channel Two (From a perspective): Capture The region. This field of view is entirely derived from... The central area is used to detect the direction of the binding wires on the back of the insulator.

[0101] Channel 3 (Left virtual field of view): Capture The region. This region is comprised of... The right edge and It is formed by merging the left edge.

[0102] Channel 4 (Virtual view on the right): Capture area.

[0103] Through this segmentation, originally in The side binding wires of the insulator, which are difficult to see due to perspective compression, are in and The image is unfolded into a planar view. This allows the detection algorithm to clearly measure the vertical distance between the side conductor and the bottom of the insulator neck slot, just like observing the front. This embodiment solves the technical problems of visual information fragmentation and side observation distortion in two-person face-to-face operation scenarios by constructing a cylindrical projection model isomorphic to the pin insulator. The nonlinear projection in step S2.1 corrects the perspective distortion caused by the wide-angle lens at the edge of the insulator, so that the wire diameter width remains consistent in the panoramic image; the four standard fields of view generated in step S2.3 transform the side area, which was originally at the visual edge, into a virtual frontal view area. ).

[0104] Next, this embodiment details the main-view operation specification detection based on side-view depth projection in the method of the present invention. The first operator is positioned directly in front of the insulator, with their line of sight perpendicular to the conductor's direction. In this monocular main-view... There is a significant depth perception blind spot: when the worker's arm is extended forward to wrap the wire, it is impossible to distinguish on the two-dimensional image whether the hand is close to the neck of the insulator (compliant operation), suspended in front of the insulator (false binding), or extended too far into the top groove of the insulator (risk of electric shock).

[0105] To address this issue, this embodiment no longer relies on single-source information from the main perspective, but instead invokes the generated channel three. (The virtual field of view on the left) serves as a virtual depth sensor. Because... visual axis and The axes of vision are orthogonal in space. The depth information in the middle is precisely transformed into The horizontal position information (X-axis) in the image.

[0106] S3.1: Extraction of hand spatial coordinates in the side view;

[0107] Three-channel image Processing, in In the unfolded image of the cylinder, the pin insulator is located near the horizontal center of the image, while the first worker's hand reaches in from one side of the image.

[0108] The system uses the HRNet pose estimation network to detect hand keypoints, including the index fingertip and the center of the wrist. Let's assume the time is... ,exist The pixel coordinates of the hand key points detected in the image are , and Indicates in The horizontal and vertical coordinates of key hand points detected in the image.

[0109] Reconstruct the pixel coordinates into spatial parameters in cylindrical coordinates. The azimuth angle of the hand keypoint in panoramic space. and vertical height The calculation is as follows:

[0110]

[0111] in, Panoramic view Total width, . This reflects the absolute position of the hand in the direction surrounding the insulator, while This reflects the vertical height of the hand relative to the insulator base.

[0112] S3.2: Construct a radial depth calculation model for insulators;

[0113] From the main perspective It seems that the change in depth corresponds to medium angle Towards Directional offset. The physical radius of the pin insulator is known to be... The wire binding requires a hand operating radius. It must be in close contact with the surface of the insulator, that is .

[0114] Defining the target operation surface from a theoretical perspective For the binding operation from the main viewpoint, the ideal contact surface is the front surface of the insulator; therefore, the following setting is made:

[0115]

[0116] Constructing the radial depth deviation operator This operator is used to quantify the physical distance between the current position of the hand and the ideal operating surface in the side-view projection direction.

[0117]

[0118] The function maps angles in cylindrical coordinates back to the projection axis of the imaging plane of the side-view camera. The value directly reflects the normalized lateral distance between the key hand points and the center line of the insulator in the side view.

[0119] like ,express This means that the hand coincides with the front surface of the insulator in the side view, proving that zero-distance contact was achieved in the main view.

[0120] like , If the preset suspension threshold is set, it means that although the hands appear to overlap in the main view, a gap is displayed in the side view.

[0121] S3.3: Constructing and Determining Operational Compliance Envelope Functions

[0122] A three-dimensional operational envelope constraint model is established. This model consists of a vertical height constraint function. and horizontal depth constraint function composition.

[0123] Vertical height constraint: The height range of the insulator neck groove (binding area) on the cylindrical unfolded surface is defined as follows: ,

[0124]

[0125] in, and This indicates the upper and lower limits of the vertical pixel height of the insulator neck groove on the cylindrical unfolded diagram.

[0126] Vertical height constraints are used to exclude actions that are too high (touching the wire) or too low (tied to the skirt).

[0127] Horizontal depth constraints:

[0128] Horizontal depth constraints are used to eliminate false bucking actions.

[0129] The comprehensive judgment logic defines the operation validity status of the current frame. :

[0130]

[0131] Only when When the frame is determined to be a valid binding action frame, the current frame is identified.

[0132] S3.4: Back projection of side-view constraints and first-person perspective guidance

[0133] In order to be in the main perspective The detection results from the side view are displayed intuitively. The boundary of the compliant area determined in the side view field is mapped back to the main view image plane using the inverse cylindrical-plane projection matrix.

[0134] Based on the constraints in step S3.3, the four corner points of the compliance area are explicitly defined in the panoramic cylindrical coordinate system. Let the depth threshold be... The exported angle boundary values ​​are According to the formula , can be obtained The coordinates of the four corner points of the compliant area are defined as follows: top left corner point Top right corner , lower left corner point bottom right corner ;

[0135] Using the inverse projection formula, the coordinates of any of the four corner points mentioned above can be... Mapping back to the main perspective pixel coordinates :

[0136]

[0137] From the main perspective Dynamically overlaid visual guide lines on the video stream: depth target line and offside penalty area line.

[0138] The depth compliance line is composed of corner points. and as well as and The curve formed by connecting the projected points. When the key hand point is located inside this curve, it indicates that the physical depth is compliant. The offside penalty area line is defined by height. The corresponding projection line indicates this. If the hand pixels are higher than this line, it indicates a risk of electric shock.

[0139] This embodiment addresses the technical challenge of monocular ranging in high-altitude operations of power distribution networks by establishing a geometric projection model to represent channel three. This is transformed into a virtual depth sensor. The radial depth calculation model in steps S3.1 and S3.2 uses the horizontal pixel distribution of the side view to resolve the depth distance of the main view, solving the problem that two-dimensional images cannot distinguish between false anchorages and overhangs. The inverse projection in step S3.4 visualizes the abstract side-view geometric constraints as specific main-view pixel boundaries. This scheme, without adding expensive hardware such as LiDAR, achieves the detection of fit and safe distance specifications in wire binding operations solely through multi-view geometric coupling at the algorithm level.

[0140] The core operation of binding conductors for double-crossarm pin insulators involves spirally winding aluminum sheathing tape or binding wire around the insulator neck multiple times. During this process, the first worker's hand must be firmly against the insulator neck to apply force. This leads to an unavoidable visual problem: the point of force application is also the point of obstruction. From the main viewpoint... In the process, the most critical wire binding and crimping points are often completely obscured by the operator's own hands or tools, creating a blind spot. Traditional monocular inspection cannot penetrate the hand to see the condition of the wires below, easily missing the detection of loose bindings, i.e., the hand is moving but the wires are not tight or overlapped, i.e., the coils are overlapping and not spread out.

[0141] Given that insulator conductor binding is a continuous spiral winding process, the texture formed by the binding wire at the insulator neck has full-cycle continuity. Specifically, if the first operator does not pull the binding wire taut on the front side (0 degrees), or if the hand makes a feint (virtual binding), then this continuous wire will inevitably be loose when it reaches the back side (180 degrees). It is impossible for the back side to be loose on the front and tight on the back. Simultaneously, binding is a spiral ascent process. If skipped turns or overlaps occur on the front side, it will inevitably cause a sudden change in the spiral angle, and this geometric change will immediately continue to the back side. This embodiment utilizes the generated panoramic image... And four-quadrant field of view, using channel two (Regarding the viewpoint) as a complementary view, when the main viewpoint When occlusion occurs, the texture features of the opposing viewpoint are analyzed to quantify the tightness and regularity of the back wiring of the insulator, thereby inferring the work quality in the occluded area of ​​the main viewpoint and achieving logical perspective detection.

[0142] S4.1: Generate dynamic occlusion masks in panoramic space;

[0143] In a panoramic cylindrical coordinate system, locate the area obstructed by the first worker's hand. Utilize the detected hand key points. and its corresponding panoramic coordinates In the panoramic view Generate a binarized occlusion mask. .

[0144] At the same time, define the main view operation area. for The area. If , If the area threshold is set, it is determined that there is substantial occlusion in the current main viewpoint, triggering the complementary detection process.

[0145] S4.2: Complementary Window ROI Mapping;

[0146] When the main view is obstructed, the automatic lock-on phase difference with the center of obstruction is [value missing]. The 180-degree area is used as a complementary view.

[0147] Let the center azimuth angle of the obstructed area be . The center angle of the complementary window is ,here Need to be done Modular operation normalization to middle.

[0148] In Channel Two Extract from Centered on the insulator neck groove, the height range covers the insulator neck groove. The region of interest (ROI) is denoted as .

[0149] exist Within the designated area, the standard binding lines should appear as a set of parallel, closely spaced stripes with a specific angle of inclination. The system employs the Gabor kernel function. Extract texture features from this region to enhance the response to linear structures.

[0150] S4.3: Solving for the energy and continuity of spiral textures;

[0151] To quantify the binding quality, the system constructs a texture energy operator. Gabor core Complementary window image Perform convolution operations:

[0152] Calculate the average texture energy within the ROI region:

[0153]

[0154] in, It is the set of pixels for the complementary viewport.

[0155] like A higher reading indicates that there are clear, consistent wire binding patterns on the back of the insulator, proving that the binding wires have been tightened and are neatly arranged. If... Below the preset threshold This indicates that the back binding is loose, messy, or nonexistent, meaning that the hand movements from the main viewpoint did not create an effective wrap around the back.

[0156] In addition, to detect skipped turns or broken strands, the periodic variance of the texture distribution is calculated. Convolutional response Along the height axis By performing projection integration, a one-dimensional waveform is obtained. Properly tied at The upper part should be represented by equally spaced peaks.

[0157]

[0158] in and For the first and +1 peak position.

[0159] S4.4: Quality judgment based on de-occlusion complementarity;

[0160] Dead zone quality discrimination function :

[0161]

[0162] From the main perspective A hand was detected to be wrapping around (occlusion occurred), and complementary perspectives were used. Calculated The back stitching is taut and the texture is clear. When the spacing between the display coils is uniform, the system determines that the binding quality of the currently obscured dead corner area is qualified.

[0163] If the hand movements in the first-person perspective are large, but If it remains at a low level, the system will interpret it as a false breakout; if A sudden increase indicates either a skipped turn or a superposition of faults.

[0164] This embodiment addresses the quality inspection challenge caused by hand obstruction during 10kV conductor binding operations by proposing a panoramic complementary de-obstruction detection scheme. A panoramic coordinate system is used to define the blind spot range; texture energy analysis is introduced, utilizing the spiral continuity of the insulator binding to convert the invisible blind spot quality into visible diagonal texture energy for calculation. This overcomes the fundamental deficiency of existing technologies where monocular cameras cannot detect obstructed areas. Even if the first worker's hand completely blocks the wire binding joint, the system can still detect hidden defects such as insufficient force, loose coils, or uneven spacing through texture feedback from the second worker's perspective, ensuring the rigor of the power distribution network construction process.

[0165] This invention proposes a detection system for construction operation behavior standards in power distribution networks, the system comprising:

[0166] Video stabilization module: Identifies pin insulators in the video streams captured by the first and second operators as operation anchor points, and generates a stabilized video stream with the insulator as the physical center of the image based on inverse image stabilization transformation;

[0167] Field of view segmentation module: Constructs a virtual cylindrical projection model coaxial with the work object, maps the stabilized video stream onto the cylindrical surface and performs weighted fusion to generate a panoramic image, and logically segments the panoramic image into quadrant fields of view that include the main viewpoint, the opposite viewpoint and the virtual field of view;

[0168] Single-point detection module: Extracts key hand points in the virtual field of view, and maps the boundary of the compliant area back to the main view through inverse cylindrical-planar projection to perform depth and safe distance detection;

[0169] Complementary detection module: Generates a dynamic occlusion mask in panoramic space. When occlusion is detected in the main viewpoint, it locks the opposite viewpoint as a complementary viewpoint. By calculating the spiral texture energy and periodic variance within the complementary viewpoint, it determines the binding quality in the occluded area.

[0170] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned detection method based on the behavior norms of power distribution network construction.

[0171] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for detecting the behavior norms of power distribution network construction operations.

[0172] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0173] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0174] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting the behavior norms of power distribution network construction operations, characterized in that, The method includes: The pin insulators in the video streams collected by the first and second operators are identified as operation anchor points. Based on the inverse image stabilization transformation, a stabilized video stream with the insulator as the physical center of the image is generated. Construct a virtual cylindrical projection model coaxial with the work object, map the stabilized video stream onto the cylindrical surface and perform weighted fusion to generate a panoramic image, and logically divide the panoramic image into quadrant fields of view that include the main viewpoint, the opposite viewpoint and the virtual field of view. Key hand points are extracted in the virtual field of view, and the boundaries of the compliant area are mapped back to the main view through inverse cylindrical-planar projection to perform depth and safe distance detection. A dynamic occlusion mask is generated in the panoramic space. When occlusion of the main view is detected, the opposite view is locked as a complementary view. The binding quality in the occluded area is determined by calculating the spiral texture energy and periodic variance in the complementary view. Calculating the energy of the spiral texture within the complementary viewport includes: using the Gabor kernel function. With area of ​​interest Perform convolution operation to obtain the response And calculate the average texture energy. : in A set of pixels for complementary viewports; like Below the preset threshold If the binding is loose, it is determined that the binding is loose.

2. The method for detecting the behavior norms of power distribution network construction operations according to claim 1, characterized in that, Performing an inverse image stabilization transformation based on anchor point centering includes: using a convolutional neural network to detect the bounding box of the region of interest of the pin insulator and extracting feature points within the region; and solving the inter-frame relative motion homography matrix by minimizing the reprojection error. : Where, time t and The first extracted within the insulator region The homogeneous coordinates of the feature points are as follows: and , Let be the set of interior points. Let be the motion matrix to be solved; This represents the third component of the vector; Constructing an inverse image stabilization matrix Transform the original image: in, Width and height are the image resolution. The coordinates of the detected insulator center. The angle of rotation of the camera relative to the direction of gravity.

3. The method for detecting the behavior norms of power distribution network construction operations according to claim 2, characterized in that, Constructing a virtual cylindrical projection model coaxial with the work object involves mapping any pixel in the stabilized video stream. Mapped to the unfolded plane coordinates of the cylinder surface Its mapping model is: in, This represents the horizontal azimuth angle relative to the optical axis. This represents the vertical height on the developed surface of the cylinder. This is the effective focal length of the camera.

4. The method for detecting the behavior norms of power distribution network construction operations according to claim 3, characterized in that, The stabilized video stream is mapped onto the cylindrical surface and then weighted and fused to generate a panoramic image, including: calculating the panoramic image using a linear weighted fusion algorithm. upper phase angle is Height is pixel value at : in, and These are the pixel values ​​from the perspectives of the first and second workers, respectively. , The weighting function for the overlapping region; The panoramic image is logically divided into quadrant fields of view, including the main viewpoint, the opposing viewpoint, and the virtual field of view. Specifically: Defined as channel one That is, the first-person perspective; Defined as channel two That is, the perspective; and Defined as channel three respectively and Channel Four Channel 3 and Channel Four This is a virtual field of view.

5. The method for detecting the behavior norms of power distribution network construction operations according to claim 4, characterized in that, Extract key hand points in the virtual field of view and perform depth and safe distance detection, including: in channel three Detection of key pixel coordinates of hand Restored to the azimuth angle in panoramic space and vertical height : in This represents the total width of the panoramic image; Constructing the radial depth deviation operator To quantify the degree of contact between the hand and the insulator surface: in From the perspective of the target operational surface theory; if Greater than the preset floating threshold If so, it is determined to be a false swab.

6. The method for detecting the behavior norms of power distribution network construction operations according to claim 5, characterized in that, Mapping the boundary of the compliant area back to the main viewpoint via inverse cylindrical-plane projection utilizes the following formula to determine the boundary points in the virtual field of view. Mapping back to the main perspective pixel coordinates : Draw depth benchmark lines and offside penalty area lines on the main view video stream using projected coordinates.

7. The method for detecting the behavior norms of power distribution network construction operations according to claim 6, characterized in that, Generate a dynamic occlusion mask in panoramic space and lock the viewpoint as a complementary viewpoint, including: generating a binarized occlusion mask. When the main view work area The area of ​​the mask inside exceeds the threshold. Time-triggered detection: Locking and blocking center azimuth angle Phase difference is The area serves as the center of the complementary view. : And in Channel 2 Extracting Area of ​​interest centered .

8. The method for detecting the behavior norms of power distribution network construction operations according to claim 7, characterized in that, Calculate the periodic variance and determine the binding quality under occlusion regions, including: the convolution response. The waveform is obtained by projecting along the height axis. Calculate the periodic variance of the wave crest spacing. And constructing a dead zone quality discrimination function : in The threshold for permissible uniformity of arrangement.

9. A detection system for construction operation behavior standards based on power distribution networks, used to execute the detection method for construction operation behavior standards based on power distribution networks as described in claim 1, characterized in that, The system includes: Video stabilization module: Identifies pin insulators in the video streams captured by the first and second operators as operation anchor points, and generates a stabilized video stream with the insulator as the physical center of the image based on inverse image stabilization transformation; Field of view segmentation module: Constructs a virtual cylindrical projection model coaxial with the work object, maps the stabilized video stream onto the cylindrical surface and performs weighted fusion to generate a panoramic image, and logically segments the panoramic image into quadrant fields of view that include the main viewpoint, the opposite viewpoint and the virtual field of view; Single-point detection module: Extracts key hand points in the virtual field of view, and maps the boundary of the compliant area back to the main view through inverse cylindrical-planar projection to perform depth and safe distance detection; Complementary detection module: Generates a dynamic occlusion mask in panoramic space. When occlusion is detected in the main viewpoint, it locks the opposite viewpoint as a complementary viewpoint. By calculating the spiral texture energy and periodic variance within the complementary viewpoint, it determines the binding quality in the occluded area.

Images