Image data quality assessment methods, devices, and computer equipment for power systems

Abstract

This application relates to a method, apparatus, and computer device for image data quality assessment of power systems. The method includes: acquiring image data of a power system collected by an inspection robot; determining the minimum observable visual region based on the inspection question-and-answer unit and three-dimensional geometric model of the target power equipment in the power system; determining imaging feasibility constraints based on the inspection trajectory point sequence and the minimum observable visual region; filtering target imaging pose data from the imaging pose data under the constraints of the imaging feasibility constraints; for each frame of the image data, determining the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data, performing visual recognition analysis on the surface imaging pose to obtain a visual recognition state, and evaluating the image based on the visual recognition state to obtain the image quality assessment result. This method can be used to assess the quality of image data of power systems.

Description

Technical Field

[0001] This application relates to the field of digital image processing technology, and in particular to a method, apparatus and computer equipment for evaluating the image data quality of a power system. Background Technology

[0002] In the process of building a customer dialogue database for the power system, the quality of the power system image data will directly affect the accuracy and reliability of subsequent question and answer generation.

[0003] Currently, image data quality assessment in power systems typically relies on image sharpness indicators (such as Laplace variance) to filter acquired images, selecting those that meet preset thresholds for database construction. However, this method cannot effectively identify images lacking semantic information due to uneven lighting, partial occlusion, or shooting angle deviations. This leads to factual errors or missing information in subsequent customer dialogues generated based on image information, thereby reducing the credibility of the power system customer service dialogue database.

[0004] Therefore, how to assess whether image data is suitable for a power system customer service dialogue database is an urgent problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, apparatus, and computer equipment for evaluating the quality of image data in a power system that can assess whether image data is suitable for a power system customer service dialogue database, in order to address the aforementioned technical problems.

[0006] In a first aspect, this application provides a method for evaluating the image data quality of a power system, including:

[0007] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0008] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0009] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0010] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0011] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0012] In one embodiment, the inspection trajectory point sequence includes multiple inspection trajectory points; based on the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence, the imaging feasibility constraints of the minimum observable visual region are determined, including:

[0013] Based on the spatial relative relationship between the minimum observable visual area and multiple inspection trajectory points, a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual area is constructed.

[0014] Perform Boolean intersection processing on the ray beam family to obtain the target ray subfamily;

[0015] Obtain the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, perform coverage density analysis on the projection point distribution, and obtain the ray direction cluster center covering the surface of the smallest observable visual area.

[0016] The direction vector of the cluster center of the ray direction is used as the optical axis of the imaging device of the inspection robot, and the target imaging cone is constructed based on the optical axis;

[0017] The actual optical cone of the imaging device is obtained, and the target imaging cone is geometrically isomorphically matched with the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual area.

[0018] In one embodiment, the target imaging cone is geometrically isomorphically matched with the actual optical cone to obtain imaging feasibility constraints for the minimum observable visual region, including:

[0019] Based on geometric isomorphic matching between the target imaging cone and the actual optical cone, the target imaging direction of the smallest observable visual area is obtained.

[0020] Obtain the rotation component of the inspection trajectory point in the target imaging direction. Based on the rotation component, transform and synthesize the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis.

[0021] Calculate the three-dimensional angular deviation between the target pointing vector and the desired observation vector, where the desired observation vector is the vector pointing from the center point of the smallest observable visual area to the target inspection trajectory point corresponding to the optical axis;

[0022] Based on the three-dimensional included angle deviation, the motion adjustability of the target inspection trajectory points is judged, and the motion adjustability judgment result is obtained. Based on the motion adjustability judgment result, candidate imaging sites are selected from the target inspection trajectory points.

[0023] Based on candidate imaging sites, a spatial bidirectional reachability constraint map is constructed. If the spatial bidirectional reachability constraint map passes the backtrackability verification, the imaging feasibility constraint of the minimum observable visual region is obtained from the spatial bidirectional reachability constraint map.

[0024] In one embodiment, determining the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data includes:

[0025] Based on the target imaging pose data, the minimum observable visual area of ​​the target is determined in the image, and the imaging ray cone of the inspection robot's imaging device is constructed.

[0026] Spatial intersection processing is performed on the imaging ray cone and the minimum observable visual area of ​​the target to obtain the surface intersection region;

[0027] Orthogonal projection analysis is performed on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection profile;

[0028] The physical dimensions of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device are obtained. Based on the physical dimensions of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour, the smallest continuous pixel covering the orthogonal projection contour on the photosensitive surface is obtained through coordinate system alignment transformation.

[0029] Based on the positional correspondence between the smallest continuous pixel and the optical center of the imaging device in different coordinate systems, the surface imaging pose of the smallest observable visual area of ​​the target is determined.

[0030] In one embodiment, based on external ambient lighting conditions, visual recognition analysis is performed on the surface imaging posture to obtain the visual recognition state of the target's minimum observable visual area, including:

[0031] Based on the spatial orientation of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture under the external ambient lighting conditions, the incident direction vector of the main light source direction vector in the minimum observable visual area is obtained through coordinate system transformation.

[0032] Calculate the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device to obtain the illumination observation angle of the minimum observable visual area;

[0033] Shadow occlusion analysis is performed based on the illumination observation angle and the local differential geometric properties of the minimum observable visual region of the target, and the shadow occlusion analysis results of the minimum observable visual region of the target are obtained.

[0034] Visual recognition analysis is performed based on the results of shadow occlusion analysis to obtain the visual recognition status of the minimum observable visual area of ​​the target.

[0035] In one embodiment, the visual recognition state includes an unrecognizable state and a recognizable state; visual recognition analysis is performed based on the shadow occlusion analysis results to obtain the visual recognition state of the minimum observable visual region of the target, including:

[0036] Based on the shadow occlusion analysis results, determine whether there is a local feature occlusion effect on the surface of the target's smallest observable visual area;

[0037] In the presence of local feature occlusion effect, the visual recognition state of the smallest observable visual area of ​​the target is determined to be an unrecognizable state.

[0038] In the absence of local feature occlusion effect, the imaging tilt angle parameter based on the surface imaging posture is used to determine whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than the preset grazing avoidance angle.

[0039] If the three-dimensional angle is greater than the preset grazing avoidance angle, the visual recognition state of the smallest observable visual area is determined to be recognizable. If the three-dimensional angle is less than or equal to the preset grazing avoidance angle, the visual recognition state of the smallest observable visual area of ​​the target is determined to be unrecognizable.

[0040] Secondly, this application also provides an image data quality assessment device for a power system, comprising:

[0041] The acquisition module is used to acquire image data of the power system collected by the inspection robot. The image data includes multiple frames of images.

[0042] The mapping module is used to determine the minimum observable visual area of ​​multiple target power devices based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power devices in the power system.

[0043] The determination module is used to acquire the inspection trajectory point sequence of the inspection robot and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0044] The filtering module is used to acquire the imaging pose data of the inspection robot and, under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0045] The evaluation module is used to determine the surface imaging pose of the minimum observable visual region of the target in each frame of the image data based on the target imaging pose data. Based on the external ambient lighting conditions, it performs visual recognition analysis on the surface imaging pose to obtain the visual recognition state of the minimum observable visual region of the target. Based on the visual recognition state, it evaluates the image to obtain the image quality evaluation result.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0047] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0048] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0049] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0050] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0051] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0052] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0053] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0054] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0055] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0056] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0057] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0058] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0059] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0060] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0061] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0062] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0063] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0064] The aforementioned image data quality assessment method, apparatus, and computer equipment for power systems first acquire image data of the power system collected by an inspection robot, including multiple frames of images. Based on the mapping relationship between the inspection question-and-answer units and three-dimensional geometric models of multiple target power devices in the power system, the minimum observable visual area of ​​multiple target power devices is determined. The minimum observable visual area determined in the above process clearly defines the visual range required to ensure the effective presentation of the inspection question-and-answer semantic units. Second, by acquiring the inspection trajectory point sequence of the inspection robot, and based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence, the imaging feasibility constraints of the minimum observable visual area can be determined, clarifying the spatial conditions for effective imaging of each minimum observable visual area, and eliminating the potential risk of areas being unable to be effectively observed due to shooting angle deviations. Finally, the imaging pose data of the inspection robot is acquired. Under the constraints of imaging feasibility, the target imaging pose data is selected from the imaging pose data. For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined according to the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality evaluation result. In the above process, by obtaining the surface imaging pose of each smallest observable visual region in each frame of the image that can accurately capture the visual area, and combining the surface imaging pose with the external ambient lighting conditions, visual recognition analysis is performed to obtain the visual recognition state of each smallest observable visual region. Based on this, the image quality of each frame of the image is evaluated, which can effectively identify images that lack semantic information due to uneven lighting, partial occlusion, or shooting angle deviation. This avoids images that, although they can pass the sharpness detection, lack effective visual content in key power equipment areas from entering the database. The final accurate image quality evaluation results can be used to guide the removal of unqualified images that lack semantic information, ensuring that all images in the power system customer service dialogue database contain effective visual content. This avoids factual errors or missing information in customer service dialogues generated based on image content, thereby improving the content reliability of the power system customer service dialogue database. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1This is an application environment diagram of an image data quality assessment method for a power system in one embodiment;

[0067] Figure 2 This is a flowchart illustrating an image data quality assessment method for a power system in one embodiment;

[0068] Figure 3 This is a flowchart illustrating the steps for obtaining imaging feasibility constraints in one embodiment;

[0069] Figure 4 This is a flowchart illustrating the steps for obtaining imaging feasibility constraints through geometric isomorphic matching in one embodiment.

[0070] Figure 5 This is a flowchart illustrating an image data quality assessment method for a power system in another embodiment;

[0071] Figure 6 This is a structural block diagram of an image data quality assessment device for a power system in one embodiment;

[0072] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0074] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0075] The image data quality assessment method for power systems provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104, or it can be located in the cloud or on another network server. Specifically, terminal 102 or server 104 executes an image data quality assessment method for a power system. This method includes: acquiring image data of the power system collected by an inspection robot, the image data including multiple frames; determining the minimum observable visual region of multiple target power devices based on the mapping relationship between inspection question-and-answer units and three-dimensional geometric models of multiple target power devices in the power system; acquiring the inspection trajectory point sequence of the inspection robot, and determining the imaging feasibility constraint of the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence; acquiring the imaging pose data of the inspection robot, and filtering out target imaging pose data from the imaging pose data under the constraints of the imaging feasibility constraint; for each frame of the image data, determining the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data, performing visual recognition analysis on the surface imaging pose based on external ambient lighting conditions to obtain the visual recognition state of the minimum observable visual region of the target, and evaluating the image based on the visual recognition state to obtain the image quality assessment result.

[0076] Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0077] In one exemplary embodiment, such as Figure 2 As shown, a method for assessing the quality of image data in a power system is provided, which can be applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 210. Wherein:

[0078] Step 202: Obtain image data of the power system collected by the inspection robot. The image data includes multiple frames of images.

[0079] The inspection robot is used to move automatically within the area where the power system is located and to inspect multiple target power devices using its onboard sensors, including a positioning module, imaging equipment, a light sensor, and a posture sensor. The image data consists of multiple frames captured by the imaging equipment during the inspection process.

[0080] Optionally, an inspection robot can be used to inspect the target power system area and acquire image data of the power system collected by the inspection robot during the inspection process.

[0081] Step 204: Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, determine the minimum observable visual area of ​​multiple target power equipment.

[0082] The target power equipment refers to the power equipment that the inspection robot can effectively observe. The inspection question-and-answer unit is a semantic set related to the target power equipment, used to reflect the key inspection areas, requirements, and features of interest, including inspection-related questions, answers, and corresponding inspection points. The 3D geometric model is a digital model matched to the target power equipment, including the 3D coordinates, geometry, surface features, and connections between all components, used to reflect the spatial geometry of the target power equipment. The minimum observable visual area is the smallest geometric area that meets the inspection requirements contained in the corresponding inspection question-and-answer unit and can be used for independent visual observation and identification.

[0083] For example, a power system includes multiple power devices, such as transformers and circuit breakers. Using these multiple devices as initial inspection targets, a sequence of inspection trajectory points for the inspection robot is obtained, for example, containing 50 trajectory points, each with clearly defined three-dimensional spatial coordinates. The geographical coordinates of the initial inspection targets are obtained; these geographical coordinates and the coordinates of the inspection trajectory points belong to the same spatial coordinate system. Based on the coordinates of each inspection trajectory point and each geographical coordinate, through spatial distance calculation and orientation relationship analysis, intermediate power devices located more than 5 meters away (beyond the imaging observation range of the inspection robot) or obstructed by other equipment (unable to be directly observed) are identified. These intermediate power devices are removed from the initial inspection targets, resulting in the target power devices.

[0084] Optionally, the geographical coordinates of the power equipment can be measured using professional positioning equipment. The geographical coordinates of the power equipment are the absolute three-dimensional spatial coordinates of the power equipment in the actual inspection scenario, used to accurately locate the position of the power equipment in the actual space.

[0085] For example, inspection question-and-answer semantic units of multiple target power devices in a power system are obtained. By semantically parsing the inspection questions and answers, the specific physical location of the target power device corresponding to each inspection question-and-answer semantic unit is determined. Each inspection question-and-answer semantic unit points to a specific inspection location and feature of the corresponding target power device, without ambiguous pointing or overlapping pointing to multiple target power devices. By determining the specific physical location, the semantic-level inspection requirements can be transformed into physical-level target power device location pointers.

[0086] Furthermore, based on the actual structure, size, component distribution, and spatial relationships of the target power equipment, a three-dimensional geometric model of the target power equipment is constructed using three-dimensional modeling technology. Each identified inspection question-and-answer semantic unit is mapped to the specific physical part of the target power equipment and the three-dimensional geometric model. Based on the mapping relationship between the specific physical part and the three-dimensional geometric model, the minimum observable visual regions of multiple target power equipment are determined in the three-dimensional geometric model. The minimum observable visual region is used to fully present the characteristics of the inspection part of the target power equipment, and the conditions within the region cannot be further subdivided; further subdivision would fail to fully reflect the inspection requirements of that inspection part. Each minimum observable visual region corresponds to a unique inspection question-and-answer semantic unit and belongs to only one target power equipment.

[0087] Step 206: Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0088] The inspection trajectory point sequence refers to a series of spatial coordinate points collected and recorded in real time by the positioning module as the inspection robot travels along a preset inspection path during the inspection process. Spatial relative relationships include spatial parameters such as spatial distance, azimuth angle, and relative height difference between the smallest observable visual area and multiple inspection trajectory points. Imaging feasibility constraints refer to the constraints that the inspection robot must meet to obtain a clear image by capturing images of the smallest observable visual area.

[0089] For example, the embedding position of the boundary vertex of the minimum observable visual region in the three-dimensional geometric model of the target power equipment is determined, and the spatial coordinates of the inspection trajectory point sequence in global geographic coordinates are determined. Based on the embedding position and spatial coordinates, a spatial relative position analysis is performed to obtain the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence, and the imaging feasibility constraints of the minimum observable visual region are determined according to the spatial relative relationship.

[0090] Step 208: Obtain the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0091] Among them, the imaging pose data consists of key parameters that affect the image imaging effect, such as the spatial position of the inspection robot, the orientation of the imaging device, the imaging angle, and the imaging distance, when the inspection robot collects image data. These parameters are used to reflect the imaging state when each frame of image is acquired.

[0092] Optionally, the imaging pose data synchronously collected by the positioning module, attitude sensor and imaging equipment of the inspection robot can be obtained.

[0093] For example, an imaging feasibility analysis is performed on the imaging pose data to determine whether the imaging pose data meets the imaging feasibility constraints. If the imaging pose data meets the imaging feasibility constraints, this imaging pose data is used as the target imaging pose data.

[0094] Step 210: For each frame of the image data, determine the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data. Based on the external ambient lighting conditions, perform visual recognition analysis on the surface imaging pose to obtain the visual recognition state of the minimum observable visual region of the target. Based on the visual recognition state, evaluate the image to obtain the image quality evaluation result.

[0095] Among them, the surface imaging pose is the spatial pose of the smallest observable visual area in the corresponding frame of the image. The external ambient lighting conditions are lighting parameters that can affect the visual effect of the image, such as the light intensity, light direction, and light uniformity at the scene when the inspection robot acquires the image, and are used to reflect the scene lighting conditions at the time of acquisition of each frame of the image.

[0096] Optionally, each frame of the image includes at least one minimum observable visual region of the target.

[0097] For example, orthogonal projection analysis is performed based on the target imaging pose data to obtain the orthogonal projection contour; based on the boundary pixel coordinates of the orthogonal projection contour, the surface imaging pose of the minimum observable visual area of ​​each target in each frame image is determined by coordinate position analysis.

[0098] For example, by combining the external ambient lighting conditions and surface imaging pose of the target's minimum observable visual area, the visual recognition state of the target's minimum observable visual area is obtained through shadow occlusion analysis.

[0099] For example, for each frame of image, based on the visual recognition status of all minimum observable visual regions of targets in the image, and combined with the importance of the inspection question-and-answer semantic units corresponding to all minimum observable visual regions of targets, a differentiated weight allocation is performed. For example, the minimum observable visual region corresponding to the semantic unit with higher importance has a greater weight in terms of its influence on image quality, thus determining the quality level of that frame of image. The quality level is used as the quality assessment result to accurately reflect whether the image meets the inspection and recognition requirements.

[0100] In the aforementioned image data quality assessment method for power systems, firstly, image data of the power system collected by the inspection robot is acquired, including multiple frames of images. Based on the mapping relationship between the inspection question-and-answer units and the three-dimensional geometric models of multiple target power devices in the power system, the minimum observable visual area of ​​multiple target power devices is determined. The minimum observable visual area determined in the above process clearly defines the visual range required to ensure the effective presentation of the inspection question-and-answer semantic units. Secondly, by acquiring the inspection trajectory point sequence of the inspection robot, and based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence, the imaging feasibility constraints of the minimum observable visual area can be determined, clarifying the spatial conditions for effective imaging of each minimum observable visual area, and eliminating the potential risk of areas being unable to be effectively observed due to shooting angle deviations. Finally, the imaging pose data of the inspection robot is acquired. Under the constraints of imaging feasibility, the target imaging pose data is selected from the imaging pose data. For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined according to the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality evaluation result. In the above process, by obtaining the surface imaging pose of each smallest observable visual region in each frame of the image that can accurately capture the visual area, and combining the surface imaging pose with the external ambient lighting conditions, visual recognition analysis is performed to obtain the visual recognition state of each smallest observable visual region. Based on this, the image quality of each frame of the image is evaluated, which can effectively identify images that lack semantic information due to uneven lighting, partial occlusion, or shooting angle deviation. This avoids images that, although they can pass the sharpness detection, lack effective visual content in key power equipment areas from entering the database. The final accurate image quality evaluation results can be used to guide the removal of unqualified images that lack semantic information, ensuring that all images in the power system customer service dialogue database contain effective visual content. This avoids factual errors or missing information in customer service dialogues generated based on image content, thereby improving the content reliability of the power system customer service dialogue database.

[0101] In one embodiment, such as Figure 3 As shown, the inspection trajectory point sequence includes multiple inspection trajectory points; based on the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence, the imaging feasibility constraints of the minimum observable visual region are determined, including steps 302 to 310, wherein:

[0102] Step 302: Based on the spatial relative relationship between the minimum observable visual area and multiple inspection trajectory points, construct a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual area.

[0103] Here, the inspection trajectory point is the three-dimensional spatial coordinate of the inspection robot's location at a given moment. The ray family is a set of multiple rays pointing from the inspection trajectory point to different boundary vertices of the minimum observable visual area, used to reflect all possible ray directions from the inspection trajectory point to the minimum observable visual area.

[0104] For example, the spatial coordinates of each inspection trajectory point in the inspection trajectory point sequence are extracted in the global geographic coordinate system. The global geographic coordinate system refers to a three-dimensional coordinate system used to uniformly describe the position of all spatial objects (including inspection robots, target power equipment, etc.) in the inspection scenario. All spatial coordinates are defined based on this coordinate system, which can ensure the uniformity and comparability of spatial positions.

[0105] For example, the embedding positions of the boundary vertices of the minimum observable visual region in the three-dimensional geometric model of the target power equipment are determined. Based on the embedding positions and spatial coordinates, a spatial relative relationship analysis is performed to obtain the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points. Based on the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points, rays are constructed one by one, with each inspection trajectory point as the ray origin and each boundary vertex of the minimum observable visual region as the ray endpoint. All rays pointing from the same inspection trajectory point to all boundary vertices of the minimum observable visual region together constitute a ray set, i.e., a ray bundle family.

[0106] In this context, the boundary vertex refers to the edge endpoint of the geometric contour of the smallest observable visual region of the target. It can completely delineate the spatial geometric range of the smallest observable visual region. The number of boundary vertices for each smallest observable visual region is determined according to its geometric shape, ensuring that the spatial boundary of the smallest observable visual region can be accurately defined. The embedding position refers to the specific spatial coordinates of the boundary vertex in the coordinate system of the three-dimensional geometric model of the target power equipment. This coordinate system is consistent with the coordinate system of the three-dimensional geometric model and is used to reflect the specific position of the boundary vertex in the three-dimensional geometric model.

[0107] Step 304: Perform Boolean intersection processing on the ray beam family to obtain the target ray subfamily.

[0108] Among them, the target ray subfamily is a subset of rays in the ray beam family that are removed from those that are blocked by adjacent power equipment and can effectively reach the minimum observable visual area. It is used to reflect the effective ray direction of each inspection trajectory point pointing to the minimum observable visual area.

[0109] For example, the first physical geometric surface of the target power equipment and the second physical geometric surface of the adjacent power equipment are obtained. Here, the first physical geometric surface refers to all the external surfaces of the target power equipment itself, which are the externally observable physical surfaces in the three-dimensional geometric model of the target power equipment, reflecting its external geometric shape; adjacent power equipment refers to other power equipment that is close to the target power equipment in the actual inspection scenario and would obstruct the observation of the target power equipment; the second physical geometric surface refers to all the external surfaces of the adjacent power equipment itself, that is, the externally observable physical surfaces in the three-dimensional geometric model of the adjacent power equipment.

[0110] For example, for each ray in each ray beam family, a Boolean intersection operation is performed with both the first and second solid geometric surfaces to obtain a target ray subfamily. Specifically, the "Boolean intersection operation" means determining whether each ray intersects with the first and second solid geometric surfaces. If a ray intersects with the first solid geometric surface but not with the second solid geometric surface, that ray is considered a target ray. All target rays in the ray beam family are then combined to form the target ray subfamily. Here, a target ray is an effective ray that is not obstructed by adjacent power equipment and can directly reach the smallest observable visual area.

[0111] Step 306: Obtain the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, perform coverage density analysis on the projection point distribution, and obtain the ray direction cluster center covering the surface of the smallest observable visual area.

[0112] The local tangent plane coordinate system is used to describe the local spatial relationships of the surface of the smallest observable visual region. The ray direction cluster center is the ray direction in the target ray subfamily that can most densely cover the surface of the smallest observable visual region, and is used to reflect the ray direction that is most suitable for imaging in that region.

[0113] For example, a local tangent plane coordinate system is constructed with the geometric center of the smallest observable visual region as the origin, the tangent plane of the surface of this region as the XY plane, and the direction perpendicular to the tangent plane as the Z-axis. Each target ray in the target ray subfamily is projected onto the XY plane of the local tangent plane coordinate system to obtain the distribution of projection points of the target ray subfamily. Here, the projection point distribution refers to the distribution of projection points of the target rays onto the XY plane of the local tangent plane coordinate system. The coordinates of multiple projection points are defined based on the local coordinate system to reflect the projection position of the rays on the surface of the smallest observable visual region.

[0114] For example, the density gradient of the projection point distribution is determined by calculating the number of projection points in different regions on the XY plane of the local tangent plane coordinate system. Based on the density gradient, regions with dense and sparse projection points are identified, and the regions with dense projection points are taken as the core regions of the minimum observable visual region. Based on the ray directions corresponding to the projection points in the core regions, rays with similar directions are grouped into one category, the average direction of each category of ray directions is calculated, and the ray corresponding to the average direction is taken as the ray direction cluster center.

[0115] Step 308: Use the direction vector of the ray direction cluster center as the optical axis of the imaging device of the inspection robot, and construct the target imaging cone based on the optical axis.

[0116] Here, the direction vector is the vector pointing in the direction of the ray direction cluster center, used to describe the orientation of the ray corresponding to the ray direction cluster center. The target imaging cone refers to the cone-shaped spatial region formed with the optical axis of the imaging device as the center and a preset imaging angle as the cone angle, used to reflect the spatial range that the imaging device can cover when the ray direction is the imaging direction. The optical axis of the imaging device is the central axis of the inspection robot imaging device (such as a camera), used to determine the imaging direction of the imaging device.

[0117] For example, the direction vector of the cluster center for each ray direction is obtained. For each direction vector, the direction vector is used as the optical axis of the imaging device, and a target imaging cone is constructed with a preset imaging angle. For example, if the preset imaging angle is 60 degrees, then the target imaging cone is centered on the optical axis and has a cone angle of 60 degrees.

[0118] Step 310: Obtain the actual optical cone of the imaging device, and perform geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual area.

[0119] Among them, the actual optical cone refers to the cone-shaped imaging area inherent to the imaging equipment of the inspection robot and determined by its optical performance. The cone-shaped imaging area is determined by factors such as the lens parameters and imaging angle of the imaging equipment, and is the spatial range in which the imaging equipment can achieve effective imaging.

[0120] For example, the target imaging cone is geometrically isomorphically matched with the actual optical cone to obtain a target imaging direction that satisfies the condition that the actual optical cone contains the target imaging cone. Based on the target imaging direction, the target pointing vector of the optical axis in the global geographic coordinate system is determined. Based on the target pointing vector, the action fine-tuning discrimination result is obtained through action adjustability discrimination. The imaging feasibility constraint of the minimum observable visual area is determined according to the action fine-tuning discrimination result.

[0121] In this embodiment, by accurately selecting effective imaging directions, eliminating occlusion interference, and matching imaging equipment performance, the spatial conditions for effective imaging of each minimum observable visual area can be clearly defined. This ensures that the surface imaging posture determined based on imaging feasibility constraints can completely present the minimum observable visual area in the image, avoiding the problem of the minimum observable visual area being unable to be effectively observed due to factors such as shooting angle deviation and occlusion. This effectively improves the reliability of the power system customer service dialogue database.

[0122] In one exemplary embodiment, such as Figure 4 As shown, step 310 includes steps 402 to 410. Wherein:

[0123] Step 402: Based on the geometric isomorphic matching between the target imaging cone and the actual optical cone, the target imaging direction of the smallest observable visual area is obtained.

[0124] Among them, the target imaging direction is the optical axis direction of the imaging device that ensures the target imaging cone is completely within the actual optical cone and that the smallest observable visual area is fully imaged.

[0125] For example, by comparing the cone angles of the target imaging cone and the actual optical cone, it is determined whether the cone angle of the target imaging cone is less than or equal to the cone angle of the actual optical cone; and by comparing the axial directions of the target imaging cone and the actual optical cone, it is determined whether the axial direction of the target imaging cone is consistent with the axial direction of the actual optical cone, or whether the deviation is within a preset allowable range. If the cone angle of the target imaging cone is less than or equal to the cone angle of the actual optical cone and the axial direction deviation is within a preset allowable range, the target imaging cone is used as the intermediate imaging cone, and the direction vector of the ray direction cluster center corresponding to the intermediate imaging cone is obtained. This direction vector is then used as the target imaging vector.

[0126] Furthermore, the preset allowable deviation range is 0 to 5 degrees, which is determined based on the attitude adjustment capability of the imaging device. The determination process must ensure that the axis directions of the two can be aligned by fine-tuning the attitude of the imaging device.

[0127] Step 404: Obtain the rotation component of the inspection trajectory point in the target imaging direction. Based on the rotation component, transform and synthesize the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis.

[0128] The preset installation offset matrix describes the offset parameters of the imaging device's installation position and orientation on the inspection robot relative to the robot's own coordinate system. This matrix is ​​preset when the imaging device is installed on the inspection robot and reflects the relative position and orientation between the imaging device and the inspection robot. The target pointing vector is the vector pointing towards the optical axis of the imaging device in the global geographic coordinate system, and its direction is consistent with the target imaging direction.

[0129] For example, a definition module and attitude sensor of the inspection robot are used to collect six-degree-of-freedom pose data at all inspection trajectory points along the target imaging direction. The six-degree-of-freedom pose data consists of parameters describing the position and attitude of the inspection robot in space, including three translational components and three rotational components. The three rotational components describe the rotation angles of the inspection robot around the X, Y, and Z axes, reflecting the robot's spatial attitude. The three translational components describe the position of the inspection robot in the X, Y, and Z axis directions of the global geographic coordinate system.

[0130] For example, the pose data of the inspection robot can be converted into the pose data of the imaging device by using a preset installation offset matrix. Specifically, based on the preset installation offset matrix, coordinate transformation and attitude correction are performed on the rotation component of the inspection robot to eliminate the influence of installation offset between the imaging device and the inspection robot, converting the rotation attitude of the inspection robot into the rotation attitude of the imaging device, and obtaining the target pointing vector of the optical axis of the imaging device in the global geographic coordinate system.

[0131] Step 406: Calculate the three-dimensional angle deviation between the target pointing vector and the desired observation vector. The desired observation vector is the vector pointing from the center point of the smallest observable visual area to the target inspection trajectory point corresponding to the optical axis.

[0132] The three-dimensional angle deviation refers to the angle formed between the target pointing vector and the desired observation vector in three-dimensional space, with a value ranging from 0 degrees to 180 degrees. The desired observation vector is used to reflect the ideal observation direction of the optical axis of the imaging device, ensuring that the smallest observable visual area can be presented in the imaging image at the optimal angle.

[0133] Optionally, based on the coordinates of the boundary vertices of the minimum observable visual region, the average of the coordinates of all boundary vertices is calculated to obtain the coordinates of the center point of the minimum observable visual region.

[0134] For example, the magnitudes of the target pointing vector and the desired observation vector are calculated, and the three-dimensional angle deviation is obtained by vector dot product operation based on the magnitudes of the target pointing vector and the desired observation vector.

[0135] Step 408: Based on the three-dimensional angle deviation, the motion adjustability of the target inspection trajectory points is judged to obtain the motion adjustability judgment result. Based on the motion adjustability judgment result, candidate imaging sites are selected from the target inspection trajectory points.

[0136] Among them, candidate imaging sites refer to inspection trajectory points that meet the conditions for collision-free fine-tuning and can achieve optimal imaging. The action adjustability judgment result includes "fine-tunable" and "non-fine-tunable".

[0137] For example, "action adjustability judgment" refers to determining whether the imaging device of the inspection robot can eliminate the deviation between the target pointing vector and the desired observation vector by fine-tuning its own posture, so that the optical axis of the imaging device reaches the desired observation direction. Specifically, a preset three-dimensional angle deviation threshold is obtained, and the three-dimensional angle deviation is compared with the preset three-dimensional angle deviation threshold. If the three-dimensional angle deviation is less than or equal to the preset three-dimensional angle deviation threshold, it indicates that the deviation is small, and the imaging device can eliminate the deviation by fine-tuning its posture, and the action fine-tuning judgment result is determined to be "fine-tunable". If the three-dimensional angle deviation is greater than the preset three-dimensional angle deviation threshold, the deviation is too large, exceeding the posture fine-tuning range of the imaging device, and the deviation cannot be eliminated by fine-tuning, and the action fine-tuning judgment result is determined to be "not fine-tunable". Here, fine-tuning refers to the inspection robot making only a small adjustment to its own posture at the current inspection trajectory point position (the adjustment range does not exceed the preset fine-tuning threshold, which is a rotation of no more than 5 degrees around each coordinate axis and a translation of no more than 10 centimeters), and the adjustment process will not collide with the target power equipment, adjacent power equipment, or other obstacles in the inspection scene.

[0138] For example, an action adjustability judgment is performed for each target inspection trajectory point. If the action adjustability judgment result is "fine-tunable", the target inspection trajectory point is used as a candidate imaging site.

[0139] Step 410: Construct a spatial bidirectional reachability constraint map based on candidate imaging sites. If the spatial bidirectional reachability constraint map passes the backtrackability verification, obtain the imaging feasibility constraint of the minimum observable visual region from the spatial bidirectional reachability constraint map.

[0140] The spatial bidirectional reachability constraint graph, with the minimum observable visual region as the root node and the inspection trajectory points associated with the geometric imaging configuration as leaf nodes, is a graphical structure used to describe the spatial reachability relationship between the minimum observable visual region and the target imaging site, representing the minimum observable visual region that needs to be imaged. The leaf nodes are the inspection trajectory points associated with the geometric imaging configuration corresponding to each target imaging site; they are the terminal nodes of the spatial bidirectional reachability constraint graph, representing the inspection trajectory points that can achieve effective imaging.

[0141] For example, based on the temporal arrangement order of candidate imaging sites in the inspection trajectory point sequence, an imaging window sliding sequence is constructed, and the six-degree-of-freedom pose change rate between adjacent candidate imaging sites in each active window of the imaging window sliding sequence is calculated. Based on the six-degree-of-freedom pose change rate, the target imaging point is selected from the candidate imaging sites through curvature constraint analysis.

[0142] The target imaging points are those that satisfy the motion smoothness differential constraint, while the remaining imaging points are those whose images are blurred due to abrupt changes in six-degree-of-freedom pose. The temporal arrangement order refers to the order in which candidate imaging points are arranged according to the robot's movement sequence during inspection, from the start to the end of the inspection, consistent with the temporal sequence of the inspection trajectory points, reflecting the order in which the robot passes each candidate imaging point. The imaging window sliding sequence refers to setting a fixed-size sliding window, starting from the first candidate imaging point in the temporal arrangement, and sliding the window sequentially at a speed of moving one candidate imaging point at a time. The set of windows formed during all sliding processes is the result. The size of the sliding window can be adjusted according to the density of the inspection trajectory points; for example, the size can be preset to three candidate imaging points, meaning each window contains three consecutive candidate imaging points. The six-degree-of-freedom pose change rate refers to the rate of change of the robot's six-degree-of-freedom pose between two adjacent candidate imaging points. Curvature constraint analysis determines whether the pose change between adjacent candidate imaging points is too drastic and whether it will lead to image blurring.

[0143] Specifically, the translational changes and rotational changes around the X, Y, and Z axes of two adjacent candidate imaging sites are calculated respectively, and then divided by the inspection time interval between the two candidate imaging sites to obtain the rate of change of each component. The rate of change of all components together constitutes the six-degree-of-freedom pose change rate.

[0144] Specifically, a preset threshold for the six-degree-of-freedom pose change rate is obtained (generally, the translation rate threshold can be set to 5 cm / s, and the rotation rate threshold to 2 degrees / s). If the six-degree-of-freedom pose change rate between any two adjacent candidate imaging sites within the sliding window exceeds the preset threshold, it indicates a sudden change in the robot's motion posture between those adjacent sites, which would cause blurring in the image captured by the imaging device. In this case, candidate imaging sites with excessive change rates within the sliding window are removed. If the six-degree-of-freedom pose change rate between all adjacent candidate imaging sites within the sliding window does not exceed the preset threshold, it indicates smooth motion posture and no blurring in the imaging. In this case, all candidate imaging sites within the sliding window are retained. The retained candidate imaging sites are then used as target imaging sites.

[0145] For example, the geometric imaging configuration is determined based on the coupling relationship between the optical axis of the imaging device corresponding to the target imaging point and the local differential geometric properties of the surface of the minimum observable visual region. Here, the local differential geometric properties are the geometric features of the surface of the minimum observable visual region at each spatial point.

[0146] Specifically, the optical axis of the imaging device corresponding to each target imaging point is obtained. Based on the sign of the Gaussian curvature and the sign of the mean curvature of the surface of the minimum observable visual region, the local differential geometric properties of the surface of the minimum observable visual region are determined. The Gaussian curvature is the product of the two principal curvatures of the surface of the minimum observable visual region at a given point. The sign of the Gaussian curvature is positive, negative, or zero: a positive sign indicates a convex surface at that point; a negative sign indicates a concave surface; and a zero sign indicates a flat surface. The mean curvature is the average of the two principal curvatures of the surface of the minimum observable visual region at a given point. The sign of the mean curvature is also positive, negative, or zero: a positive sign indicates a generally convex surface at that point; a negative sign indicates a generally concave surface; and a zero sign indicates a flat surface.

[0147] Specifically, the coupling relationship between the optical axis of the imaging device corresponding to the target imaging point and the local differential geometric properties of the surface of the minimum observable visual region is analyzed. Based on this coupling relationship, the geometric imaging configuration is determined. The coupling relationship refers to the adaptation relationship between the direction of the imaging device's optical axis and the local concave-convex shape of the surface of the minimum observable visual region, i.e., whether the imaging device's optical axis can be perpendicular to the tangent plane of the surface, whether it can avoid concave surface occlusion, and whether it can completely cover convex surface features. The geometric imaging configuration is the set of imaging parameters that enable the imaging device to adapt to the local differential geometric properties of the surface of the minimum observable visual region at the target imaging point and ensure the complete presentation of the region's features. These parameters include the angle of the imaging device's optical axis, imaging distance, and imaging focal length.

[0148] For example, the smallest observable visual region is taken as the root node, and each target imaging point is taken as a leaf node. Based on the geometric imaging configuration, the connection relationship between the root node and each leaf node is established. According to spatial reachability, the connection relationship between adjacent leaf nodes is established. Based on the connection relationship, the root node, and the leaf nodes, a spatial bidirectional reachability constraint graph is constructed. Among them, the target imaging point corresponding to the leaf node can effectively image the root node (smallest observable visual region) through the geometric imaging configuration, and the attitude transition between adjacent leaf nodes can be achieved through smooth movement, ensuring that the graphic structure can reflect the spatial correlation between the target imaging point and the smallest observable visual region, as well as between the target imaging points themselves.

[0149] For example, starting from the target imaging point corresponding to the leaf node of the spatial bidirectional reachability constraint graph, a ray is drawn in the opposite direction of the optical axis of the imaging device. It is determined whether the ray will collide with or be blocked by the target power equipment, adjacent power equipment, and surrounding obstacles. If the ray can reach the area of ​​the minimum observable visual region without obstruction, the verification is passed; if the ray is blocked, the verification is failed, and the leaf node is removed. Based on the geometric imaging configuration, target pointing vector, and action fine-tuning range corresponding to all verified leaf nodes (target imaging points), imaging feasibility constraints for each minimum observable visual region are formed.

[0150] In this embodiment, the above process makes the obtained imaging feasibility constraints more targeted and operable, and has collision-free, blur-free, occlusion-free, and adaptable regional geometric features. It can effectively guide the inspection robot imaging equipment to achieve effective imaging of the smallest observable visual area in the optimal posture, ensuring that the features of the smallest observable visual area can be completely presented in the image. This ensures that the images used to build the power system customer service dialogue database can contain effective visual content, avoid the loss of semantic information due to unreasonable imaging positions or unsuitable imaging parameters, and improve the content reliability of the power system customer service dialogue database.

[0151] In an exemplary embodiment, determining the surface imaging pose of the minimum observable visual region of a target in an image based on target imaging pose data includes: determining the minimum observable visual region of a target in the image based on target imaging pose data, and constructing an imaging ray cone of the imaging device of the inspection robot; performing spatial intersection processing on the imaging ray cone and the minimum observable visual region of the target to obtain a surface intersection region; performing orthogonal projection analysis on the surface intersection region and the optical axis of the imaging device to obtain an orthogonal projection contour; obtaining the physical size of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device, and obtaining the minimum continuous pixels covering the orthogonal projection contour on the photosensitive surface through coordinate system alignment transformation processing based on the physical size of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour; and determining the surface imaging pose of the minimum observable visual region of the target based on the positional correspondence between the minimum continuous pixels and the optical center of the imaging device in different coordinate systems.

[0152] The target imaging pose data includes position and rotation components. The surface intersection region refers to the smallest observable visual region that can be covered by the imaging ray cone and presented in the imaging device's image. The orthogonal projection profile refers to the projection shape of the surface intersection region on the imaging plane, reflecting the approximate outline of the surface intersection region in the image. The smallest continuous pixel refers to the smallest pixel on the photosensitive surface that can completely wrap the orthogonal projection profile, and that is interconnected without gaps. Surface imaging pose includes features such as the region's orientation, tilt angle, and presentation ratio in the image, reflecting the actual imaging state of the visual region.

[0153] For example, the minimum observable visual region corresponding to the target imaging pose data is obtained, and the minimum observable visual region corresponding to the target imaging pose data is compared with the minimum observable visual region identified in the image. The same minimum observable visual region is taken as the minimum observable visual region of the target.

[0154] For example, based on the position component, the three-dimensional coordinates of the optical center of the imaging device are calculated by combining the preset installation offset matrix of the imaging device with respect to the inspection robot. Based on the three-dimensional coordinates and rotation component of the optical center of the imaging device, an imaging ray cone is constructed with the optical center of the imaging device as the vertex and the optical axis of the imaging device as the axis of symmetry.

[0155] For example, a spatial intersection operation is performed on the imaging ray cone and the minimum observable visual region of the target. Based on the operation result, it is determined whether there is an overlapping area between the spatial range of the imaging ray cone and the spatial range of the minimum observable visual region of the target. If there is, the overlapping area is taken as the surface intersection area between the imaging ray cone and the minimum observable visual region of the target.

[0156] Optionally, the boundary vertices of the surface intersection region are obtained, and the spatial geometric range of the surface intersection region is determined based on the boundary vertices. The three-dimensional spatial coordinates of each boundary vertex in the global geographic coordinate system are extracted. Using the optical axis of the imaging device as a reference, an imaging plane perpendicular to the optical axis of the imaging device is determined. Based on the three-dimensional spatial coordinates, orthogonal projection is performed by projecting all boundary vertices of the surface intersection region onto this imaging plane. For each boundary vertex, a ray parallel to the optical axis of the imaging device is drawn from that vertex, and the intersection of this ray with the imaging plane is the projection point of that boundary vertex on the imaging plane. Connecting the projection points of all boundary vertices in sequence forms a closed shape, which is the orthogonal projection profile. The imaging plane refers to the plane perpendicular to the optical axis of the imaging device and passing through the optical center of the imaging device, used to receive the projected image of the surface intersection region. Its spatial position is determined by the optical center of the imaging device and the target pointing vector.

[0157] For example, the boundary pixel coordinates of the orthogonal projection contour, as well as the physical dimensions of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device, are obtained by reading the imaging device parameters. The boundary pixel coordinates refer to the pixel coordinates of the boundary projection points of the orthogonal projection contour in the imaging plane coordinate system, reflecting the pixel distribution position of the projection contour on the imaging plane; the physical dimensions of the photosensitive surface refer to the length and width of the photosensitive surface of the imaging sensor, in millimeters, reflecting the actual size of the photosensitive surface; the pixel array arrangement parameters refer to the arrangement of pixels on the photosensitive surface of the imaging sensor, the pixel spacing, and the number of pixels. The pixel spacing refers to the distance between the centers of two adjacent pixels, and the number of pixels refers to the total number of pixels in the horizontal and vertical directions on the photosensitive surface.

[0158] Furthermore, taking the optical center of the imaging device as a reference, and combining the physical dimensions of the photosensitive surface and the pixel array arrangement parameters, the transformation relationship between the photosensitive surface coordinate system and the imaging plane coordinate system is calculated. The boundary pixel coordinates of the orthogonal projection contour are converted into pixel coordinates in the photosensitive surface coordinate system, thereby aligning the two coordinate systems and obtaining the converted boundary pixel coordinates. Based on the boundary pixel coordinates, a rectangular pixel region is defined, and all pixels in this region are taken as the minimum continuous pixels that cover the orthogonal projection contour on the photosensitive surface.

[0159] For example, the two-dimensional coordinates of the pixel center point of each smallest continuous pixel in the photosensitive surface coordinate system and the three-dimensional coordinates of the optical center of the imaging device in the global geographic coordinate system are obtained. Based on the above two-dimensional and three-dimensional coordinates, a family of reverse imaging rays starting from each pixel center point and passing through the optical center of the imaging device is constructed. The family of reverse imaging rays is spatially intersected with the smallest observable visual region to obtain the geometric landing point of each pixel center point on the surface of the smallest observable visual region. Based on the two-dimensional coordinates of each geometric landing point in the local tangent plane coordinate system and the local differential geometric properties of the surface of the smallest observable visual region at the geometric landing point, the differential structure consistency analysis between the pixel center point and the geometric landing point is performed to obtain the target pixel. Based on the distribution pattern of each geometric landing point in the target pixel in the local tangent plane coordinate system and the principal direction field of the smallest observable visual region in the local tangent plane coordinate system, the surface imaging pose of each target smallest observable visual region in each frame image is determined.

[0160] In this context, the pixel center point refers to the geometric center of each pixel, and the optical center of the imaging device refers to the center of the lens of the imaging device for the inspection robot. The reverse imaging ray family comprises a set of reverse imaging rays corresponding to multiple smallest consecutive pixels, reflecting the reverse ray connection between each smallest consecutive pixel and the optical center of the imaging device. A reverse imaging ray is a ray whose propagation direction is opposite to that of the actual imaging ray from the optical center of the imaging device, pointing towards the smallest observable visual area. A geometric landing point refers to the specific spatial point on the surface of the smallest observable visual area mapped from the pixel center point through the reverse imaging ray, used to establish a one-to-one correspondence between the photosensitive pixel and the surface of the smallest observable visual area. A target pixel is a pixel that is precisely mapped to the surface of the smallest observable visual area and can clearly capture the local geometric features of the surface. The distribution pattern refers to the arrangement, density, and overall contour of all geometric landing points on the local tangent plane coordinate system (XY plane), reflecting the projection distribution of the smallest observable visual area in the image; the principal direction field refers to the set of directions formed by the main extension directions of the surface of the smallest observable visual area.

[0161] Specifically, the coordinates of the top-left corner of each pixel in the photosensitive surface coordinate system are calculated, and half the pixel spacing is added to obtain the two-dimensional coordinates of the pixel center point of each smallest consecutive pixel in the photosensitive surface coordinate system. The position components of the inspection robot's imaging pose data are combined with the imaging device's preset installation offset matrix to calculate the three-dimensional coordinates of the imaging device's optical center in the global geographic coordinate system. Based on these two-dimensional and three-dimensional coordinates, starting from the pixel center point of each smallest consecutive pixel and passing through the optical center of the imaging device, reverse imaging rays are constructed one by one. All rays originating from the pixel center point of the smallest consecutive pixel and passing through the optical center of the imaging device form a family of reverse imaging rays.

[0162] Specifically, it is determined whether there is a unique intersection point between the spatial extension trajectory of each reverse imaging ray and the spatial range of the minimum observable visual region. The unique intersection point must be located on the surface of the minimum observable visual region and on the extension segment of the reverse imaging ray after it originates from the pixel center point and passes through the optical center of the imaging device (excluding the line segment between the optical center and the pixel center point). If a reverse imaging ray has a unique and compliant intersection point with the minimum observable visual region, this intersection point is taken as the geometric landing point of the pixel center point corresponding to the ray on the surface of the minimum observable visual region; if a reverse imaging ray has no intersection point with the minimum observable visual region or has multiple intersection points, it means that the pixel center point cannot be mapped to the surface of the minimum observable visual region, and the ray and the corresponding pixel center point are discarded.

[0163] Specifically, a local tangent plane coordinate system is constructed with the geometric center of the smallest observable visual area as the origin, the tangent plane of the surface of this area as the XY plane, and the direction perpendicular to the tangent plane as the Z-axis. The two-dimensional coordinates of each geometric point in this local tangent plane coordinate system are calculated. Based on these two-dimensional coordinates, the angle between the reverse imaging ray and the surface normal vector at the geometric point is calculated. Combined with the local differential geometric properties at the geometric point, it is determined whether the ray direction can clearly represent the local geometric feature. If the angle is less than or equal to 10 degrees, and the ray direction can adapt to the concave-convex shape of the geometric point (e.g., the ray direction is not obstructed on a convex surface, or the ray direction can penetrate into a concave surface), then the reverse imaging ray is determined to be consistent with the differential structure at the geometric point. If the angle is greater than 10 degrees, or the ray direction cannot adapt to the local concave-convex shape, then the reverse imaging ray is determined to be inconsistent with the differential structure at the geometric point. Pixel center points with consistent differential structures are selected as target pixels.

[0164] For example, based on the distribution pattern of each geometric point in the target pixel in the local tangent plane coordinate system, and the principal direction field of the minimum observable visual area in the local tangent plane coordinate system, the directional alignment degree between the principal axis of the geometric point distribution and the principal direction field is calculated to obtain directional aligned pixels that make the angle between the principal axis of the geometric point distribution and the principal direction field less than a preset differential alignment tolerance; based on the bidirectional mapping relationship between the pixel center point of each directional aligned pixel and its corresponding geometric point, and the projection direction of the optical axis of the imaging device in the local tangent plane coordinate system, the three-dimensional angle deviation between the optical axis of the imaging device and the normal vector of the local tangent plane is calculated to obtain the imaging tilt angle parameter; based on the imaging tilt angle parameter, the corresponding pixels of the surface intersection area in the image frame acquired by the imaging device at the corresponding inspection trajectory point are reverse geometrically reprojected to obtain the surface imaging pose of each minimum observable visual area in each frame image.

[0165] The principal orientation field reflects the overall extension trend of the surface area. For example, the principal orientation field of the smallest observable visual region in a planar region typically contains two mutually perpendicular principal extension directions, corresponding to the length and width directions of the region, respectively. The preset differential alignment tolerance is the maximum allowable angle set to ensure the accuracy of orientation alignment. Orientation-aligned pixels refer to target pixels whose geometric point distribution matches the principal orientation field of the smallest observable visual region and accurately reflects the overall extension trend of the region. The bidirectional mapping relationship refers to the mapping of the pixel center point to the geometric point through the reverse imaging ray. The projection direction refers to the target pointing vector of the imaging device's optical axis, used to reflect the projection orientation of the imaging device's optical axis onto the tangent plane of the smallest observable visual region surface.

[0166] For example, the coordinates of all geometric landing points on the local tangent plane coordinate system XY plane are statistically analyzed, the distribution center of the coordinates is calculated, and then an axis that runs through the distribution area and can cover all geometric landing points to the greatest extent is fitted based on the distribution center. This axis is the geometric landing point distribution principal axis. The directional alignment between the geometric landing point distribution principal axis and the main direction field is determined based on the degree of coincidence between the geometric landing point distribution principal axis and each of the main extension directions in the main direction field. Here, the geometric landing point distribution principal axis refers to the axis that can reflect the overall distribution trend of geometric landing points based on the distribution pattern of geometric landing points. The directional alignment is measured by the angle between the geometric landing point distribution principal axis and each of the main extension directions in the main direction field. The smaller the angle, the higher the directional alignment.

[0167] For example, based on the inspection accuracy requirements of the minimum observable visual area, a preset differential alignment tolerance is determined, such as 5 degrees. That is, when the angle between the principal axis of the geometric point distribution and a certain main extension direction in the principal direction field is less than 5 degrees, the two are determined to be aligned. All direction-aligned pixels that make the angle between the principal axis of the geometric point distribution and the principal direction field less than the preset differential alignment tolerance are selected from the target pixels.

[0168] For example, by mapping the forward imaging ray to the corresponding pixel center point, a bidirectional mapping relationship between the pixel center point of each aligned pixel and its corresponding geometric landing point is extracted. After converting the three-dimensional coordinates of the target pointing vector to coordinates in the local tangent plane coordinate system, the direction vector of the XY plane is extracted to obtain the projection direction of the imaging device's optical axis in the local tangent plane coordinate system. Through vector dot product operations, combined with the magnitudes of the target pointing vector of the imaging device's optical axis and the normal vector of the local tangent plane, the imaging tilt angle parameter formed by the target pointing vector and the imaging device's optical axis in three-dimensional space is calculated. The imaging tilt angle parameter refers to the tilt angle between the imaging device's optical axis and the vertical direction (normal vector direction) of the surface of the minimum observable visual area. It reflects the imaging device's shooting tilt angle and directly affects the surface imaging posture of the minimum observable visual area. The smaller the tilt angle, the closer the imaging is to a frontal shot, and the clearer the regional features are presented. The local tangent plane normal vector refers to the vector perpendicular to the tangent plane of the surface of the minimum observable visual area, and its direction is determined by the Z-axis direction of the local tangent plane coordinate system, used to reflect the vertical direction of the surface of the minimum observable visual area.

[0169] For example, images are acquired by the inspection robot at the corresponding target imaging site, and target pixels corresponding to the surface intersection regions in the images are extracted. Using the imaging tilt angle parameter as a calibration basis, the mapping relationship between target pixels and geometric landing points in the image is adjusted. Pixel offsets caused by the imaging tilt angle are corrected based on the mapping relationship. The pixel distribution in the image is then reversed to restore the spatial geometric distribution of the surface of the minimum observable visual area. Based on the spatial geometric distribution, the actual presentation state of the minimum observable visual area in the image is determined, and this actual presentation state is used as the surface imaging posture in the image.

[0170] In this embodiment, the above process provides support for visual recognition analysis and image quality assessment based on surface imaging posture, ensuring that the visual recognition status of the smallest observable visual area can be accurately determined, effectively identifying unqualified images with missing semantic information due to imaging posture deviation, and ensuring that the images used to construct the power system customer service dialogue database contain valid visual content, thereby improving the content reliability of the power system customer service dialogue database.

[0171] In an exemplary embodiment, based on external ambient lighting conditions, visual identification analysis is performed on the surface imaging posture to obtain the visual identification state of the target's minimum observable visual region. This includes: obtaining the incident direction vector of the main light source direction vector in the minimum observable visual region through coordinate system transformation based on the spatial orientation of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture in the external ambient lighting conditions; calculating the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device to obtain the illumination observation angle of the minimum observable visual region; performing shadow occlusion analysis based on the illumination observation angle and the local differential geometric properties of the target's minimum observable visual region to obtain the shadow occlusion analysis result of the target's minimum observable visual region; and performing visual identification analysis based on the shadow occlusion analysis result to obtain the visual identification state of the target's minimum observable visual region.

[0172] The external lighting environment refers to the relevant parameters of natural or artificial lighting that affect imaging effects in the inspection scene, including parameters such as the direction of the main light source, light intensity, and lighting uniformity. The main light source direction vector refers to the vector corresponding to the propagation direction of the light from the most important light source in the inspection scene (such as sunlight or main lighting), used to reflect the illumination direction of the main light source. The incident direction vector is the propagation direction vector of the main light source in the local tangent plane coordinate system, used to reflect the incident direction of the main light source relative to the surface of the smallest observable visual area. The observation direction vector refers to the direction vector of the imaging device's optical axis in the local tangent plane coordinate system. The lighting observation angle refers to the angle between the mirror image direction of the main light source's incident direction and the observation direction of the imaging device; the smaller the angle, the more favorable the lighting is for the imaging device to capture the surface features of the smallest observable visual area.

[0173] For example, the direction vector of the main light source is obtained by the light sensor carried by the inspection robot.

[0174] For example, based on the three-dimensional angle relationship indicated by the imaging tilt angle parameter, and combined with the transformation relationship between the local tangent plane coordinate system and the global geographic coordinate system, the three-dimensional coordinates of the main light source direction vector are transformed and calculated to eliminate the position offset and angle deviation between the two coordinate systems, so as to obtain the incident direction vector of the main light source direction vector in the local tangent plane coordinate system.

[0175] Optionally, the observation direction vector of the imaging device's optical axis in the local tangent plane coordinate system is obtained, and the vector perpendicular to the tangent plane of the minimum observable visual area (whose direction is determined by the Z-axis direction of the local tangent plane coordinate system) is determined as the local tangent plane normal vector. Using the local tangent plane normal vector as the axis of symmetry, the incident direction vector is mirrored around the local tangent plane normal vector to obtain a vector symmetric to the incident direction vector about the normal vector; the illumination observation angle between this mirror-symmetric vector and the observation direction vector is calculated.

[0176] For example, the surface morphology of the region is determined by combining local differential geometric properties. If the surface is convex or tends to be convex, the convex portion is assessed based on the illumination observation angle to determine whether it will block light and create shadows on the surface. If the surface is concave or tends to be concave, the system determines whether light can reach the concave area and whether shadows will form due to the inability of light to reach it. If the surface is flat, the system determines whether the illumination observation angle will cause uneven local illumination, resulting in differences in brightness. Based on the assessment results, the shadow occlusion analysis results are determined. These results include "existence of local feature occlusion effect" and "absence of local feature occlusion effect," used to determine whether each minimum observable visual area surface has a local feature occlusion effect caused by asymmetric illumination. Asymmetric illumination refers to an illumination state where the incident direction of the main light source and the observation direction of the imaging device are not symmetrical, resulting in uneven light distribution on the surface.

[0177] In this embodiment, the visual recognition and analysis process based on surface imaging posture and lighting conditions described above can effectively identify potential semantic information loss caused by uneven lighting and local occlusion, ensuring the accuracy of image quality assessment, thereby eliminating unqualified images, ensuring that all images used to construct the power system customer service dialogue database contain valid visual content, avoiding factual errors or missing information in customer service dialogues generated based on image content, and improving the content reliability of the power system customer service dialogue database.

[0178] In one embodiment, the visual identification state includes an unidentifiable state and an identifiable state. Visual identification analysis based on shadow occlusion analysis results is performed to obtain the visual identification state of the minimum observable visual region of the target, including: determining whether a local feature occlusion effect exists on the surface of the minimum observable visual region of the target based on the shadow occlusion analysis results; if a local feature occlusion effect exists, then the visual identification state of the minimum observable visual region of the target is determined to be an unidentifiable state; if no local feature occlusion effect exists, then based on the imaging tilt angle parameter of the surface imaging posture, determining whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than a preset grazing avoidance angle; if the three-dimensional angle is greater than the preset grazing avoidance angle, then the visual identification state of the minimum observable visual region is determined to be an identifiable state; if the three-dimensional angle is less than or equal to the preset grazing avoidance angle, then the visual identification state of the minimum observable visual region of the target is determined to be an unidentifiable state.

[0179] The "unrecognizable state" refers to the state where the features of the smallest observable visual region (corresponding to the inspection requirements of the inspection question-and-answer semantic unit) cannot be identified. This may be due to factors such as overexposure caused by excessively strong lighting, underexposure caused by insufficient lighting, or feature occlusion or distortion caused by improper imaging posture. These factors make it impossible to accurately distinguish the details and contours of the smallest observable visual region, thus failing to meet the inspection requirements. The "recognizable state" refers to the state where the features of the smallest observable visual region can be clearly identified, and the details, contours, and feature points of the region in the image can be accurately distinguished. The local feature occlusion effect refers to the phenomenon where, due to the uneven structure of the region's surface, under asymmetrical lighting, the raised parts block light, causing the recessed parts to form shadows, thus making it impossible to clearly observe the local features of the recessed parts.

[0180] For example, when the shadow occlusion analysis result is "local feature occlusion effect exists", it is determined that the visual recognition state of the minimum observable visual area of ​​the target is unrecognizable. When the shadow occlusion analysis result is "no local feature occlusion effect exists", it is determined that the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than the preset grazing avoidance angle based on the imaging tilt angle parameter of the surface imaging posture.

[0181] Optionally, if the three-dimensional angle is greater than the preset grazing avoidance angle, it indicates that the imaging light will experience grazing incidence blurring, and the grazing avoidance result is a grazing avoidance state that fails to avoid grazing incidence blurring. In this case, the visual recognition state of the minimum observable visual area is determined to be a recognizable state. If the three-dimensional angle is less than or equal to the preset grazing avoidance angle, it indicates that the imaging light will not experience grazing incidence blurring, and the grazing avoidance result is a grazing avoidance state that avoids grazing incidence blurring. In this case, the visual recognition state of the minimum observable visual area of ​​the target is determined to be an unrecognizable state.

[0182] In this embodiment, by performing layered judgment on the visual recognition status, the confirmation process of the recognition status not only takes into account the influence of shadow occlusion caused by illumination, but also takes into account the influence of grazing incidence blur caused by imaging angle. This ensures that the obtained visual recognition status can truly reflect the recognizability of the smallest observable visual area, providing a basis for image quality assessment. In turn, it effectively identifies images with missing semantic information due to uneven illumination and shooting angle deviation, preventing such unqualified images from entering the power system customer service dialogue database, ensuring the effective visual content of the database images, and improving the content reliability of the power system customer service dialogue database.

[0183] In one of the most specific embodiments, such as Figure 5 As shown, the following uses a transformer as an example to illustrate a method for evaluating the image data quality of a power system according to this application.

[0184] Step 1: Use an inspection robot to perform inspections and acquire image data collected by the inspection robot during the inspection process. The image data includes multiple frames of images.

[0185] Step 2: Identify the transformer as the target power equipment. Based on the transformer's inspection question-and-answer unit and three-dimensional geometric model, determine the minimum observable visual area of ​​the transformer.

[0186] Obtain the semantic units of the inspection questions and answers corresponding to the transformer. Assuming one of these units is "Is there any oil leakage on the surface of the transformer tank?", perform semantic analysis to determine that this unit points to the surface of the transformer tank as the inspection location. The inspection requirement is to observe whether there are visual features related to oil leakage on the tank surface. Obtain the 3D geometric model of the transformer, including all components such as the tank, core, coils, and radiator. Based on the mapping relationship between the inspection requirement and the inspection location and the 3D geometric model of the transformer, locate the complete outer surface area of ​​the transformer tank in the 3D geometric model. This outer surface area is then used as the minimum observable visual area for the corresponding inspection question and answer semantic unit.

[0187] Step 3: Obtain the inspection trajectory point sequence of the inspection robot during the inspection process, and determine the imaging feasibility constraints based on the spatial relative relationship between the inspection trajectory point sequence and the minimum observable visual area.

[0188] Assuming the outer surface of the oil tank has a rectangular geometry with four boundary vertices, denoted as vertex 1, vertex 2, vertex 3, and vertex 4, the embedding positions of these four boundary vertices in the transformer's 3D geometric model are determined, and their respective 3D coordinates are obtained. Any one inspection trajectory point from the inspection trajectory point sequence is extracted, and its spatial coordinates in the global geographic coordinate system are obtained. Starting from this inspection trajectory point, four rays are constructed with the four boundary vertices as endpoints. These four rays together form a ray bundle family pointing from the inspection trajectory point to the smallest observable visual area. The above steps are repeated to construct a corresponding ray bundle family for each trajectory point in the inspection trajectory point sequence, ultimately obtaining a ray bundle family that corresponds one-to-one with all inspection trajectory points.

[0189] The first solid geometric surface of the transformer's 3D geometric model and the second solid geometric surface of the transformer's adjacent electrical equipment are obtained. The first solid geometric surface includes all external surfaces of the transformer itself, such as the outer surface of the tank, the radiator surface, and the terminal surface. The adjacent electrical equipment is the circuit breaker. The second solid geometric surface includes all external surfaces of the circuit breaker, such as the outer casing surface and the wiring surface. By performing Boolean intersection operations on each ray in the ray beam family corresponding to all inspection trajectory points, the target ray subfamily corresponding to each inspection trajectory point is obtained.

[0190] For each inspection trajectory point, a target ray subfamily containing three effective rays is obtained. These three rays are projected onto the XY plane of the local tangent plane coordinate system, resulting in three projection points. These three projection points are concentrated in the central region of the XY plane, which is the core region with the highest projection point density. Cluster analysis is performed on the ray directions corresponding to these three projection points. It is found that the directions of the three rays are similar. The average direction of the three rays is calculated, resulting in an average ray direction, which is the cluster center of the ray direction corresponding to the target ray subfamily. Based on the cluster center of the ray direction corresponding to each target ray subfamily, coverage density gradient analysis is performed on the target ray subfamily corresponding to each inspection trajectory point, obtaining the cluster center of the ray direction corresponding to each target ray subfamily.

[0191] The direction vector of each ray direction cluster center is obtained, and this direction vector is used as the optical axis of the imaging device. A target imaging cone is constructed with a preset imaging angle, assuming its cone angle is 60 degrees and its axis direction is perpendicular to the geometric center of the smallest observable visual area on the transformer tank surface. The actual optical cone of the inspection robot imaging device is obtained, with an inherent cone angle of 65 degrees and a preset allowable deviation range of 0 to 5 degrees for the axis direction. Geometric isomorphic matching is performed between the two: the 60-degree cone angle of the target imaging cone is less than the 65-degree cone angle of the actual optical cone, satisfying the cone angle matching requirement; the deviation of the axis direction of the target imaging cone from the initial axis direction of the actual optical cone is 3 degrees, which is within the preset allowable range of 0 to 5 degrees, satisfying the axis direction matching requirement. Therefore, the direction vector of the ray direction cluster center corresponding to this target imaging cone is the target imaging direction; if the cone angle of another target imaging cone is 70 degrees (greater than the 65-degree cone angle of the actual optical cone), the matching requirement is not met, and its corresponding ray direction is not selected as the target imaging direction. At this moment, at the inspection trajectory point corresponding to the target imaging direction, the rotation components in the six-degree-of-freedom pose data of the inspection robot are 0 degrees around the X-axis, 10 degrees around the Y-axis, and 5 degrees around the Z-axis.

[0192] The preset installation offset matrix of the imaging device relative to the inspection robot is set to 0 degrees offset around the X-axis, Y-axis, and Z-axis of the inspection robot, ensuring that the imaging device and the inspection robot have completely identical postures. The rotation component of the inspection robot is transformed and synthesized with this preset installation offset matrix. Since the installation offset is 0, the rotation component of the inspection robot is directly used as the rotational posture of the imaging device, thus obtaining the target pointing vector of the imaging device's optical axis in the global geographic coordinate system. The direction of this vector is consistent with the target imaging direction. If the preset installation offset matrix is ​​3 degrees offset around the Z-axis of the inspection robot, then during the transformation and synthesis, the rotation component of the inspection robot (5 degrees around the Z-axis) is superimposed with the installation offset (3 degrees around the Z-axis) to obtain the posture of the imaging device rotated 8 degrees around the Z-axis, resulting in the corresponding adjusted target pointing vector.

[0193] The center point coordinates are calculated by averaging the coordinates of the four boundary vertices of the rectangular area on the surface of the fuel tank. Based on the center point coordinates and the coordinates of the corresponding inspection trajectory points along the optical axis of the imaging device, the desired observation vector pointing from the inspection trajectory point to the center point is calculated. Simultaneously, the target pointing vector is called, and the three-dimensional angle deviation between the two vectors is calculated. If the deviation is 2 degrees, which is less than or equal to a preset threshold of 5 degrees, it indicates that the imaging device can eliminate the deviation by fine-tuning its attitude, and the action fine-tuning judgment result is "fine-tunable". If the three-dimensional angle deviation corresponding to another inspection trajectory point is 7 degrees, which exceeds the preset threshold of 5 degrees, it indicates that the deviation cannot be eliminated by fine-tuning the attitude of the imaging device, and the action fine-tuning judgment result is "not fine-tunable".

[0194] Calculations show that, assuming the inspection trajectory point sequence contains 10 inspection trajectory points corresponding to the target imaging direction, each trajectory point has a corresponding action fine-tuning judgment result. Among them, the action fine-tuning judgment result of 3 trajectory points is "not fine-tunable", that is, their three-dimensional angle deviation cannot be eliminated by the collision-free fine-tuning action at the current position (during simulated fine-tuning, they will collide with adjacent circuit breakers, or the deviation exceeds the fine-tuning range), and these 3 trajectory points will be removed; the action fine-tuning judgment result of the remaining 7 trajectory points is "fine-tunable", and the simulated fine-tuning action is collision-free and can eliminate the deviation, resulting in a candidate imaging site containing 7 inspection trajectory points.

[0195] Candidate imaging sites were arranged chronologically as sites 1 to 7. A sliding window size of 3 sites was set, and an imaging window sliding sequence was constructed, including window 1 (site 1, site 2, site 3), window 2 (site 2, site 3, site 4), window 3 (site 3, site 4, site 5), window 4 (site 4, site 5, site 6), and window 5 (site 6, site 7) (if the last window has fewer than 3 sites, the actual number of sites is used). The six-degree-of-freedom pose change rate of adjacent sites within each window was calculated: In window 1, the rotation change rate between site 2 and site 3 was 3 degrees / second, exceeding the preset threshold of 2 degrees / second, indicating a sudden change in pose, and site 3 was removed; In window 2, the translation change rate between site 2 and site 4 (after removing site 3) was 6 cm / second, exceeding the preset threshold of 5 cm / second, and site 4 was removed; the pose change rates of adjacent sites in the remaining windows did not exceed the limit, and the corresponding sites were retained. Based on the above operations, sites 1, 2, 5, 6, and 7 were ultimately retained, resulting in 5 target imaging sites.

[0196] By analyzing the coupling relationship between the optical axis of the imaging device and the flat plane, and determining whether the optical axis is perpendicular to the plane (optimal imaging angle), the direction of the optical axis of the imaging device corresponding to each target imaging point can be determined as the vector pointing from the target. For point 1, the optical axis of the imaging device is perpendicular to the surface plane of the fuel tank, and the coupling relationship is good. Its geometric imaging configuration is determined as follows: the optical axis is perpendicular to the surface of the target area, the imaging distance is 3 meters, and the imaging focal length is 50 millimeters. For point 5, the optical axis of the imaging device forms an 85-degree angle with the surface plane of the fuel tank, and the coupling relationship is relatively good. After fine-tuning the optical axis angle to 90 degrees, its geometric imaging configuration is determined as follows: the optical axis is perpendicular to the surface of the target area, the imaging distance is 3.2 meters, and the imaging focal length is 50 millimeters, ensuring that the flat features of the fuel tank surface can be fully presented.

[0197] At this point, there are 5 target imaging sites (site 1, site 2, site 5, site 6, and site 7). Taking the smallest observable visual area on the transformer tank surface as the root node and these 5 target imaging sites as leaf nodes, based on the geometric imaging configuration of each site, the connection relationships between the root node and leaf nodes, and between adjacent leaf nodes, are established, constructing a spatial bidirectional reachability constraint graph. Backward ray traceability verification is performed starting from each leaf node: the backward rays corresponding to sites 1, 2, 5, and 7 are all unobstructed and can reach the root node area, thus the verification passes; the backward ray corresponding to site 6 is obstructed by the transformer radiator and cannot reach the root node area, thus the verification fails, and site 6 is eliminated. Integrating the geometric imaging configurations and target pointing vectors of the 4 verified sites, the imaging feasibility constraints for this smallest observable visual area are finally obtained.

[0198] Step 4: Acquire the imaging pose data of the inspection robot. Based on the imaging pose data and imaging feasibility constraints, obtain the target imaging pose data. For each frame of the image data, determine the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data. Based on the external ambient lighting conditions, perform visual recognition analysis on the surface imaging pose to obtain the visual recognition state of the minimum observable visual region of the target.

[0199] Four target imaging sites are associated with the imaging feasibility constraints. The imaging pose data of the inspection robot at one of these target imaging sites is obtained. Its position components are (100m, 50m, 10m) in the global geographic coordinate system, and its rotation components are 0 degrees around the X-axis, 10 degrees around the Y-axis, and 5 degrees around the Z-axis. Combined with the preset installation offset matrix of the imaging device relative to the inspection robot (with an offset of 0 degrees around each axis), the three-dimensional coordinates of the optical center of the imaging device are calculated to be (100m, 50m, 10.2m). With this optical center as the vertex, the target pointing vector (perpendicular to the minimum observable visual area of ​​the transformer tank surface) as the axis of symmetry, and the actual optical cone angle of the imaging device (65 degrees) as the cone angle, an imaging ray cone is constructed. The spatial intersection operation of this imaging ray cone with the minimum observable visual area of ​​the transformer tank surface is performed to obtain an overlapping rectangular area. This rectangular area is the surface intersection area, which is smaller than the minimum observable visual area and is the area that can be captured by the imaging device.

[0200] At this point, there are four boundary vertices of the smallest observable visual region. The three-dimensional spatial coordinates of these four boundary vertices in the global geographic coordinate system are obtained. The four boundary vertices are projected onto the imaging plane along a direction parallel to the optical axis of the imaging device, resulting in four projection points. These four projection points are connected in sequence to form a rectangle with the same shape as the surface intersection area but scaled down proportionally, which is the orthogonal projection profile.

[0201] The boundary pixel coordinates of the orthogonal projection contour are (100 pixels, 200 pixels), (300 pixels, 200 pixels), (300 pixels, 400 pixels), and (100 pixels, 400 pixels) in the imaging plane coordinate system. The physical dimensions of the photosensitive surface of the imaging device's imaging sensor are 10 mm × 8 mm, and the pixel array arrangement parameters are 1000 pixels horizontally and 800 pixels vertically, with a pixel pitch of 0.01 mm. Using the optical center of the imaging device as a reference, the boundary pixel coordinates of the imaging plane coordinate system are converted to pixel coordinates in the photosensitive surface coordinate system. The converted boundary pixel coordinates are (150 pixels, 250 pixels), (350 pixels, 250 pixels), (350 pixels, 450 pixels), and (150 pixels, 450 pixels). Find the leftmost pixel coordinates 150, the rightmost pixel coordinates 350, the topmost pixel coordinates 250, and the bottommost pixel coordinates 450, and define a rectangular pixel area. All pixels in this area from 150 to 350 pixels (horizontal) and from 250 to 450 pixels (vertical) are the smallest consecutive pixels covering the orthographic projection outline, totaling 200 × 200 = 40,000 pixels.

[0202] Therefore, there are 40,000 minimum consecutive pixels. The two-dimensional coordinates of the pixel center point of each pixel in the photosensitive surface coordinate system have been determined. Three pixel center points are selected, with two-dimensional coordinates of (150 pixels, 250 pixels), (250 pixels, 350 pixels), and (350 pixels, 450 pixels). The three-dimensional coordinates of the optical center of the imaging device in the global geographic coordinate system are (100 meters, 50 meters, 10.2 meters). Starting from these three pixel center points, and passing through the optical center of the imaging device, three reverse imaging rays are constructed. Each ray originates from the pixel center point, passes through the optical center, and extends towards the minimum observable visual area. Following the above method, corresponding reverse imaging rays are constructed for the pixel center points of all 40,000 minimum consecutive pixels. All rays together constitute a family of reverse imaging rays.

[0203] Each ray is spatially intersected with the smallest observable visual area on the transformer tank surface. 39,800 reverse imaging rays have a unique intersection point with this smallest observable visual area, and these intersection points are located on the extension segments of the rays after they pass through the optical center of the imaging device. These intersection points are the geometric landing points of the corresponding pixel centers. The remaining 200 reverse imaging rays, because their extension trajectories deviate from the smallest observable visual area, do not form intersection points. These 200 rays and their corresponding pixel centers are discarded, retaining the 39,800 geometric landing points and their corresponding pixel centers. All 39,800 geometric landing points are located on the transformer tank surface (with local differential geometric properties of zero Gaussian curvature and zero mean curvature, i.e., a flat plane), and their surface normal vectors are perpendicular to the tank surface. Calculate the angle between each reverse imaging ray and the normal vector at its corresponding geometric landing point. Among them, 39,700 rays have an angle ≤10 degrees with the normal vector, and the ray direction is perpendicular to the surface of the fuel tank, which can clearly present the flat surface features. They are judged to have consistent differential structures, and the corresponding pixel center point is the target pixel. The remaining 100 rays have an angle greater than 10 degrees with the normal vector, and the ray direction is too tilted, which cannot clearly present the surface features. They are judged to have inconsistent differential structures, and the corresponding pixel center points are removed, finally obtaining 39,700 target pixels.

[0204] Since there are 39,700 target pixels, their corresponding geometric landing points are all located on the surface of the transformer tank (a flat plane in the local tangent plane coordinate system). The geometric landing points are rectangularly distributed on the XY plane of the local tangent plane coordinate system, and the principal axes of the fitted geometric landing point distribution are two axes along the X-axis and Y-axis. The principal direction field of the smallest observable visual area contains two mutually perpendicular principal extension directions, corresponding to the length direction (X-axis direction) and width direction (Y-axis direction) of the tank surface, respectively. With a preset differential alignment tolerance of 5 degrees, the angle between the principal axis of the geometric landing point distribution and the principal extension direction of the principal direction field is calculated. Among them, the angle between the principal axis of the geometric landing point distribution corresponding to 39,600 target pixels and the principal extension direction of the principal direction field is less than 5 degrees, and the direction alignment meets the requirements. These 39,600 target pixels are selected as direction-aligned pixels; the angles corresponding to the remaining 100 target pixels are greater than 5 degrees, and the direction alignment does not meet the requirements, so they are discarded. Among them, the 39,600 direction-aligned pixels all have a clear bidirectional mapping relationship between the pixel center point and the geometric landing point. The projection direction of the target pointing vector of the imaging device's optical axis in the local tangent plane coordinate system is the positive X-axis direction, and the local tangent plane normal vector is the positive Z-axis direction (perpendicular to the fuel tank surface). By performing a vector dot product operation, the three-dimensional angular deviation between the target pointing vector of the imaging device's optical axis and the local tangent plane normal vector is calculated, yielding a deviation of 3 degrees. This 3-degree deviation is used as the imaging tilt angle parameter, indicating that the tilt angle between the imaging device's optical axis and the fuel tank surface perpendicular to the local tangent plane is 3 degrees.

[0205] With an imaging tilt angle of 3 degrees, image frames are acquired by the inspection robot at the corresponding target imaging point. Pixels corresponding to the surface intersection area in these image frames (a total of 39,600, consistent with the number of orientation-aligned pixels) are extracted. Based on the 3-degree imaging tilt angle parameter, these pixels undergo inverse geometric reprojection to correct the slight pixel shift caused by the 3-degree tilt angle. The pixel distribution in the image frame is then reversed to restore the spatial geometric distribution of the minimum observable visual area on the transformer tank surface. Finally, the surface imaging posture of the minimum observable visual area in the image frame is determined to be: facing the imaging device, tilted at a 3-degree angle, and presented at a scale consistent with the actual area, clearly showing the flatness of the tank surface. The 3-degree imaging tilt angle parameter indicates that the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane of the minimum observable visual area (transformer tank surface) is 3 degrees. The main light source in the external ambient lighting conditions is sunlight, and its main light source direction vector points downwards at 45 degrees in the global geographic coordinate system, obtained by the inspection robot's light sensor. Using a 3-degree imaging tilt angle as a reference, and combining the transformation relationship between the local tangent plane coordinate system (the origin is the geometric center of the smallest observable visual area on the fuel tank surface, the XY plane is the fuel tank surface, and the Z-axis is perpendicular to the fuel tank surface) and the global geographic coordinate system, the main light source direction vector is transformed to obtain the incident direction vector in the local tangent plane coordinate system as follows: positive X-axis direction, negative Z-axis direction (obliquely illuminating the fuel tank surface downwards). This vector is the incident direction vector.

[0206] The incident direction vector points in the positive X-axis and negative Z-axis directions in the local tangent plane coordinate system; the observation direction vector of the imaging device's optical axis points in the negative Z-axis direction (approaching perpendicular illumination of the fuel tank surface) in the local tangent plane coordinate system; the local tangent plane normal vector points in the positive Z-axis direction. First, using the positive Z-axis as the axis of symmetry, the mirror-symmetric vector of the incident direction vector is calculated, resulting in a vector pointing in the negative X-axis and negative Z-axis directions; then, the three-dimensional angle between this mirror-symmetric vector and the observation direction vector (negative Z-axis direction) is calculated, yielding an angle of 30 degrees. This 30-degree angle is the illumination observation angle, indicating a 30-degree deviation between the mirror direction of the main light source's incident direction and the observation direction of the imaging device.

[0207] Based on the illumination observation angle and the plane's properties, a shadow occlusion analysis was performed to determine whether a local feature occlusion effect existed at a 30-degree illumination observation angle. Since the surface of the area is a flat plane without any protrusions or depressions, the light can evenly illuminate the entire area, and no shadows are formed. Therefore, the shadow occlusion analysis result is "no local feature occlusion effect." However, if the smallest observable visual area is the convex structure of the transformer terminal block (with a positive Gaussian curvature), and the illumination observation angle is 60 degrees, the convex portion will block light, forming a shadow below the convex surface. This makes the local features of the shadowed area unobservable, and the shadow occlusion analysis result is "a local feature occlusion effect exists." The transformer terminal surface was determined to have a local feature occlusion effect. This area has a convex structure, and the protrusions block light, forming a shadow below the convex surface. This prevents the imaging equipment from clearly capturing features such as the terminal screws and connections in the shadowed area, making visual identification impossible. Therefore, the visual identification state of this smallest observable visual area is determined to be unidentifiable.

[0208] When the surface of the transformer tank is determined to be free of local feature occlusion effects, and its surface is a flat plane with no features obscured by shadows, the imaging tilt angle parameter for that area indicates a 3-degree three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane. The preset grazing avoidance angle is 10 degrees. Comparing 3 degrees with 10 degrees, 3 degrees is less than 10 degrees. Therefore, it is determined that the 3-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is not greater than the preset grazing avoidance angle, resulting in a grazing avoidance state that avoids grazing incidence blur. If another smallest observable visual area without local feature occlusion effects has an imaging tilt angle parameter indicating a 12-degree three-dimensional angle, which is greater than the preset grazing avoidance angle of 10 degrees, then the grazing avoidance result is a grazing avoidance state that does not avoid grazing incidence blur.

[0209] The grazing avoidance result on the transformer tank surface is a grazing avoidance state that avoids grazing incidence ambiguity. This area has no local feature occlusion effect, and the imaging light has no grazing incidence ambiguity. Features (tank surface welds, markings, etc.) can be clearly captured by the imaging equipment. Therefore, the visual recognition state of this smallest observable visual area is determined to be recognizable. If another smallest observable visual area does not have a local feature occlusion effect, its grazing avoidance result is a grazing avoidance state that does not avoid grazing incidence ambiguity. The imaging light has grazing incidence ambiguity, regional feature details are lost, and it cannot be clearly recognized. Therefore, its visual recognition state is determined to be unrecognizable.

[0210] Step 5: Evaluate the image based on the visual recognition status to obtain the image quality evaluation result.

[0211] Based on the above process, the shadow occlusion analysis results for the transformer tank surface (flat plane) and the transformer terminal surface (convex structure) are as follows: the shadow occlusion analysis result for the transformer tank surface is "no local feature occlusion effect," while the shadow occlusion analysis result for the transformer terminal surface is "local feature occlusion effect exists." Analysis of these results confirms that the transformer tank surface does not exhibit a local feature occlusion effect due to asymmetrical lighting, while the transformer terminal surface does exhibit a local feature occlusion effect due to asymmetrical lighting.

[0212] Assuming that when the inspection robot acquires a frame of image, the surface imaging posture of the smallest observable visual area is facing the imaging device, the imaging angle is perpendicular to the fuel tank surface, and the imaging scale is moderate, capable of completely presenting the area of ​​the fuel tank surface; at this time, the external ambient lighting conditions are moderate light intensity (500-800 lux), uniform light direction, and no obvious reflections or shadows. Based on the surface imaging posture and lighting conditions, visual recognition analysis can clearly identify the flatness of the fuel tank surface, and no obvious oil stain color difference or oil stain outline is found. Therefore, the visual recognition state of the smallest observable visual area is determined to be recognizable.

[0213] Because the visual recognition state of the smallest observable visual area corresponding to "whether the transformer radiator fins are blocked" is identifiable, while the visual recognition state of the smallest observable visual area corresponding to "whether the transformer terminals are loose" is unidentifiable due to overexposure caused by excessive lighting, the features cannot be recognized. Considering the importance of the inspection question-and-answer semantic units corresponding to the three smallest observable visual areas, the weights are assumed to be: "whether the transformer terminals are loose" (weight 40%) > "whether there are oil leaks on the surface of the transformer tank" (weight 30%) > "whether the transformer radiator fins are blocked" (weight 30%). The total identifiable weight of this frame is calculated as: 40% × 0 (unidentifiable) + 30% × 1 (identifiable) + 30% × 1 (identifiable) = 60%. The preset image quality assessment standard can be: a total identifiable weight ≥ 80% is a high-quality image, 60%-79% is a qualified image, and < 60% is an unqualified image. Therefore, the quality level of this frame is qualified.

[0214] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0215] Based on the same inventive concept, this application also provides an image data quality assessment device for power systems to implement the image data quality assessment method for power systems described above. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more embodiments of the image data quality assessment device for power systems provided below can be found in the limitations of the image data quality assessment method for power systems described above, and will not be repeated here.

[0216] In one exemplary embodiment, such as Figure 6 As shown, an image data quality assessment device 600 for a power system is provided, comprising: an acquisition module 602, a mapping module 604, a determination module 606, a filtering module 608, and an assessment module 610, wherein:

[0217] The acquisition module 602 is used to acquire image data of the power system collected by the inspection robot. The image data includes multiple frames of images.

[0218] The mapping module 604 is used to determine the minimum observable visual area of ​​multiple target power equipment based on the mapping relationship between the inspection question and answer unit and the three-dimensional geometric model of multiple target power equipment in the power system.

[0219] The determination module 606 is used to obtain the inspection trajectory point sequence of the inspection robot and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0220] The filtering module 608 is used to acquire the imaging pose data of the inspection robot and, under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0221] The evaluation module 610 is used to determine the surface imaging pose of the minimum observable visual region of the target in each frame of the image data based on the target imaging pose data, perform visual recognition analysis on the surface imaging pose based on the external ambient lighting conditions, obtain the visual recognition state of the minimum observable visual region of the target, evaluate the image based on the visual recognition state, and obtain the image quality evaluation result.

[0222] In one embodiment, the determining module 606 is further configured to: construct a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points; perform Boolean intersection processing on the ray beam family to obtain a target ray subfamily; obtain the distribution of the projection points of the target ray subfamily in the local tangent plane coordinate system, perform coverage density analysis on the projection point distribution to obtain the ray direction cluster centers covering the surface of the minimum observable visual region; use the direction vector of the ray direction cluster centers as the optical axis of the imaging device of the inspection robot, and construct a target imaging cone based on the optical axis; obtain the actual optical cone of the imaging device, and perform geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraints of the minimum observable visual region.

[0223] In one embodiment, the determining module 606 is further configured to perform geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the target imaging direction of the minimum observable visual region; obtain the rotation component of the inspection trajectory point in the target imaging direction; based on the rotation component, transform and synthesize the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis; calculate the three-dimensional angle deviation between the target pointing vector and the desired observation vector, where the desired observation vector is the vector pointing from the center point of the minimum observable visual region to the target inspection trajectory point corresponding to the optical axis; perform motion adjustability judgment on the target inspection trajectory point according to the three-dimensional angle deviation to obtain the motion adjustability judgment result; based on the motion adjustability judgment result, select candidate imaging sites from the target inspection trajectory points; construct a spatial bidirectional reachability constraint graph based on the candidate imaging sites; and, if the spatial bidirectional reachability constraint graph passes the backtracking verification, obtain the imaging feasibility constraint of the minimum observable visual region from the spatial bidirectional reachability constraint graph.

[0224] In one embodiment, the evaluation module 610 is further configured to: determine the minimum observable visual region of the target in the image based on the target imaging pose data; construct the imaging ray cone of the imaging device of the inspection robot; perform spatial intersection processing on the imaging ray cone and the minimum observable visual region of the target to obtain the surface intersection region; perform orthogonal projection analysis on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection contour; obtain the physical size of the photosensitive surface of the imaging sensor and the pixel array arrangement parameters; based on the physical size of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour, obtain the minimum continuous pixel covering the orthogonal projection contour on the photosensitive surface through coordinate system alignment transformation processing; and determine the surface imaging pose of the minimum observable visual region of the target based on the positional correspondence between the minimum continuous pixel and the optical center of the imaging device in different coordinate systems.

[0225] In one embodiment, the evaluation module 610 is further configured to, based on the spatial orientation of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture under external ambient lighting conditions, obtain the incident direction vector of the main light source direction vector in the minimum observable visual region through coordinate system transformation; calculate the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device to obtain the illumination observation angle of the minimum observable visual region; perform shadow occlusion analysis based on the illumination observation angle and the local differential geometric properties of the minimum observable visual region of the target to obtain the shadow occlusion analysis result of the minimum observable visual region of the target; and perform visual recognition analysis based on the shadow occlusion analysis result to obtain the visual recognition state of the minimum observable visual region of the target.

[0226] In one embodiment, the evaluation module 610 is further configured to determine, based on the shadow occlusion analysis results, whether there is a local feature occlusion effect on the surface of the minimum observable visual area of ​​the target; if there is a local feature occlusion effect, then the visual identification state of the minimum observable visual area of ​​the target is determined to be unidentifiable; if there is no local feature occlusion effect, then based on the imaging tilt angle parameter of the surface imaging posture, it is determined whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than a preset grazing avoidance angle; if the three-dimensional angle is greater than the preset grazing avoidance angle, then the visual identification state of the minimum observable visual area is determined to be identifiable; if the three-dimensional angle is less than or equal to the preset grazing avoidance angle, then the visual identification state of the minimum observable visual area of ​​the target is determined to be unidentifiable.

[0227] Each module in the aforementioned power system image data quality assessment device 600 can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0228] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores image data of the power system collected by the inspection robot. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for evaluating the image data quality of a power system.

[0229] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0230] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0231] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0232] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0233] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0234] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0235] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0236] In one embodiment, when the processor executes the computer program, it further implements the following steps: constructing a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points; performing Boolean intersection processing on the ray beam family to obtain a target ray subfamily; obtaining the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, performing coverage density analysis on the projection point distribution to obtain the ray direction cluster center covering the surface of the minimum observable visual region; using the direction vector of the ray direction cluster center as the optical axis of the imaging device of the inspection robot, and constructing a target imaging cone based on the optical axis; obtaining the actual optical cone of the imaging device, and performing geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual region.

[0237] In one embodiment, when the processor executes the computer program, it further implements the following steps: geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the target imaging direction of the minimum observable visual area; obtaining the rotation component of the inspection trajectory point along the target imaging direction; and, based on the rotation component, transforming and synthesizing the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis; calculating the three-dimensional angle deviation between the target pointing vector and the desired observation vector, where the desired observation vector is the vector pointing from the center point of the minimum observable visual area to the target inspection trajectory point corresponding to the optical axis; performing motion adjustability judgment on the target inspection trajectory point according to the three-dimensional angle deviation to obtain the motion adjustability judgment result; and, based on the motion adjustability judgment result, selecting candidate imaging sites from the target inspection trajectory points; constructing a spatial bidirectional reachability constraint graph based on the candidate imaging sites; and, if the spatial bidirectional reachability constraint graph passes backtracking verification, obtaining the imaging feasibility constraint of the minimum observable visual area from the spatial bidirectional reachability constraint graph.

[0238] In one embodiment, when the processor executes the computer program, it further performs the following steps: based on the target imaging pose data, it determines the minimum observable visual region of the target in the image and constructs the imaging ray cone of the imaging device of the inspection robot; it performs spatial intersection processing on the imaging ray cone and the minimum observable visual region of the target to obtain the surface intersection region; it performs orthogonal projection analysis on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection contour; it obtains the physical size of the photosensitive surface of the imaging sensor and the pixel array arrangement parameters, and based on the physical size of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour, it obtains the minimum continuous pixel covering the orthogonal projection contour on the photosensitive surface through coordinate system alignment transformation processing; based on the positional correspondence between the minimum continuous pixel and the optical center of the imaging device in different coordinate systems, it determines the surface imaging pose of the minimum observable visual region of the target.

[0239] In one embodiment, when the processor executes the computer program, it further performs the following steps: based on the spatial pointing of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture in the external ambient lighting conditions, the incident direction vector of the main light source direction vector in the minimum observable visual area is obtained through coordinate system transformation; the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device is calculated to obtain the illumination observation angle of the minimum observable visual area; shadow occlusion analysis is performed based on the illumination observation angle and the local differential geometric properties of the minimum observable visual area of ​​the target to obtain the shadow occlusion analysis result of the minimum observable visual area of ​​the target; and visual recognition analysis is performed based on the shadow occlusion analysis result to obtain the visual recognition state of the minimum observable visual area of ​​the target.

[0240] In one embodiment, when the processor executes the computer program, it further implements the following steps: based on the shadow occlusion analysis results, it determines whether there is a local feature occlusion effect on the surface of the minimum observable visual area of ​​the target; if there is a local feature occlusion effect, it determines that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state; if there is no local feature occlusion effect, it determines whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than a preset grazing avoidance angle based on the imaging tilt angle parameter of the surface imaging posture; if the three-dimensional angle is greater than the preset grazing avoidance angle, it determines that the visual identification state of the minimum observable visual area is an identifiable state; if the three-dimensional angle is less than or equal to the preset grazing avoidance angle, it determines that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state.

[0241] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0242] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0243] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0244] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0245] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0246] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0247] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: constructing a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points; performing Boolean intersection processing on the ray beam family to obtain a target ray subfamily; obtaining the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, performing coverage density analysis on the projection point distribution to obtain the ray direction cluster center covering the surface of the minimum observable visual region; using the direction vector of the ray direction cluster center as the optical axis of the imaging device of the inspection robot, and constructing a target imaging cone based on the optical axis; obtaining the actual optical cone of the imaging device, and performing geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual region.

[0248] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the target imaging direction of the minimum observable visual area; obtaining the rotation component of the inspection trajectory point along the target imaging direction; and, based on the rotation component, transforming and synthesizing the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis; calculating the three-dimensional angle deviation between the target pointing vector and the desired observation vector, where the desired observation vector is the vector pointing from the center point of the minimum observable visual area to the target inspection trajectory point corresponding to the optical axis; judging the motion adjustability of the target inspection trajectory point based on the three-dimensional angle deviation to obtain the motion adjustability judgment result; and, based on the motion adjustability judgment result, selecting candidate imaging sites from the target inspection trajectory points; constructing a spatial bidirectional reachability constraint graph based on the candidate imaging sites; and, if the spatial bidirectional reachability constraint graph passes backtracking verification, obtaining the imaging feasibility constraint of the minimum observable visual area from the spatial bidirectional reachability constraint graph.

[0249] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the minimum observable visual region of the target in the image based on the target imaging pose data, and constructing the imaging ray cone of the imaging device of the inspection robot; performing spatial intersection processing on the imaging ray cone and the minimum observable visual region of the target to obtain the surface intersection region; performing orthogonal projection analysis on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection contour; obtaining the physical size of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device, and obtaining the minimum continuous pixels covering the orthogonal projection contour on the photosensitive surface through coordinate system alignment transformation processing based on the physical size of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour; and determining the surface imaging pose of the minimum observable visual region of the target based on the positional correspondence between the minimum continuous pixels and the optical center of the imaging device in different coordinate systems.

[0250] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the spatial pointing of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture in the external ambient lighting conditions, the incident direction vector of the main light source direction vector in the minimum observable visual area is obtained through coordinate system transformation; the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device is calculated to obtain the illumination observation angle of the minimum observable visual area; shadow occlusion analysis is performed based on the illumination observation angle and the local differential geometric properties of the minimum observable visual area of ​​the target to obtain the shadow occlusion analysis result of the minimum observable visual area of ​​the target; and visual recognition analysis is performed based on the shadow occlusion analysis result to obtain the visual recognition state of the minimum observable visual area of ​​the target.

[0251] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the shadow occlusion analysis results, determine whether there is a local feature occlusion effect on the surface of the minimum observable visual area of ​​the target; if there is a local feature occlusion effect, determine that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state; if there is no local feature occlusion effect, determine whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than a preset grazing avoidance angle based on the imaging tilt angle parameter of the surface imaging posture; if the three-dimensional angle is greater than the preset grazing avoidance angle, determine that the visual identification state of the minimum observable visual area is an identifiable state; if the three-dimensional angle is less than or equal to the preset grazing avoidance angle, determine that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state.

[0252] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0253] The inspection robot collects image data of the power system, which includes multiple frames of images.

[0254] Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​multiple target power equipment is determined.

[0255] Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence.

[0256] Acquire the imaging pose data of the inspection robot, and under the constraints of imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data.

[0257] For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined based on the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality assessment result.

[0258] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: constructing a family of ray beams pointing from multiple inspection trajectory points to the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and multiple inspection trajectory points; performing Boolean intersection processing on the ray beam family to obtain a target ray subfamily; obtaining the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, performing coverage density analysis on the projection point distribution to obtain the ray direction cluster center covering the surface of the minimum observable visual region; using the direction vector of the ray direction cluster center as the optical axis of the imaging device of the inspection robot, and constructing a target imaging cone based on the optical axis; obtaining the actual optical cone of the imaging device, and performing geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual region.

[0259] In one embodiment, when the computer program is executed by the processor, it further implements the following steps: geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the target imaging direction of the minimum observable visual area; obtaining the rotation component of the inspection trajectory point along the target imaging direction; and, based on the rotation component, transforming and synthesizing the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis; calculating the three-dimensional angle deviation between the target pointing vector and the desired observation vector, where the desired observation vector is the vector pointing from the center point of the minimum observable visual area to the target inspection trajectory point corresponding to the optical axis; judging the motion adjustability of the target inspection trajectory point based on the three-dimensional angle deviation to obtain the motion adjustability judgment result; and, based on the motion adjustability judgment result, selecting candidate imaging sites from the target inspection trajectory points; constructing a spatial bidirectional reachability constraint graph based on the candidate imaging sites; and, if the spatial bidirectional reachability constraint graph passes backtracking verification, obtaining the imaging feasibility constraint of the minimum observable visual area from the spatial bidirectional reachability constraint graph.

[0260] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: determining the minimum observable visual region of the target in the image based on the target imaging pose data, and constructing the imaging ray cone of the imaging device of the inspection robot; performing spatial intersection processing on the imaging ray cone and the minimum observable visual region of the target to obtain the surface intersection region; performing orthogonal projection analysis on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection contour; obtaining the physical size of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device, and obtaining the minimum continuous pixels covering the orthogonal projection contour on the photosensitive surface through coordinate system alignment transformation processing based on the physical size of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour; and determining the surface imaging pose of the minimum observable visual region of the target based on the positional correspondence between the minimum continuous pixels and the optical center of the imaging device in different coordinate systems.

[0261] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the spatial pointing of the main light source direction vector and the imaging tilt angle parameters of the surface imaging posture in the external ambient lighting conditions, the incident direction vector of the main light source direction vector in the minimum observable visual area is obtained through coordinate system transformation; the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device is calculated to obtain the illumination observation angle of the minimum observable visual area; shadow occlusion analysis is performed based on the illumination observation angle and the local differential geometric properties of the minimum observable visual area of ​​the target to obtain the shadow occlusion analysis result of the minimum observable visual area of ​​the target; and visual recognition analysis is performed based on the shadow occlusion analysis result to obtain the visual recognition state of the minimum observable visual area of ​​the target.

[0262] In one embodiment, when the computer program is executed by the processor, it further performs the following steps: based on the shadow occlusion analysis results, determine whether there is a local feature occlusion effect on the surface of the minimum observable visual area of ​​the target; if there is a local feature occlusion effect, determine that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state; if there is no local feature occlusion effect, determine whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than a preset grazing avoidance angle based on the imaging tilt angle parameter of the surface imaging posture; if the three-dimensional angle is greater than the preset grazing avoidance angle, determine that the visual identification state of the minimum observable visual area is an identifiable state; if the three-dimensional angle is less than or equal to the preset grazing avoidance angle, determine that the visual identification state of the minimum observable visual area of ​​the target is an unidentifiable state.

[0263] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0264] 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, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0265] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0266] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for assessing the quality of image data in a power system, characterized in that, The method includes: The inspection robot collects image data of the power system, which includes multiple frames of images. Based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of multiple target power equipment in the power system, the minimum observable visual area of ​​the multiple target power equipment is determined. Obtain the inspection trajectory point sequence of the inspection robot, and determine the imaging feasibility constraints of the minimum observable visual area based on the spatial relative relationship between the minimum observable visual area and the inspection trajectory point sequence. Acquire the imaging pose data of the inspection robot, and under the constraints of the imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data; For each frame of the image data, the surface imaging pose of the minimum observable visual region of the target in the image is determined according to the target imaging pose data. Based on the external ambient lighting conditions, the surface imaging pose is visually identified and analyzed to obtain the visual identification state of the minimum observable visual region of the target. Based on the visual identification state, the image is evaluated to obtain the image quality evaluation result of the image.

2. The method according to claim 1, characterized in that, The inspection trajectory point sequence includes multiple inspection trajectory points; determining the imaging feasibility constraints of the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence includes: Based on the spatial relative relationship between the minimum observable visual region and the plurality of inspection trajectory points, a family of ray beams pointing from the plurality of inspection trajectory points to the minimum observable visual region is constructed; Perform Boolean intersection processing on the aforementioned ray beam family to obtain the target ray subfamily; Obtain the projection point distribution of the target ray subfamily in the local tangent plane coordinate system, perform coverage density analysis on the projection point distribution, and obtain the ray direction cluster centers covering the surface of the minimum observable visual area; The direction vector of the cluster center of the ray direction is used as the optical axis of the imaging device of the inspection robot, and a target imaging cone is constructed based on the optical axis; The actual optical cone of the imaging device is obtained, and the target imaging cone is geometrically isomorphically matched with the actual optical cone to obtain the imaging feasibility constraint of the minimum observable visual region.

3. The method according to claim 2, characterized in that, The step of performing geometric isomorphic matching between the target imaging cone and the actual optical cone to obtain the imaging feasibility constraints for the minimum observable visual region includes: Based on the geometric isomorphic matching between the target imaging cone and the actual optical cone, the target imaging direction of the minimum observable visual region is obtained. Obtain the rotation component of the inspection trajectory point in the target imaging direction. Based on the rotation component, transform and synthesize the preset installation bias matrix of the inspection robot to obtain the target pointing vector of the optical axis. Calculate the three-dimensional angle deviation between the target pointing vector and the desired observation vector, wherein the desired observation vector is a vector pointing from the center point of the minimum observable visual area to the target inspection trajectory point corresponding to the optical axis; Based on the three-dimensional included angle deviation, the motion adjustability of the target inspection trajectory point is judged to obtain the motion adjustability judgment result. Based on the motion adjustability judgment result, candidate imaging sites are selected from the target inspection trajectory points. Based on the candidate imaging sites, a spatial bidirectional reachability constraint graph is constructed. If the spatial bidirectional reachability constraint graph passes the backtrackability verification, the imaging feasibility constraint of the minimum observable visual region is obtained from the spatial bidirectional reachability constraint graph.

4. The method according to claim 1, characterized in that, The step of determining the surface imaging pose of the minimum observable visual region of the target in the image based on the target imaging pose data includes: Based on the target imaging pose data, the minimum observable visual area of ​​the target is determined in the image, and the imaging ray cone of the imaging device of the inspection robot is constructed. Spatial intersection processing is performed on the imaging ray cone and the minimum observable visual region of the target to obtain the surface intersection region; Orthogonal projection analysis is performed on the surface intersection region and the optical axis of the imaging device to obtain the orthogonal projection profile; The physical dimensions of the photosensitive surface and the pixel array arrangement parameters of the imaging sensor of the imaging device are obtained. Based on the physical dimensions of the photosensitive surface, the pixel array arrangement parameters and the orthogonal projection contour, the minimum continuous pixels covering the orthogonal projection contour on the photosensitive surface are obtained through coordinate system alignment transformation. Based on the positional correspondence between the minimum continuous pixel and the optical center of the imaging device in different coordinate systems, the surface imaging pose of the minimum observable visual area of ​​the target is determined.

5. The method according to claim 1, characterized in that, The step of performing visual recognition analysis on the surface imaging posture based on external ambient lighting conditions to obtain the visual recognition state of the minimum observable visual area of ​​the target includes: Based on the spatial orientation of the main light source direction vector under the external ambient lighting conditions and the imaging tilt angle parameter of the surface imaging posture, the incident direction vector of the main light source direction vector in the minimum observable visual area is obtained through coordinate system transformation. Calculate the mirror deviation angle between the incident direction vector and the observation direction vector of the optical axis of the imaging device to obtain the illumination observation angle of the minimum observable visual area; Shadow occlusion analysis is performed based on the illumination observation angle and the local differential geometric properties of the minimum observable visual region of the target to obtain the shadow occlusion analysis results of the minimum observable visual region of the target. Based on the shadow occlusion analysis results, visual recognition analysis is performed to obtain the visual recognition status of the minimum observable visual area of ​​the target.

6. The method according to claim 5, characterized in that, The visual recognition states include unrecognizable states and recognizable states; The step of performing visual recognition analysis based on the shadow occlusion analysis results to obtain the visual recognition state of the minimum observable visual region of the target includes: Based on the shadow occlusion analysis results, determine whether there is a local feature occlusion effect on the surface of the minimum observable visual area of ​​the target; In the presence of local feature occlusion effect, the visual recognition state of the minimum observable visual region of the target is determined to be an unrecognizable state. In the absence of local feature occlusion effect, based on the imaging tilt angle parameter of the surface imaging posture, it is determined whether the three-dimensional angle between the optical axis of the imaging device and the normal vector of the local tangent plane is greater than the preset grazing avoidance angle. If the three-dimensional angle is greater than the preset grazing avoidance angle, the visual recognition state of the minimum observable visual area is determined to be recognizable. If the three-dimensional angle is less than or equal to the preset grazing avoidance angle, the visual recognition state of the minimum observable visual area of ​​the target is determined to be unrecognizable.

7. An image data quality assessment device for a power system, characterized in that, The device includes: The acquisition module is used to acquire image data of the power system collected by the inspection robot, the image data including multiple frames of images; The mapping module is used to determine the minimum observable visual area of ​​the multiple target power devices based on the mapping relationship between the inspection question-and-answer unit and the three-dimensional geometric model of the multiple target power devices in the power system. The determination module is used to acquire the inspection trajectory point sequence of the inspection robot and determine the imaging feasibility constraints of the minimum observable visual region based on the spatial relative relationship between the minimum observable visual region and the inspection trajectory point sequence. The filtering module is used to acquire the imaging pose data of the inspection robot and, under the constraints of the imaging feasibility constraints, filter out the target imaging pose data from the imaging pose data. The evaluation module is used to determine the surface imaging pose of the minimum observable visual region of the target in each frame of the image data based on the target imaging pose data, perform visual recognition analysis on the surface imaging pose based on the external ambient lighting conditions, obtain the visual recognition state of the minimum observable visual region of the target, evaluate the image based on the visual recognition state, and obtain the image quality evaluation result of the image.

8. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images