Method for panoramic acquisition of a nuclear power plant

Abstract

This invention provides a method for panoramic data acquisition of nuclear power plants, comprising the following steps: Step S1: Acquire spatial reference data of the nuclear power plant area, and calculate and generate an optimal acquisition path covering the target acquisition area using a path planning algorithm; Step S2: Acquire on-site images and pose data acquired according to the optimal acquisition path; Step S3: Process the on-site images based on the pose data to generate a panoramic image. Through the integrated collaboration of intelligent path planning, multi-source fusion positioning, and pose constraint processing, the method achieves standardization and efficiency in data acquisition operations, attachment of spatial information to the panoramic image, and full-process automation of acquisition and processing.

Description

Technical Field

[0001] This invention relates to the field of image acquisition and processing technology, and more specifically, to a method for panoramic acquisition of nuclear power plants. Background Technology

[0002] Currently, the acquisition of panoramic image data for nuclear power plants mainly relies on traditional, manual, and non-standardized on-site operation modes. Operators need to carry general-purpose panoramic cameras by hand or on their backs, and plan routes, select points, and take pictures manually in complex plant environments based on experience. This mode has fundamental defects such as extremely low operation efficiency and standardization, serious lack of spatial location information in panoramic images, poor adaptability to the on-site environment, and fragmented data processing flow. It is difficult to meet the requirements of building a high-fidelity, real-time updated real-scene twin platform for efficient, accurate, and standardized data sources. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method for panoramic data acquisition of nuclear power plants, addressing the technical shortcomings of existing technologies such as low operational efficiency and low standardization.

[0004] The technical solution adopted in this application to solve its technical problem is: a method for constructing a panoramic data acquisition of a nuclear power plant, including the following steps: Step S1: Obtain spatial reference data of the nuclear power plant area, and generate the optimal acquisition path covering the target acquisition area through path planning algorithm; Step S2: Obtain the on-site images and pose data acquired according to the optimal acquisition path; Step S3: Based on the pose data, process the on-site images to generate a panoramic image.

[0005] Further, the spatial reference data includes one or more of two-dimensional plan views, lightweight BIM models, and point cloud data; step S1 includes: step S11: generating a candidate collection point set in one or more of two-dimensional plan views, lightweight BIM models, and point cloud data according to preset collection density parameters; step S12: when the spatial reference data includes a lightweight BIM model, obtaining room function labels in the lightweight BIM model, identifying key areas according to the room function labels, assigning a first weight value to candidate points located in key areas, and assigning a second weight value to candidate points located in non-key areas, wherein the first weight value is greater than the second weight value; Step S13: Based on A The algorithm calculates the actual passable path length between each point in the candidate collection point set, and assigns a penalty coefficient to the path segment passing through the radiation control area in combination with safety rules to construct a cost matrix; Step S14: Using the Traveling Salesman Problem optimization algorithm, the optimal collection path sequence is calculated and output from the candidate collection point set according to the cost matrix and the weight of each candidate point.

[0006] Furthermore, the acquisition density parameters include the acquisition spacing along the channel direction and the grid point spacing within the equipment room; step S11 includes: determining whether the acquisition area type in the two-dimensional plan view, lightweight BIM model, and point cloud data is a channel area or an equipment room area; when the acquisition area type is a channel area, generating candidate acquisition points within the channel area according to the acquisition spacing; when the acquisition area type is an equipment room area, generating grid-like candidate acquisition points within the equipment room area according to the grid point spacing; summarizing the candidate acquisition points and the grid-like candidate acquisition points to obtain a candidate acquisition point set.

[0007] Further, step S13 includes: constructing a navigation grid based on one or more of a two-dimensional plan view, a lightweight BIM model, or point cloud data; for any two candidate points in the candidate collection point set, using A... The algorithm calculates the actual passable path length from the starting point to the ending point on the navigation grid, which is used as the passage cost between the two points; identifies the radiation control zone in the navigation grid, and adds a preset penalty coefficient to the passage cost for path segments that pass through the radiation control zone; and constructs a cost matrix based on the passage costs between all candidate points.

[0008] Further, step S14 includes: adjusting the passage cost between any two points according to the following formula: in, Let be the original travel cost from candidate point i to candidate point j. The revised passage cost. and Let i and j be the weights of candidate point i and candidate point j, respectively. and The preset adjustment coefficient is used; it is determined whether the number of candidate points in the candidate collection point set exceeds a preset scale threshold; when the number of candidate points in the candidate collection point set does not exceed the preset scale threshold, all candidate points in the candidate collection point set are taken as nodes of the traveling salesman problem, and the corrected passage cost is applied. As the edge cost between nodes, the optimal traversal order that minimizes the total correction cost is found. When the number of candidate points in the candidate collection point set exceeds a preset scale threshold, a hierarchical planning strategy is used to find the optimal traversal order. Based on the optimal traversal order, the optimal collection path sequence is output. The optimal collection path sequence includes an ordered collection point sequence and the walking path connecting each collection point.

[0009] Furthermore, the hierarchical planning strategy for finding the optimal traversal order includes: dividing the data collection area into several logical sub-regions based on factory buildings or floors; within each logical sub-region, using candidate points within the logical sub-region as nodes, and using the corrected passage cost... As edge costs, solve the Traveling Salesman Problem separately to obtain the locally optimal path sequence within each logical sub-region; treat each logical sub-region as a super point, and use the travel cost between sub-regions as edge costs to solve the access order between super points; concatenate the locally optimal path sequences within each logical sub-region according to the access order between super points to obtain the optimal traversal order.

[0010] Furthermore, the method also includes: obtaining the navigation mode selected by the user; when the navigation mode is a two-dimensional map mode, obtaining the current location information, rendering the optimal acquisition path sequence on the two-dimensional map layer, and displaying the path between the current location and the next acquisition point in the optimal acquisition path sequence on the two-dimensional map layer; when the navigation mode is an augmented reality mode, obtaining the real-scene preview image stream, the current location information, and the attitude information, calculating the superposition position of the virtual navigation mark in the real-scene preview image stream based on the current location information and the attitude information, rendering the virtual navigation mark on the real-scene preview image stream, and superimposing and displaying prompts for the standard shooting height and level on the real-scene preview image stream.

[0011] Further, the pose number includes a positional confidence parameter, and step S3 includes: Step S31: Based on the positional confidence parameter, remove the field images whose positional confidence is lower than a preset threshold to generate a first candidate image set; acquire ambient lighting data, and based on the ambient lighting data, remove field images whose image quality does not meet the preset standard from the first candidate image set to generate a second candidate image set; Step S32: Group the field images in the second candidate image set according to the optimal acquisition path to generate a grouped field image sequence; perform image enhancement processing on the grouped field image sequence to generate a preprocessed field image sequence; Step S33: Based on the pose data, stitch the preprocessed field image sequence to generate a panoramic image in equidistant cylindrical projection format; Step S34: Based on the pose data, generate a spatial information file.

[0012] Further, the pose data includes three-dimensional coordinates and attitude angles; step S33 includes: acquiring the preprocessed field image sequence and the corresponding three-dimensional coordinates and attitude angles; calculating the initial spherical projection transformation matrix for each preprocessed field image based on the three-dimensional coordinates and attitude angles; limiting the search range of feature points from the entire image to a preset window centered on the pose prediction position, performing feature matching on the preprocessed field image sequence, and generating preliminary matching results; performing bundle adjustment on the preliminary matching results, globally optimizing the camera pose and three-dimensional point coordinates, eliminating accumulated errors, and generating optimized stitching results; and generating a panoramic image in equidistant cylindrical projection format based on the optimized stitching results.

[0013] Further, step S34 includes: acquiring the three-dimensional coordinates and attitude angles corresponding to the panoramic image in equidistant cylindrical projection format; organizing the three-dimensional coordinates and attitude angles according to a preset format to generate a spatial information file; and outputting the panoramic image and spatial information file in equidistant cylindrical projection format as the result for three-dimensional spatial fusion.

[0014] The method for panoramic data acquisition of nuclear power plants according to the present invention has the following beneficial effects: It includes: Step S1: acquiring spatial reference data of the nuclear power plant area, and calculating and generating an optimal acquisition path covering the target acquisition area using a path planning algorithm; Step S2: acquiring on-site images and pose data obtained according to the optimal acquisition path; Step S3: processing the on-site images based on the pose data to generate a panoramic image. Through the integrated collaboration of intelligent path planning, multi-source fusion positioning, and pose constraint processing, the method achieves standardization and efficiency in data acquisition operations, attachment of spatial information to the panoramic image, and full-process automation of acquisition and processing. Attached Figure Description

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a logic flowchart of a method for panoramic data acquisition of a nuclear power plant. Detailed Implementation

[0016] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the invention are now described in detail with reference to the accompanying drawings. In the following description, specific details such as particular structures and techniques are set forth for illustrative purposes and not for limitation, so as to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known devices, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0017] like Figure 1 As shown, Figure 1 This is a logic flowchart of a method for panoramic data acquisition of a nuclear power plant, as described in one embodiment. The provided method for panoramic data acquisition of a nuclear power plant includes the following steps: Step S1: Obtain spatial reference data of the nuclear power plant area, and generate the optimal acquisition path covering the target acquisition area through path planning algorithm; In this step, the spatial reference data includes one or more of the following: a two-dimensional plan view of the nuclear power plant site, a lightweight BIM model, and laser point cloud data; the optimal acquisition path includes a candidate acquisition point set generated according to the region type, the optimal passage path considering radiation control zones and safety rules, and an ordered sequence of acquisition points weighted by key regions.

[0018] Step S2: Obtain the on-site images and pose data acquired according to the optimal acquisition path; In this step, the on-site images are high-definition panoramic images of the nuclear power plant's equipment, structures, and passageways taken at designated locations along the optimal acquisition path; the pose data are the three-dimensional spatial coordinates (X, Y, Z), camera orientation, pitch angle, roll angle, and positional confidence parameters acquired synchronously at the moment of capture.

[0019] Step S3: Based on the pose data, process the on-site images to generate a panoramic image.

[0020] In this step, the on-site images are processed by filtering images based on location confidence and illumination quality, grouping them according to the optimal acquisition path, image enhancement, feature matching and cluster adjustment optimization under pose constraints; and generating a panoramic image in the standard panoramic format of equidistant cylindrical projection with spatial pose information, which can be directly used for fusion of 3D reality twin platforms.

[0021] Furthermore, the spatial reference data includes one or more of the following: two-dimensional plan view, lightweight BIM model, and point cloud data; step S1 includes: Step S11: Generate a candidate collection point set from one or more of the following: two-dimensional plan view, lightweight BIM model and point cloud data, according to the preset collection density parameters; Specifically, the system utilizes 2D floor plans of the nuclear power plant area, lightweight BIM models, or laser-scanned point cloud data as spatial references. Point cloud data, acquired through laser scanning, provides centimeter-level precision 3D spatial geometric information for visibility analysis, determining whether there is visual obstruction between two points and ensuring sufficient overlap between adjacent acquisition points. The BIM model, derived from the nuclear power plant design model, provides semantic information on equipment locations, room functions, piping layouts, and safety zones, used to identify critical areas and avoid radiation control zones during path planning. The 2D floor plan serves as a basic spatial reference, used when BIM or point cloud data is missing, and a simplified 2D topology map is constructed by identifying wall, door, and passageway elements for path planning.

[0022] Furthermore, the acquisition density parameters include the acquisition spacing along the channel direction and the grid point spacing within the equipment room; step S11 includes: determining whether the acquisition area type in the two-dimensional plan view, lightweight BIM model, and point cloud data is a channel area or an equipment room area; when the acquisition area type is a channel area, generating candidate acquisition points within the channel area according to the acquisition spacing; when the acquisition area type is an equipment room area, generating grid-like candidate acquisition points within the equipment room area according to the grid point spacing; summarizing the candidate acquisition points and the grid-like candidate acquisition points to obtain a candidate acquisition point set.

[0023] Specifically, linear candidate collection points are generated along the channel area at intervals of 3–5 meters, and uniformly distributed grid-like candidate collection points are generated in the equipment inter-area area at preset grid intervals. The linear collection points and grid collection points are merged to form a complete set of candidate collection points covering the target area.

[0024] Step S12: When the spatial reference data contains a lightweight BIM model, obtain the room function labels in the lightweight BIM model, identify key areas based on the room function labels, assign a first weight value to candidate points located in key areas, and assign a second weight value to candidate points located in non-key areas, with the first weight value being greater than the second weight value. Specifically, the system reads the room function attribute fields from the BIM model, marking main equipment rooms, valve operation areas, and inspection route points as critical areas, and ordinary passages and auxiliary rooms as non-critical areas. For example, candidate points in critical areas are assigned a first weight value of no less than 1.5, and candidate points in non-critical areas are assigned a second weight value of no more than 1.0. The BIM model is a pre-built and stored standardized building information model of a nuclear power plant. The model contains standardized attribute information such as room function attribute fields, area type fields, critical area marking fields, equipment layout fields, and inspection route fields, which can be directly read and used for area division, critical area identification, candidate collection point weight assignment, and collection path planning.

[0025] Step S13: Based on A The algorithm calculates the actual passable path length between each point in the candidate collection point set, and assigns a penalty coefficient to the path segment that passes through the radiation control area in combination with safety rules to construct a cost matrix; Specifically, a navigation grid is constructed based on spatial reference data, using A The algorithm calculates the actual passable path length between any two candidate acquisition points on the navigation grid; using this path length as the basic passage cost, it identifies radiation control zones within the navigation grid and adds a penalty coefficient (e.g., not less than 2.0) to path segments crossing these zones, obtaining the corrected passage cost; it then iterates through all candidate acquisition points, constructing a cost matrix between each pair of points. The navigation grid is generated by rasterizing the spatial reference data into passable areas, marking walls, equipment, and obstacles as impassable areas, and passageways, rooms, and passable areas as passable grids, forming a basic passable spatial model for path search.

[0026] Further, step S13 includes: constructing a navigation grid based on one or more of a two-dimensional plan view, a lightweight BIM model, or point cloud data; for any two candidate points in the candidate collection point set, using A... The algorithm calculates the actual passable path length from the starting point to the ending point on the navigation grid, which is used as the passage cost between the two points; identifies the radiation control zone in the navigation grid, and adds a preset penalty coefficient to the passage cost for path segments that pass through the radiation control zone; and constructs a cost matrix based on the passage costs between all candidate points.

[0027] Specifically, the navigation grid is generated by rasterizing spatial reference data into passable areas, marking walls, equipment, and obstacles as impassable areas; A The algorithm takes the starting point and ending point as input, uses the passable grid as the search space, and outputs the shortest passable path length; the radiation control area is identified according to the preset safety partition layer, and the penalty coefficient is directly added in the path cost calculation.

[0028] Step S14: Using the Traveling Salesman Problem optimization algorithm, the optimal collection path sequence is calculated and output from the candidate collection point set based on the cost matrix and the weight of each candidate point.

[0029] Specifically, candidate collection points are used as nodes in the Traveling Salesman Problem, and the corrected passage cost is used as the edge cost between nodes. The objective function is constructed by combining the weights of each point. A genetic algorithm is used to solve for the minimum value of the objective function, and the optimal path sequence for traversing all candidate collection points is obtained.

[0030] Path planning algorithm based on A To optimize the Traveling Salesman Problem (TSP), this multi-objective optimization solution combines preset data acquisition density parameters, key area weights, safety rules, and visibility analysis to calculate the optimal data acquisition path sequence that covers the target area, has the shortest total travel distance, and sufficient overlap between adjacent data acquisition points. Specifically, it first generates a candidate data acquisition point set based on BIM / point cloud / CAD spatial data and density parameters, then assigns high weights to key area points based on BIM room function labels, and finally... The algorithm calculates the actual traversable cost on the navigation grid and applies a penalty coefficient to the path in the radiation control area to construct a cost matrix. Then, it abstracts the problem into a weighted traveling salesman problem, prioritizes traversing high-weight points by modifying the cost function, and uses a hierarchical planning strategy to process large-scale point sets. Finally, it outputs ordered collection points and the optimal walking path and sends them to the field terminal. The navigation interface pushes the planning results to a ruggedized tablet computer, providing two modes: two-dimensional map and augmented reality (AR). The two-dimensional mode can intuitively display the path and current location, while the AR mode calls the camera to overlay virtual arrows and destination markers on the real scene, guiding the operator to accurately reach the preset point and prompting the standard shooting height and horizontal posture.

[0031] The hierarchical planning strategy divides the nuclear power plant into regions based on the building, floors, and functional zones. It decomposes the large-scale collection points into multiple sub-region point sets, first solving the optimal path within each sub-region, then determining the optimal access order between sub-regions, and finally connecting the paths of each sub-region end to end to form a globally optimal path covering all collection points.

[0032] Further, step S14 includes: adjusting the passage cost between any two points according to the following formula: in, Let be the original travel cost from candidate point i to candidate point j. The revised passage cost. and Let i and j be the weights of candidate point i and candidate point j, respectively. and The preset adjustment coefficient is used; it is determined whether the number of candidate points in the candidate collection point set exceeds a preset scale threshold; when the number of candidate points in the candidate collection point set does not exceed the preset scale threshold, all candidate points in the candidate collection point set are taken as nodes of the traveling salesman problem, and the corrected passage cost is applied. As the edge cost between nodes, the optimal traversal order that minimizes the total correction cost is found. When the number of candidate points in the candidate collection point set exceeds a preset scale threshold, a hierarchical planning strategy is used to find the optimal traversal order. Based on the optimal traversal order, the optimal collection path sequence is output. The optimal collection path sequence includes an ordered collection point sequence and the walking path connecting each collection point.

[0033] For example, The value range is 0.3–0.7, and the scale threshold is set to 200 points; if the value is less than the threshold, an exact algorithm is used to solve the problem, and if the value is greater than the threshold, a hierarchical programming strategy is used to solve the problem.

[0034] Furthermore, the hierarchical planning strategy for finding the optimal traversal order includes: dividing the data collection area into several logical sub-regions based on factory buildings or floors; within each logical sub-region, using candidate points within the logical sub-region as nodes, and using the corrected passage cost... As edge costs, solve the Traveling Salesman Problem separately to obtain the locally optimal path sequence within each logical sub-region; treat each logical sub-region as a super point, and use the travel cost between sub-regions as edge costs to solve the access order between super points; concatenate the locally optimal path sequences within each logical sub-region according to the access order between super points to obtain the optimal traversal order.

[0035] Specifically, logical sub-regions are divided according to factory buildings and floor boundaries, and the internal paths of each sub-region are solved separately; the passage cost between the entrances and exits of the sub-regions is calculated, a super point cost matrix is ​​constructed, and the access order of the sub-regions is solved; the paths of each sub-region are connected end to end according to the access order.

[0036] Furthermore, the method also includes: obtaining the navigation mode selected by the user; when the navigation mode is a two-dimensional map mode, obtaining the current location information, rendering the optimal acquisition path sequence on the two-dimensional map layer, and displaying the path between the current location and the next acquisition point in the optimal acquisition path sequence on the two-dimensional map layer; when the navigation mode is an augmented reality mode, obtaining the real-scene preview image stream, the current location information, and the attitude information, calculating the superposition position of the virtual navigation mark in the real-scene preview image stream based on the current location information and the attitude information, rendering the virtual navigation mark on the real-scene preview image stream, and superimposing and displaying prompts for the standard shooting height and level on the real-scene preview image stream.

[0037] Specifically, the 2D map mode displays the optimal acquisition path, acquisition point, and current location as a vector layer overlay; the AR mode performs projection transformation based on camera intrinsic parameters and pose data, rendering the virtual navigation arrow, target box, and shooting height prompt to the corresponding positions in the real scene.

[0038] Further, step S3 includes: step S31: based on the positional confidence parameter in the pose data, remove the on-site images with positional confidence lower than a preset threshold to generate a first candidate image set; acquire ambient lighting data, and based on the ambient lighting data, remove on-site images whose image quality does not meet the preset standard from the first candidate image set to generate a second candidate image set; In one embodiment, the location confidence threshold is set to 0.7, and images below the threshold are directly rejected; the image quality is judged by image sharpness, exposure value, and signal-to-noise ratio, and blurry, overexposed, and underexposed images are rejected.

[0039] Step S32: Group the field images in the second candidate image set according to the optimal acquisition path to generate a grouped field image sequence; perform image enhancement processing on the grouped field image sequence to generate a preprocessed field image sequence; Specifically, images are sorted and grouped according to the optimal acquisition path, ensuring consistency between the temporal and spatial sequences of the images and avoiding stitching chaos. Addressing issues such as uneven lighting, high contrast between light and dark areas, and equipment reflections in nuclear power plant environments, the system performs enhancement processing including white balance correction, brightness equalization, contrast enhancement, defogging, and noise reduction. This ensures that all images have consistent colors, are clear and transparent, and have complete details, improving the smoothness and realism of the panoramic stitching.

[0040] Step S33: Based on the pose data, stitch together the preprocessed on-site image sequence to generate a panoramic image in equidistant cylindrical projection format; Specifically, using synchronously acquired high-precision 3D coordinates and camera pose data as strong constraints, multiple images are stitched and fused to generate a panoramic image in equidistant cylindrical projection (ERP) format that conforms to the 3D reality twin standard. The stitching process requires no manual intervention, and there are no ghosting, misalignment, or distortion. The panoramic image can be directly loaded and used in the digital twin platform.

[0041] Further, the pose data includes three-dimensional coordinates and attitude angles; step S33 includes: acquiring the preprocessed field image sequence and the corresponding three-dimensional coordinates and attitude angles; calculating the initial spherical projection transformation matrix for each preprocessed field image based on the three-dimensional coordinates and attitude angles; limiting the search range of feature points from the entire image to a preset window centered on the pose prediction position, performing feature matching on the preprocessed field image sequence, and generating preliminary matching results; performing bundle adjustment on the preliminary matching results, globally optimizing the camera pose and three-dimensional point coordinates, eliminating accumulated errors, and generating optimized stitching results; and generating a panoramic image in equidistant cylindrical projection format based on the optimized stitching results.

[0042] Specifically, pose data significantly narrows the feature matching search range, greatly improving matching speed and robustness, and avoiding matching errors that are prone to occur in traditional panoramic stitching. Through bundle adjustment, the camera pose and spatial point coordinates are globally optimized to eliminate accumulated errors and local deformations, ultimately outputting a standard panoramic image with ultra-high precision, high fidelity, and seamless stitching.

[0043] This invention makes key optimizations to the open-source stitching algorithm kernel. It uses the collected accurate pose data to calculate the initial spherical projection transformation matrix for each image, and restricts the search range of feature points from the entire image to a small window centered on the pose prediction position during the feature matching stage. This reduces the computational complexity by 1-2 orders of magnitude, significantly improving the processing speed and shortening the processing time for 100 panoramic image sequences from hours to minutes. At the same time, a robust optimization strategy is adopted to ensure stitching reliability. First, a fast and high-success-rate preliminary matching is achieved based on pose constraints. Then, a cluster adjustment is used to globally optimize the camera pose and 3D point coordinates. This utilizes prior pose information to accelerate calculation and eliminates accumulated errors to ensure stitching accuracy. In addition, considering the characteristics of repetitive textures and similar structures in nuclear power plant scenes, local geometric context information is added to the feature descriptor to further improve the discriminativeness and accuracy of feature matching. The relevant optimization code is a proprietary technological achievement. Finally, a panoramic image in standard isometric cylindrical projection format is output.

[0044] Step S34: Generate a spatial information file based on the pose data.

[0045] Specifically, an independent spatial information file is generated for each panoramic image, which fully records information such as the three-dimensional coordinates of the shooting point, camera orientation, pitch angle, roll angle, positioning reliability, and shooting time, so that each panoramic image has a unique "spatial ID card" and achieves precise binding between the image and spatial location.

[0046] Further, step S34 includes: acquiring the three-dimensional coordinates and attitude angles corresponding to the panoramic image in equidistant cylindrical projection format; organizing the three-dimensional coordinates and attitude angles according to a preset format to generate a spatial information file; and outputting the panoramic image and spatial information file in equidistant cylindrical projection format as the result for three-dimensional spatial fusion.

[0047] Specifically, the spatial information files adopt a universal standardized format, which can be imported into the real-scene twin platform along with panoramic images to achieve full, high-precision, and human-intervention-free spatial registration. This completely solves the bottlenecks of low efficiency, large errors, and long cycles in traditional manual registration. The output results can be directly used in business scenarios such as digital delivery of nuclear power plants, 3D visualization, equipment management, maintenance planning, and emergency simulation.

[0048] Generate a JSON-formatted spatial information file containing the 3D coordinates and orientation of the center point of the panoramic sphere. This result can be directly imported into a 3D engine to achieve "zero-configuration" spatial alignment.

[0049] To ensure the standardized, high-precision, safe, and compliant execution of the aforementioned panoramic acquisition method, this invention also provides a dedicated integrated high-precision acquisition hardware device. This device corresponds one-to-one with the aforementioned method and is used in conjunction with it. The integrated high-precision acquisition system is the core for achieving accurate attachment of spatial information in panoramic images. It is the hardware device and hardware support foundation that matches the method. It adopts an integrated hardware design and mainly consists of a panoramic acquisition unit, a multi-source fusion positioning and attitude determination unit, an environmental perception and quality control unit, and an industrial-grade protective structure. The panoramic acquisition unit uses an industrial-grade multi-lens panoramic camera module to ensure… With high resolution, low distortion, and excellent low-light performance, the multi-source fusion positioning and attitude determination unit integrates GNSS, IMU, UWB, and visual laser SLAM modules. It can monitor the status of each positioning source in real time and switch dynamically and smoothly. It synchronously outputs accurate three-dimensional coordinates and attitude angles at the moment the shutter is triggered. The environmental perception and quality control unit integrates a photometer and temperature and humidity sensors. It can record environmental information in real time and perform quality detection and on-site prompts for image blurring, overexposure, etc. The entire hardware is packaged in a special housing that is radiation-proof, explosion-proof, dustproof, and electromagnetic interference-resistant, fully meeting the safety regulations and complex operating conditions of nuclear power plant sites.

[0050] After acquisition, the raw data packet containing image, pose, and environmental data is uploaded to the cloud via a secure network. The processing pipeline is built on Docker+Kubernetes containerization technology, supporting multi-task parallel scheduling. The core processing flow includes three steps: filtering and grouping, batch preprocessing, and intelligent stitching. The system first removes images with positioning failures and substandard quality based on pose information and groups them according to the optimal acquisition path. Then, it performs batch preprocessing such as color balancing, exposure compensation, and noise reduction on the panoramic image sequence. Finally, it uses an optimized stitching algorithm kernel, with the accurate pose data obtained in the acquisition stage as a strong constraint, to narrow the feature matching range from the entire image to a preset window, greatly improving the stitching speed and accuracy, avoiding traditional manual point selection operations, and achieving high precision, full generation, and spatial alignment of the panoramic sphere.

[0051] The panoramic data acquisition method for nuclear power plants provided by this invention possesses seamless indoor and outdoor positioning capabilities throughout the entire plant. Through UWB / SLAM+IMU multi-source fusion positioning, it can still provide centimeter-level high-precision continuous positioning and attitude determination even in indoor areas without GNSS signals, underground pipe corridors, and densely equipped areas, truly achieving full plant coverage without blind spots. Simultaneously, it employs a portable, integrated handheld hardware platform, which is lightweight, flexible, and operable by a single person, enabling access to all personnel-permitted areas within the nuclear power plant, ensuring comprehensive and blind-spot-free data acquisition. The tool is equipped with intelligent path planning based on 3D models, generating optimal coverage and highest operational efficiency acquisition schemes. Combined with AR real-scene navigation, it precisely guides operators to preset acquisition points and standardizes shooting postures, ensuring standardized execution and high coverage of panoramic data acquisition operations. The system has built-in... The real-time quality pre-inspection algorithm can instantly identify image quality issues such as blurriness, overexposure, and underexposure at the acquisition site and prompt for reshoots, effectively ensuring the success rate of first-time acquisition and data quality. The tool is equipped with a cloud-based, fully automated data processing pipeline. Relying on the precise pose data acquired synchronously during acquisition, it can quickly stitch panoramic images together and align them spatially. It can directly output standardized results adapted to the 3D twin platform, completing an end-to-end closed loop from raw data to usable data, improving data processing efficiency by an order of magnitude. In addition, the hardware platform adopts an industrial-grade protective design that is radiation-proof, explosion-proof, and resistant to electromagnetic interference. The software process strictly complies with the safety regulations of nuclear power plants and avoids high-risk areas during the path planning stage, forming a truly customized, safe, compliant, stable, and reliable integrated panoramic acquisition solution specifically for nuclear power scenarios.

[0052] It is understood that the above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can freely combine the above technical features without departing from the concept of the present invention, and can also make several modifications and improvements, all of which fall within the protection scope of the present invention. Therefore, all equivalent transformations and modifications made with respect to the scope of the claims of the present invention should fall within the scope of the claims of the present invention.

Claims

1. A method of panoramic acquisition of a nuclear power plant, characterized in that, Includes the following steps: Step S1: Obtain spatial reference data of the nuclear power plant area, and generate the optimal acquisition path covering the target acquisition area through path planning algorithm; Step S2: Obtain the on-site images and pose data acquired according to the optimal acquisition path; Step S3: Based on the pose data, process the scene image to generate a panoramic image.

2. The method for panoramic acquisition of a nuclear power plant according to claim 1, characterized in that, The spatial reference data includes one or more of the following: two-dimensional plan view, lightweight BIM model and point cloud data; Step S1 includes: Step S11: Generate a candidate collection point set from one or more of the two-dimensional plan view, the lightweight BIM model, and the point cloud data according to the preset collection density parameters; Step S12: When the spatial reference data contains the lightweight BIM model, obtain the room function labels in the lightweight BIM model, identify key areas based on the room function labels, assign a first weight value to candidate points located in the key areas, assign a second weight value to candidate points located in non-key areas, and the first weight value is greater than the second weight value. Step S13: Based on A The algorithm calculates the actual passable path length between each point in the candidate collection point set, and gives a penalty coefficient to the path segment passing through the radiation control area in combination with the safety rules to construct a cost matrix. Step S14: Using the Traveling Salesman Problem optimization algorithm, the optimal collection path sequence is calculated and output from the candidate collection point set based on the cost matrix and the weight of each candidate point.

3. The method for panoramic acquisition of a nuclear power plant according to claim 2, characterized in that, The acquisition density parameters include the acquisition spacing along the channel direction and the grid spacing within the equipment room; Step S11 includes: Determine whether the type of the collection area in the two-dimensional plan view, the lightweight BIM model, and the point cloud data is a channel area or an equipment room area. When the acquisition area type is a channel area, candidate acquisition points are generated within the channel area according to the acquisition interval; When the collection area type is an inter-device area, grid-shaped candidate collection points are generated in the inter-device area according to the grid point spacing. The candidate collection points and the grid-shaped candidate collection points are combined to obtain the candidate collection point set.

4. The method for panoramic acquisition of a nuclear power plant according to claim 2, characterized in that, Step S13 includes: A navigation grid is constructed based on one or more of the two-dimensional plan view, the lightweight BIM model, or the point cloud data. For any two candidate points in the candidate collection point set, A is used. The algorithm calculates the actual traversable path length from the starting point to the ending point on the navigation grid, which is used as the traversal cost between the two points. Identify the radiation control zones in the navigation grid, and for path segments that pass through the radiation control zones, add a preset penalty coefficient to the passage cost; The cost matrix is ​​constructed based on the travel costs between all pairs of candidate points.

5. The method for panoramic acquisition of a nuclear power plant according to claim 2, characterized in that, Step S14 includes: The passage cost between any two points is adjusted according to the following formula: in, Let be the original travel cost from candidate point i to candidate point j. The revised passage cost. and Let i and j be the weights of candidate point i and candidate point j, respectively. and This is the preset adjustment coefficient; Determine whether the number of candidate points in the candidate collection point set exceeds a preset scale threshold; When the number of candidate points in the candidate collection point set does not exceed the preset scale threshold, all candidate points in the candidate collection point set are taken as nodes of the Traveling Salesman Problem, and the corrected passage cost is applied. As the edge cost between nodes, find the optimal traversal order that minimizes the total correction cost; When the number of candidate points in the candidate collection point set exceeds the preset scale threshold, a hierarchical planning strategy is adopted to solve for the optimal traversal order. Based on the optimal traversal order, the optimal acquisition path sequence is output, which includes an ordered sequence of acquisition points and the walking path connecting each acquisition point.

6. The method of panoramic acquisition of a nuclear power plant according to claim 5, characterized in that, The method of using hierarchical planning to solve for the optimal traversal order includes: The data collection area is divided into several logical sub-areas based on the factory buildings or floors; In each logical sub-region, taking the candidate points in the logical sub-region as nodes, taking the modified passing cost As an edge cost, the traveling salesman problem is solved respectively to obtain a local optimal path sequence in each logical sub-region. Treat each logical sub-region as a super point, and use the travel cost between sub-regions as the edge cost to solve the access order between super points; The optimal traversal order is obtained by concatenating the locally optimal path sequences within each logical sub-region according to the access order between the superpoints.

7. The method for panoramic acquisition of a nuclear power plant according to claim 2, characterized in that, The method further includes: Get the navigation mode selected by the user; When the navigation mode is a two-dimensional map mode, the current location information is obtained, the optimal collection path sequence is rendered on the two-dimensional map layer, and the path between the current location and the next point to be collected in the optimal collection path sequence is displayed on the two-dimensional map layer. When the navigation mode is augmented reality mode, the system acquires the real-view preview image stream, current location information, and attitude information. Based on the current location information and the attitude information, it calculates the superposition position of the virtual navigation mark in the real-view preview image stream, renders the virtual navigation mark on the real-view preview image stream, and superimposes and displays prompts indicating the standard shooting height and level on the real-view preview image stream.

8. The method for panoramic acquisition of a nuclear power plant according to claim 1, characterized in that, The pose number includes a position confidence parameter, and step S3 includes: Step S31: Based on the location confidence parameter, remove the on-site images whose location confidence is lower than a preset threshold to generate a first candidate image set; acquire ambient lighting data, and based on the ambient lighting data, remove on-site images whose image quality does not meet a preset standard from the first candidate image set to generate a second candidate image set; Step S32: Group the on-site images in the second candidate image set according to the optimal acquisition path to generate a grouped on-site image sequence; perform image enhancement processing on the grouped on-site image sequence to generate a preprocessed on-site image sequence; Step S33: Based on the pose data, stitch together the preprocessed scene image sequence to generate a panoramic image in equidistant cylindrical projection format; Step S34: Generate a spatial information file based on the pose data.

9. The method for panoramic acquisition of nuclear power plants according to claim 8, characterized in that, The pose data includes three-dimensional coordinates and pose angles; Step S33 includes: Obtain the preprocessed on-site image sequence and the corresponding three-dimensional coordinates and attitude angles; Based on the three-dimensional coordinates and the attitude angle, calculate the initial spherical projection transformation matrix for each preprocessed on-site image; The search range of feature points is restricted from the entire image to a preset window centered on the pose prediction position. Feature matching is then performed on the preprocessed on-site image sequence to generate preliminary matching results. The initial matching results are subjected to bundle adjustment, and the camera pose and 3D point coordinates are globally optimized to eliminate accumulated errors and generate optimized stitching results. Based on the optimized stitching results, a panoramic image in equidistant columnar projection format is generated.

10. The method of panoramic acquisition of a nuclear power plant according to claim 9, characterized in that, Step S34 includes: Obtain the three-dimensional coordinates and attitude angles corresponding to the panoramic image in the equidistant cylindrical projection format; The three-dimensional coordinates and the attitude angles are organized according to a preset format to generate a spatial information file; The panoramic image in the equidistant cylindrical projection format and the spatial information file are used as the output for three-dimensional spatial fusion.

Images