Building facade 3D mapping system and method based on drone and AI

Abstract

The present invention discloses a system and method for three-dimensional mapping of building facades based on drones and AI, relating to the field of building surveying technology. The system comprises: a scene parsing module for performing information parsing of calibrated scene information and marking each surveying sub-area as a node to be collected; a collection planning module for calculating the node transition probability of each node to be collected; a surveying search module for configuring a virtual drone and iteratively selecting surveying sub-areas using pheromone concentration and heuristic factors; a clustering and sharing module for performing three-dimensional mapping of building facades and clustering to establish a shared information pool; and a surveying optimization module for performing collaborative decision-making optimization of local paths and updating the collection path based on the collaborative decision-making optimization results. This system solves the technical problem in existing technologies that it is difficult to meet the requirements of large-scale, high-precision surveying and mapping of modern buildings, resulting in low surveying and mapping efficiency and large measurement errors, thereby achieving the technical effect of improving the efficiency and accuracy of building surveying and mapping.

Description

Technical Field

[0001] The present application relates to the technical field related to building surveying and mapping, and specifically to a three-dimensional surveying system and method for building facades based on drones and AI. Background Art

[0002] With the acceleration of urbanization, the construction industry is booming, and the demand for accurate surveying of building facades is growing. Traditional methods of surveying building facades, such as manual field surveying using simple surveying instruments, require professionals to carry equipment and climb to various parts of the building facade to conduct measurements. This is not only inefficient and time-consuming, but also has large measurement errors, is limited by human physical strength and operating accuracy, and is significantly affected by environmental factors. For example, in some high-rise or complex structures, it is difficult for humans to reach certain locations, resulting in missing data. In addition, conventional instrument surveying, such as total stations, although they can improve accuracy to a certain extent, require stations to be set up in multiple locations, the process is cumbersome, and has high requirements for the surrounding environment. It is difficult to effectively perform in narrow spaces or when there are many surrounding obstacles. The acquired data is difficult to process, and it is impossible to quickly generate intuitive three-dimensional models. It is difficult to meet the requirements of large-scale, high-precision surveying of modern buildings.

[0003] Therefore, at the current stage, relevant technologies have technical problems that are difficult to meet the requirements of large-scale and high-precision surveying and mapping of modern buildings, resulting in low surveying and mapping efficiency and large measurement errors. Summary of the Invention

[0004] This application provides a three-dimensional surveying and mapping system and method for building facades based on drones and AI, which solves the technical problems in the existing technology that are difficult to meet the large-scale and high-precision surveying and mapping requirements of modern buildings, resulting in low surveying and mapping efficiency and large measurement errors, and achieves the technical effect of improving the efficiency and accuracy of building surveying and mapping.

[0005] The present application provides a three-dimensional mapping system for building facades based on drones and AI. The system includes: a scene analysis module for reading the calibration scene information of the scene to be mapped, performing information analysis of the calibration scene information, establishing N mapping sub-areas, and marking each mapping sub-area as a node to be collected; an acquisition planning module for allocating initial pheromone values ​​on each mapping path after the collection starting point of the distributed drone group, configuring heuristic factors, and calculating the node transfer probability of each node to be collected; a mapping search module for configuring a virtual drone to start from the collection starting point and calculate the node transfer probability of each node to be collected according to the initial pheromone value. The pheromone concentration is calculated, and the pheromone concentration and the heuristic factor are used to iteratively select the mapping sub-area to establish an optimization result, wherein the initial pheromone value is cumulatively updated after each completion of the global path planning; a clustering sharing module is used to control the drone group to perform three-dimensional mapping of the building facade based on the optimization result, and to establish a shared information pool according to the real-time mapping position clustering of the drone group, and upload the real-time mapping results to the shared information pool; a mapping optimization module is used to perform collaborative decision-making optimization of the local path through the shared information pool and the optimization result, and to update the collection path according to the collaborative decision-making optimization result.

[0006] In a possible implementation, the drone and AI-based three-dimensional mapping system for building facades further performs the following processing: a drawing reading module for reading the scene drawings of the scene to be surveyed and using the scene drawings as calibration scene information; a grid segmentation module for establishing a median grid granularity, performing grid division of the building facade of the scene to be surveyed based on the median grid granularity, and generating an initial grid segmentation result; a grid update module for identifying the integrity of elements within the grid on the initial grid segmentation result, performing grid migration clustering based on the integrity identification result, generating a grid migration clustering result, and establishing N mapping sub-areas based on the grid migration clustering result.

[0007] In a possible implementation, the drone- and AI-based three-dimensional mapping system for building facades further performs the following processing: obtaining a set of features not covered by the initial grid segmentation result and marking it as an updated feature set; obtaining the size data of each feature in the updated feature set, and selecting clustering features based on the matching degree between the size data and the median grid granularity; and using the clustering features to complete grid migration clustering.

[0008] In a possible implementation, the drone and AI-based building facade 3D mapping system further performs the following processing: establishing a collection node transfer cost function as follows:

[0009] ;

[0010] in, Characterization from the acquisition node Transfer to the collection node The acquisition node transfer cost function, Characterization acquisition node and collection nodes distance, and Characterize the acquisition nodes separately and collection nodes risk factors, and Characterize the acquisition nodes separately and collection nodes The node priority, 、 、 They are distance weight factor, risk weight factor, and node priority weight factor respectively; according to the acquisition node transfer cost function, the heuristic factor is configured, and the node transfer probability is calculated using the heuristic factor as follows:

[0011] ;

[0012] in, Characterization from the acquisition node To the collection node The node transition probability, Characterization acquisition node To the collection node The pheromone concentration, Is the heuristic factor, specifically representing the collection node To the collection node Heuristic information, , Characterizes the set of optional collection nodes that the drone has not visited, Represents any optional acquisition node, Characterization acquisition node To the collection node The pheromone concentration, Characterization acquisition node To the collection node Heuristic information, Characterize pheromone influencing factors, Characterize the heuristic factors affecting the factors.

[0013] In a possible implementation, the drone- and AI-based three-dimensional mapping system for building facades also performs the following processing: obtaining the spatial position coordinates of each drone in the drone swarm at the current time node; obtaining the next-round mapping sub-area of ​​each drone in the drone swarm; performing real-time mapping position clustering based on the next-round mapping sub-area and the spatial position coordinates to generate a real-time mapping position clustering result; identifying the drone with the largest storage space in each cluster in the real-time mapping position clustering result, creating a shared information pool for the identified drones, and receiving the real-time mapping results of other drones in the real-time mapping position clustering result.

[0014] In a possible implementation, the drone- and AI-based three-dimensional mapping system for building facades further performs the following processing: obtaining environmental shared information in a shared information pool, and reconstructing local environmental mutation constraints based on the environmental shared information; obtaining mapping anomalies based on real-time mapping results in the shared information pool, and creating additional mapping tasks using the mapping anomalies; and performing collaborative decision-making optimization of local paths based on the local environmental mutation constraints, the additional mapping tasks, and the optimization results.

[0015] In a possible implementation, the drone- and AI-based three-dimensional mapping system for building facades also performs the following processing: an anomaly recognition module is used to record the collection anomalies of the drone group and perform anomaly level judgment, match the response schemes in the response scheme library based on the anomaly level judgment results, and perform anomaly management based on the response scheme matching results. The response scheme library includes an anomaly marking response scheme and a temporary task creation response scheme.

[0016] The present application also provides a three-dimensional mapping method for building facades based on drones and AI, including: reading calibration scene information of a scene to be mapped, performing information analysis of the calibration scene information, establishing N mapping sub-areas, and marking each mapping sub-area as a node to be collected; after distributing the collection starting point of the drone group, allocating an initial pheromone value on each mapping path, configuring an inspiration factor, and calculating the node transfer probability of each node to be collected; configuring a virtual drone to start from the collection starting point, calculating the pheromone concentration based on the initial pheromone value, and using the pheromone concentration and the inspiration factor to iteratively select the mapping sub-area to establish an optimization result, wherein, after each global path planning is completed, the initial pheromone value is accumulated and updated; based on the optimization result, the drone group is controlled to perform three-dimensional mapping of the building facade, and a shared information pool is established according to the real-time mapping position clustering of the drone group, and the real-time mapping results are uploaded to the shared information pool; collaborative decision optimization of the local path is performed through the shared information pool and the optimization result, and the collection path is updated according to the collaborative decision optimization result.

[0017] The proposed drone- and AI-based 3D building facade mapping system and method includes a scene analysis module for performing information analysis of calibrated scene information and marking each mapping sub-area as a node to be collected; a collection planning module for calculating the node transition probability of each node to be collected; a mapping search module for configuring a virtual drone and iteratively selecting mapping sub-areas using pheromone concentration and heuristic factors; a clustering and sharing module for performing 3D mapping of building facades and clustering to establish a shared information pool; and a mapping optimization module for collaborative decision-making optimization of local paths and updating the collection path based on the collaborative decision-making optimization results. This solves the technical problem in existing technologies that it is difficult to meet the requirements of large-scale, high-precision mapping of modern buildings, resulting in low mapping efficiency and large measurement errors, thereby achieving the technical effect of improving the efficiency and accuracy of building mapping. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0019] Figure 1 Schematic diagram of the structure of the drone- and AI-based building facade 3D mapping system provided in an embodiment of the present application.

[0020] Figure 2 Schematic diagram of the flow chart of the three-dimensional mapping method of building facades based on drones and AI provided in an embodiment of the present application.

[0021] Description of the accompanying drawings: scene analysis module 10, acquisition planning module 20, surveying and mapping search module 30, clustering and sharing module 40, surveying and mapping optimization module 50. DETAILED DESCRIPTION

[0022] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0023] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0024] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0025] The present application embodiment provides a building facade 3D mapping system based on drones and AI, such as Figure 1 As shown, the system includes:

[0026] The scene analysis module 10 is used to read the calibration scene information of the scene to be surveyed, perform information analysis on the calibration scene information, establish N survey sub-areas, and mark each survey sub-area as a node to be collected.

[0027] Preferably, reading the calibration scene information of the scene to be surveyed may include obtaining detailed information such as the basic size, shape, structural layout, floor height, door and window positions of the building from the building's design drawings; then using GIS data to obtain the building's geographical location information, including latitude and longitude, altitude, etc., as well as environmental information such as surrounding terrain and landforms, to determine the position of the surveying area in the entire geographic space, and consider the impact of the surrounding environment on the surveying work, such as obstruction by surrounding buildings, flight airspace restrictions, etc.; obtaining key dimensional data and feature point information through field measurement, such as the actual height of the building, the inclination of the facade, the size of special structures, etc.; and using sensor equipment such as laser rangefinders and total stations to obtain distance information and spatial coordinate data of the building facade.

[0028] Preferably, information parsing is to convert the read original calibration scene information into actionable data. Specifically, the various data read are format converted and normalized so that they can be processed by unified algorithms and models. For example, coordinate data from different sources are converted into a unified coordinate system, image data are adjusted to a standard resolution and format, noise and outliers in the data are removed, and laser ranging data, sensor measurement data, etc. are processed through filtering, smoothing and other algorithms to improve the quality and reliability of the data; computer vision algorithms and machine learning models are used to extract building feature information from image data, such as edges, corners, textures, etc. For example, the Canny edge detection algorithm is used to extract the contour edges of the building facade, and the Harris corner detection algorithm is used to extract the contour edges of the building facade. The algorithm identifies key corner points in building structures. For laser ranging data and spatial coordinate data, it extracts the geometric features of the building, such as the parameters and position information of structures such as planes, surfaces, and columns, through cluster analysis, surface fitting, and other methods. For example, the DBSCAN clustering algorithm is used to cluster laser point cloud data into different structural units, and the plane equation is fitted by the least squares method to determine the plane area of ​​the building facade; then, the extracted features are semantically understood and annotated by combining knowledge and semantic models in the architectural field. For example, by training deep learning models such as convolutional neural networks (CNN) or semantic segmentation models, different building components such as windows, doors, balconies, columns, etc. are identified and their types and location information are annotated to achieve semantic segmentation and recognition of building images and point cloud data.

[0029] Preferably, N mapping sub-areas are established, where N is a positive integer representing the total number of mapping sub-areas. Specifically, division rules are formulated according to the structural characteristics of the building and the mapping accuracy requirements. For example, the building facade is divided into several horizontal sub-areas according to the floor height. The height of each sub-area can be determined according to the flight performance and mapping accuracy of the drone, generally about 5-10 meters. Alternatively, the building facade is divided into several rectangular or polygonal sub-areas according to the shape of the building facade and the structural unit. Each sub-area contains as many complete building components or structural features as possible. Alternatively, an image segmentation algorithm or a spatial clustering algorithm is used to automatically divide the feature data of the building. For example, an image segmentation algorithm based on region growing is used to start from the seed point of the building image and gradually grow different sub-areas according to the similarity of pixels. Alternatively, a K-means clustering algorithm is used to cluster the laser point cloud data. Classification, divide the point cloud data into N different sub-regions, each sub-region corresponds to a mapping sub-region, and optimize and adjust the preliminary divided sub-regions to ensure that the boundaries between sub-regions are clear and seamless, and the size and shape of each sub-region are relatively uniform, avoiding sub-regions that are too large or too small, affecting the mapping efficiency and accuracy. At the same time, considering the flight path and acquisition angle of the drone, some sub-regions with special shapes or difficult to collect are appropriately merged or split; and a unique identifier is assigned to each mapping sub-region, marking it as a point to be collected. The annotation content may also include relevant attribute information of the mapping sub-region, such as the location coordinates, size, shape, and building structure type of the sub-region, which helps the drone to accurately locate and identify each sub-region during the collection process. For example, the coordinates of the upper left and lower right corners of the sub-region, as well as the corresponding floor and building facade information, are recorded.

[0030] Furthermore, the specific configuration of the scene parsing module 10 also includes a drawing reading module for reading the scene drawings of the scene to be surveyed and using the scene drawings as calibration scene information; a grid segmentation module for establishing a median grid granularity, performing grid division of the building facade of the scene to be surveyed based on the median grid granularity, and generating an initial grid segmentation result; a grid update module for identifying the integrity of elements within the grid on the initial grid segmentation result, performing grid migration clustering based on the integrity identification result, generating a grid migration clustering result, and establishing N surveying sub-areas based on the grid migration clustering result.

[0031] Preferably, the main function of the drawing reading module is to read the scene drawings of the scene to be surveyed and use these drawings as calibration scene information, wherein the scene drawings contain various detailed information of the building to be surveyed. Specifically, the drawings of the scene to be surveyed are obtained from different data sources, such as electronic documents provided by the construction unit, scanned versions of archive drawings, etc., and the drawing formats are uniformly processed. For example, for drawings in CAD format, it may be necessary to use a professional CAD library for reading and conversion. For drawings in image format, image preprocessing may be required, such as adjusting the resolution, grayscale, etc.; extract information related to surveying and mapping from the drawings, such as the building's external dimensions, structural layout, door and window positions, floor height, etc., and use them as calibration scene information.

[0032] Preferably, the core task of the grid segmentation module is to establish a median grid granularity, and based on this granularity, the building facade of the surveying scene is grid-divided, and finally an initial grid segmentation result is generated. Specifically, the median grid granularity is determined by comprehensively considering multiple factors, such as the surveying accuracy requirements, the complexity of the building, the flight performance of the drone, etc., wherein the median grid granularity is a key parameter that determines the grid size. Generally speaking, for scenes with high accuracy requirements or complex building structures, the median grid granularity should be set smaller, otherwise it can be appropriately increased; then, according to the determined median grid granularity, the building facade is divided into several grids, for example, a regular grid division method, such as a square grid or a rectangular grid, can be used, or an irregular grid division method can be used according to the shape and structural characteristics of the building. During the division process, it is ensured that the grid can cover the entire building facade and there is no overlap or omission between the grids; finally, the divided grid information is sorted and recorded to form an initial grid segmentation result, which usually includes information such as the position, size, and number of each grid.

[0033] Preferably, the grid update module mainly identifies the integrity of the elements in the grid based on the initial grid segmentation result, performs grid migration and clustering based on the identification result, and finally generates a grid migration and clustering result, and establishes N mapping sub-areas based on this. Specifically, the integrity of the elements is judged by analyzing the information in the drawings or combining image recognition technology, that is, each grid in the initial grid segmentation result is checked to identify whether the building elements (such as doors and windows, walls, decorative components, etc.) contained in the grid are complete. For example, if a grid only contains some doors and windows, then the elements in the grid are incomplete; according to the element integrity identification result, the grid is migrated and clustered, and for the grid with incomplete elements, it is migrated to an adjacent grid. In a grid containing complete elements, or merging multiple adjacent grids with incomplete elements into a new grid, the elements in each grid are made as complete as possible, thereby improving the accuracy and efficiency of surveying and mapping, and the grid information after migration and clustering operations is sorted and recorded to form a grid migration clustering result, reflecting the optimized grid division, and the elements in each grid are more complete and reasonable; finally, based on the grid migration clustering results, the building facade is divided into N surveying and mapping sub-areas, such as the sub-areas are divided according to factors such as the location of the grid, the type of elements, and the difficulty of surveying and mapping. Each surveying and mapping sub-area should have relatively independent surveying and mapping tasks and targets so that the drone can collect data more efficiently.

[0034] Furthermore, the specific configuration of the scene parsing module 10 also includes obtaining a set of features not covered by the initial grid segmentation result and marking it as an updated feature set; obtaining the size data of each feature in the updated feature set, and selecting clustering elements based on the matching degree between the size data and the median grid granularity; and using the clustering elements to complete grid migration clustering.

[0035] Preferably, when meshing the building facade, due to the complexity of the building structure, data errors, or limitations of the meshing rules, the initial meshing result may not completely cover all building elements (such as doors and windows, decorative components, special texture areas, etc. on the building facade). By comparing the initial meshing result with the original scene drawings, the elements not included in any initial mesh are found and marked as an updated feature set, which contains the elements omitted in the meshing process. For each element in the updated feature set, its detailed dimension data is obtained, which may include extracting it from the original design drawings, or obtaining the length, width, area, etc. of the element through image measurement, three-dimensional modeling, etc.; then, the dimension data of each element is compared with the median mesh granularity to evaluate the matching degree between them. For example, if the area of ​​an element is similar to the mesh area corresponding to the median mesh granularity, or the length and width of the element can be well integrated into a mesh based on the median mesh granularity, then it is considered that the element has a high matching degree with the median mesh granularity; based on the matching degree evaluation results, those elements with high matching degrees are selected from the updated feature set as clustering elements, thereby improving the rationality and completeness of the meshing.

[0036] Preferably, considering the location of the grid, the type of elements, the integrity of the elements already included, etc., in the initial grid segmentation result, a suitable migration target grid is found for each cluster element. For example, if a cluster element is a window, the grid containing part of the window element or adjacent to the window position can be preferentially selected as the migration target, and the cluster element is migrated to the selected target grid, which may include updating the attribute information of the target grid, such as area, type and number of elements included, etc., while ensuring that the migrated grid meets the integrity of the elements and the connectivity of the grid; after completing the migration of the cluster elements, some adjacent grids with similar elements may appear, and these grids are clustered and merged. Through the grid migration clustering method, the grid division results are further optimized, making the elements in each grid more complete and unified, ensuring that all elements of the building facade can be reasonably included in the division of the surveying and mapping sub-areas, thereby improving the accuracy and efficiency of the surveying and mapping.

[0037] Preferably, performing grid migration clustering on the initial grid segmentation result specifically includes: first clarifying the attribute information of each grid in the initial grid segmentation result, including the position (coordinates), size, type of building elements contained and related characteristics (such as the number of doors and windows, area ratio, etc.) of the grid, establishing a two-dimensional array to represent the grid set, and each element corresponds to a grid; then setting an integrity threshold to evaluate the integrity of each grid, if the proportion of the area of ​​key elements such as doors and windows in the grid to the total area of ​​the grid is lower than the threshold, the grid element is considered incomplete; or judging based on whether the grid contains a complete building structure unit (such as the wall area corresponding to a complete room); then determining the adjacent grids for each grid by defining neighborhood rules, for example, for a rectangular grid on a two-dimensional plane, defining the grids adjacent to the current grid in the horizontal, vertical or diagonal direction as its adjacent grids, and then establishing an adjacency list or adjacency matrix to represent the grid adjacency relationship, so as to quickly find the adjacent grids of each grid.

[0038] Preferably, mesh migration clustering is then performed, that is, all meshes are traversed to determine their integrity. If the current mesh element is complete, the mesh is skipped and the next mesh is processed. If the current mesh element is incomplete, the mesh with complete elements is searched for in the adjacent meshes of the current mesh based on the relationship between adjacent meshes and the result of element integrity assessment. If multiple adjacent complete element meshes are found, the mesh with the closest distance (calculated based on the distance between mesh centers) or the most similar element features (such as door and window types, wall materials, etc.) is selected as the target mesh. The attribute information of the current incomplete element mesh (such as location, included partial elements, etc.) is then merged into the target mesh, and then the mesh is clustered. Delete the current incomplete feature grid during merging, and update the adjacent grid relationship data structure at the same time to ensure that the adjacent relationship between other grids and the deleted grid is correctly handled (for example, update the adjacency list or adjacency matrix). Iterate until all grids are traversed. After the grid migration is completed, there may be some grids with complete adjacent features but high similarity. These grids are clustered and merged. For example, by calculating the similarity between adjacent grids (based on a comprehensive evaluation of factors such as feature type, location relationship, and area ratio), grids with similarity higher than a certain threshold are merged into a larger grid. Repeat the clustering optimization until there are no more grids that can be merged, and finally obtain the grid migration clustering result.

[0039] The collection planning module 20 is used to allocate initial pheromone values ​​on each mapping path after distributing the collection starting point of the drone group, configure the heuristic factor, and calculate the node transfer probability of each node to be collected.

[0040] Furthermore, the specific configuration of the collection planning module 20 also includes establishing a collection node transfer cost function as follows:

[0041] ;

[0042] in, Characterization from the acquisition node Transfer to the collection node The acquisition node transfer cost function, Characterization acquisition node and collection nodes distance, and Characterize the acquisition nodes separately and collection nodes risk factors, and Characterize the acquisition nodes separately and collection nodes The node priority, 、 、 They are distance weight factor, risk weight factor, and node priority weight factor respectively; according to the acquisition node transfer cost function, the heuristic factor is configured, and the node transfer probability is calculated using the heuristic factor as follows:

[0043] ;

[0044] in, Characterization from the acquisition node To the collection node The node transition probability, Characterization acquisition node To the collection node The pheromone concentration, Is the heuristic factor, specifically representing the collection node To the collection node Heuristic information, , Characterizes the set of optional collection nodes that the drone has not visited, Represents any optional acquisition node, Characterization acquisition node To the collection node The pheromone concentration, Characterization acquisition node To the collection node Heuristic information, Characterize pheromone influencing factors, Characterize the heuristic factors affecting the factors.

[0045] Preferably, in the three-dimensional mapping scenario of the building facade, multiple factors are comprehensively considered, such as the location, shape, surrounding environment and overall mapping needs of the building, to determine the location where the drone group starts data collection, that is, the collection starting point. The collection starting point can be determined based on the geometric center of the building, the main entrance and exit, or other representative locations. For example, for a building with a regular shape, the starting point can be set directly in front of or directly above the building. For a building with a complex shape, it may be necessary to distribute multiple starting points based on multiple key feature points to ensure that the drone can comprehensively and efficiently map the building facade; assign an identical initial pheromone value to all mapping paths, which is usually a small constant, so that all paths have a chance to be selected, avoiding some paths from being overly preferred or ignored from the beginning. Among them, pheromone is an important concept in the ant colony algorithm. In drone mapping path planning, the pheromone value represents the attractiveness or quality of a mapping path. The initial pheromone value is the initial value assigned to each possible mapping path at the beginning of the algorithm.

[0046] Preferably, the heuristic factor is a parameter used to guide the drone to select a path, reflecting the visibility or intuitive appeal of the path. Specifically, the heuristic factor is determined according to the specific surveying and mapping scenario and requirements (path distance, angle, obstacle conditions, etc.). For example, a shorter path or a path with a better vertical angle to the building facade may have a higher heuristic factor. Then, the node transfer probability of each node to be collected is calculated, where the node transfer probability refers to the probability of the drone transferring from the current node to the next node to be collected. According to the principle of the ant colony algorithm, the node transfer probability is calculated by the formula The calculation is carried out by comprehensively considering the pheromone value and heuristic factor on the path to simulate the behavior of ants in choosing a path based on factors such as pheromone and distance when searching for food. By calculating the transfer probability of each node to be collected, the drone can randomly select the next node to be collected according to the probability, thereby gradually planning a mapping path, achieving more efficient and reasonable mapping path planning, and improving the efficiency and quality of drone mapping. Among them, Characterization from the acquisition node To the collection node The node transition probability, Characterization acquisition node To the collection node The pheromone concentration, Is the heuristic factor, specifically representing the collection node To the collection node Heuristic information, , Characterizes the set of optional collection nodes that the drone has not visited, Represents any optional acquisition node, Characterization acquisition node To the collection node The pheromone concentration, Characterization acquisition node To the collection node Heuristic information, Characterization of pheromone influencing factors, Characterize the influence factor of the heuristic factor, which is used to adjust the influence of pheromone concentration and heuristic factor on the transfer probability.

[0047] Preferably, Characterization from the acquisition node Transfer to the collection node The collection node transfer cost function is used to quantify the cost of the UAV from the collection node Transfer to the collection node The price paid is: ,in, Characterization acquisition node and collection nodes distance, and Characterize the acquisition nodes separately and collection nodes risk factors, and Characterize the acquisition nodes separately and collection nodes The node priority, 、 、 These factors are distance weighting, risk weighting, and node priority weighting. Specifically, the longer the distance, the greater its weight in the cost function, meaning the transfer cost increases with distance. For example, in actual surveying and mapping, longer flight distances mean the drone consumes more power and may also increase the risk of external interference. Therefore, distance is an important factor affecting transfer costs. Risk factors can encompass a variety of situations, such as strong air currents, signal interference, and building obstructions in the area, which can affect drone flight safety. The higher the flight risk, the greater the risk factor value, and the higher the transfer cost. High-priority nodes may be locations containing important surveying and mapping information, and appear in the cost function as minus the corresponding product. This means that transferring to high-priority nodes reduces transfer costs, guiding drones to prioritize these key nodes for data collection.

[0048] The surveying and searching module 30 is used to configure the virtual drone to start from the acquisition starting point, calculate the pheromone concentration based on the initial pheromone value, and use the pheromone concentration and the heuristic factor to iteratively select the surveying sub-area to establish the optimization result, wherein the initial pheromone value is cumulatively updated after each completion of the global path planning.

[0049] Preferably, a virtual drone is configured to perform tasks from a determined collection starting point, wherein the virtual drone is used to simulate the behavior of a real drone in the mapping area, and explore different mapping paths and sub-area selection schemes through path planning simulation; the virtual drone calculates the pheromone concentration on each path at the current moment based on the initial pheromone value. In the ant colony algorithm, pheromones will evaporate over time and will also be enhanced by the passage of the drone. Specifically, a pheromone volatility coefficient ρ is set, which represents the volatilization ratio of pheromones per unit time, and the value range is between 0 and 1. For example, ρ=0.1 means that the pheromone on the path will evaporate by 10% every unit time. A constant Q is set to control the degree of pheromone enhancement.

[0050] Preferably, since pheromones evaporate naturally, pheromone volatilization calculation is performed in each time interval (which can be regarded as an iteration cycle), that is, at time t+1, the pheromone concentration on path (i, j) is (t+1) First of all, we need to consider the remaining amount after volatilization, and the calculation formula is: (t+1)=(1-ρ) (t), where (t) is the pheromone concentration on the path (i, j) at time t. At the beginning of the iteration, (t) = (0), that is, after a time interval, the remaining pheromone on the path is the pheromone concentration at the previous moment multiplied by (1-ρ). When a virtual drone passes through a path, it will leave pheromones on the path, thereby increasing its concentration. Assume that in one iteration, there are m virtual drones participating in path planning. When the kth virtual drone passes through path (i, j), the amount of pheromone added to the path is ∆ , then after all virtual drones pass through, the total amount of pheromone enhancement on path (i, j) is: , usually, ∆ The calculation is related to the performance of the virtual drone in this path planning, for example, the distance the drone flies can be To calculate: If the kth UAV passes through the path (i, j), then ∆ ; If the kth UAV does not pass through the path (i, j), then ∆ , meaning that the shorter the flight distance, the more pheromone it adds to the path it traverses. Finally, the remaining pheromone after volatilization is added to the enhanced amount to obtain the final pheromone concentration on path (i, j) at time t+1. Through continuous iterative calculations, over time and with multiple virtual drone path planning simulations, the pheromone concentration on each path will dynamically change, reflecting the quality of the selected path and guiding subsequent mapping sub-area selection and path planning.

[0051] Preferably, based on the pheromone concentration and the heuristic factor, a virtual drone is used to simulate the optimization and perform iterative selection of the mapping sub-areas. Specifically, the parameters such as the pheromone volatility coefficient ρ, the pheromone influence factor δ, the heuristic factor influence factor φ, and the upper limit of the number of iterations N are determined, and the initial pheromone values ​​are assigned to the paths between all the mapping sub-areas (nodes to be collected). (0), and calculate the heuristic factor based on the acquisition node transfer cost function , then calculate the probability of the virtual drone transferring to each unvisited node. Based on the calculated transfer probability, the roulette wheel algorithm is used to randomly select the next mapping sub-area to be visited. That is, the transfer probability of each node is regarded as a sector area on the roulette wheel. The larger the probability, the larger the sector area. The selected node is then determined by randomly rotating the roulette wheel. The virtual drone continues to move to the next unvisited sub-area until all mapping sub-areas are visited. A global path planning is completed, and the total path length L or other evaluation indicators (such as total collection time, collection cost, etc.) of this path planning are recorded. Then, pheromone accumulation and update (including volatilization and pheromone enhancement operations) are performed, and the updated pheromone concentration is calculated. Then, multiple iterations are performed. After each iteration, the evaluation indicators of the current path planning are compared with the previously recorded optimal evaluation indicators. If the current result is better, the optimal path and optimal evaluation indicators are updated. When the number of iterations reaches the preset upper limit, the iteration is stopped, and the recorded optimal path is used as the final optimization result. This path is the recommended path for the drone swarm to perform 3D mapping of building facades.

[0052] The clustering sharing module 40 is used to control the drone group to perform three-dimensional mapping of the building facade based on the optimization result, establish a shared information pool according to the real-time mapping position clustering of the drone group, and upload the real-time mapping results to the shared information pool.

[0053] Preferably, the optimization result (optimal mapping path planning) is used to control the drone group to perform three-dimensional mapping of the building facade. Specifically, each drone starts from the collection starting point according to the path segment assigned to it and flies to each mapping sub-area (node ​​to be collected) in a predetermined order. For example, the drone collects images or data from sub-areas on different floors and facades of the building in turn according to the flight route determined by the optimization result, ensuring that the entire building facade can be fully and efficiently covered; and during the flight, the drone uses its own navigation system (such as GPS, inertial navigation, etc.) and flight control algorithm to accurately control the flight attitude and speed to fly according to the path of the optimization result. At the same time, it makes real-time adjustments according to the actual environmental conditions (such as wind speed, airflow, etc.) to ensure flight stability and safety. For example, when encountering strong winds, the drone automatically adjusts the flight parameters to remain on the predetermined mapping path.

[0054] Preferably, each drone is equipped with a positioning device (such as a GPS module) that can obtain its own location information (longitude, latitude, altitude, etc.) in real time. The location information is continuously updated as the drone flies and is transmitted to a ground control center or a designated server via a wireless communication module (such as 4G, 5G or Wi-Fi). After receiving the real-time location information of the drone swarm, the ground control center or server uses a clustering algorithm (such as the K-Means clustering algorithm, the DBSCAN clustering algorithm, etc.) to perform cluster analysis on these location points. The purpose is to group the spatially similar drone locations into a group, reflecting the concentration of drones in different areas. For example, if multiple drones are surveying and mapping near a specific floor or facade of a building, they are clustered into a cluster, indicating that the area is the focus of the current surveying and mapping. Based on the clustering results, a shared information pool is established, that is, a data storage and management platform for storing the real-time location clustering information of the drones and other related data (such as flight status, battery power, etc.), to ensure the collaborative operation of the drone swarm and enable each drone to obtain the location and working status information of other drones. Drones use the sensors they carry (such as high-definition cameras, lidar, etc.) to collect data on building facades (which may include image data, point cloud data, three-dimensional model data, etc., recording detailed information of the building facades, such as texture, shape, size, etc.). After preliminary processing (such as data compression, format conversion, etc.), the data is uploaded to a shared information pool in real time via a wireless communication link. The shared information pool stores and manages the real-time mapping results to ensure the security and accessibility of the data, thereby improving the efficiency of drone swarm mapping and the accuracy of data processing.

[0055] Furthermore, the specific configuration of the cluster sharing module 40 also includes obtaining the spatial position coordinates of each drone in the drone group at the current time node; obtaining the next round of mapping sub-area of ​​each drone in the drone group; performing real-time mapping position clustering according to the next round of mapping sub-area and the spatial position coordinates to generate a real-time mapping position clustering result; identifying the drone with the largest storage space in each cluster in the real-time mapping position clustering result, creating a shared information pool for the identified drone, and receiving the real-time mapping results of other drones in the real-time mapping position clustering result.

[0056] Preferably, each drone is equipped with a corresponding positioning device (such as GPS, Beidou satellite navigation system, etc.), which can determine its own position in space in real time and present it in the form of coordinates (such as longitude, latitude, altitude, etc.). Through wireless communication technology (such as 4G, 5G or dedicated drone communication link), these spatial position coordinate information are transmitted to the ground control center in real time to obtain the specific position of each drone in space at the current time node; clarify the sub-area (node ​​to be collected) that each drone will go to for mapping in the next round. For example, a drone has completed the mapping task of the current sub-area, and its next round of mapping is determined according to the preset path sequence. The next mapping sub-area is mapped so that the drone can continue its mapping work as planned. The obtained information of each drone's next mapping sub-area is then combined with the current spatial position coordinates. The drone's position is analyzed using a clustering algorithm (such as K-Means clustering, DBSCAN clustering, etc.). Drones that are close in spatial position or about to go to the same or similar mapping sub-area are grouped together to form different cluster sets. For example, if the next mapping sub-areas of multiple drones are all concentrated on a certain side of a certain floor of a building, and their current spatial positions are also relatively close, these drones are clustered into the same set, thereby generating a real-time mapping position clustering result.

[0057] Preferably, the storage space of the drones is compared to find the drone with the largest storage space and identify it. The drone with the largest storage space is selected as the information storage center of the cluster set, and a shared information pool is created. The other drones in the cluster set transmit their real-time mapping results (including collected image data, point cloud data, mapping parameters, etc.) to the shared information pool of the identified drone via wireless communication. The purpose is to centrally store and manage mapping data in a local area, reduce the burden of data transmission, and facilitate data sharing and collaborative operations between drones in the same cluster set, thereby improving the mapping efficiency and data processing capabilities of the entire drone group. For example, when there is doubt about the data collected by a certain drone, the mapping results of other drones at similar locations can be obtained from the shared information pool for comparative analysis.

[0058] The surveying and mapping optimization module 50 is used to perform collaborative decision optimization of the local path through the shared information pool and the optimization result, and update the acquisition path according to the collaborative decision optimization result.

[0059] The specific configuration of the surveying and mapping optimization module 50 further includes obtaining environmental shared information in a shared information pool, and reconstructing local environmental mutation constraints based on the environmental shared information; obtaining surveying and mapping anomalies based on real-time surveying and mapping results in the shared information pool, and creating additional surveying and mapping tasks using the surveying and mapping anomalies; and performing collaborative decision-making optimization of local paths based on the local environmental mutation constraints, the additional surveying and mapping tasks, and the optimization results.

[0060] Preferably, during the three-dimensional mapping of the building facade by the drone swarm, the local path is dynamically adjusted and optimized based on the data in the shared information pool and the previous optimization results to cope with environmental changes and abnormal situations occurring during the mapping process. Specifically, the environmental shared information in the shared information pool is obtained, such as meteorological conditions (wind speed, wind direction, temperature, etc.) and obstacle information (new temporary obstacles, changes in local building structures, etc.). Then, based on the environmental shared information, it is determined whether there is an environmental mutation, and the factors causing the environmental mutation are determined, and then the corresponding constraint parameters are determined. Finally, based on the determined constraint parameters, specific local environmental mutation constraint conditions are constructed. That is, if there are situations such as sudden strong winds or new obstacles, the local environmental mutation constraints are reconstructed. For example, for a sudden change in wind speed, the maximum flight speed allowed at the current wind speed, the flight altitude range, and the wind direction area to be avoided can be determined based on the performance parameters of the drone. For new obstacles, the scope of the no-fly zone around them must be clarified, that is, the area at a certain distance from the obstacle that the drone cannot enter. Local environmental mutation constraints cover various environmental changes that may affect UAV flight and mapping tasks, ensuring that the UAV can comply with these constraints during flight and complete the mapping tasks safely and efficiently.

[0061] Preferably, by analyzing the real-time mapping results in the shared information pool, abnormal situations that occur during the mapping process are discovered, such as blurred image data, missing point cloud data, and excessive deviation between mapping data and expectations. If mapping abnormalities are found, additional mapping tasks are created for these abnormal situations. For example, if the image data of a certain area is blurred, high-definition image acquisition of the area needs to be performed again. If point cloud data is missing, a laser scanning task for the area needs to be added. These additional mapping tasks are entered into the path planning to ensure the integrity and accuracy of the mapping data.

[0062] Preferably, the local environmental mutation constraints, additional mapping tasks, and optimization results are used as the basis for collaborative decision-making. The optimization results provide the initial global optimal path planning, while the local environmental mutation constraints and additional mapping tasks reflect the changes in the current actual situation. Collaborative decision-making optimization of the local path is performed, that is, adjustments are made on the basis of the original path planning. Specifically, a path optimization algorithm (such as an ant colony algorithm, a genetic algorithm, etc.) is used in combination with the above information to perform collaborative decision-making optimization of the local path, comprehensively considering the new constraints and additional tasks. Under the premise of meeting the environmental constraints, the additional mapping tasks are completed as efficiently as possible while minimizing the impact on the original planned path. For example, the algorithm calculates how to reasonably arrange drones to complete the additional mapping tasks while avoiding new obstacles, so as to minimize the overall mapping time and cost. Finally, based on the collaborative decision-making optimization results of the local path, the drone collection path is updated, and the optimized path information is sent to the corresponding drones so that they can perform mapping operations according to the new path. The updated collection path can better adapt to environmental changes and mapping anomalies, improve the efficiency and quality of mapping, and ensure the smooth completion of the mapping task.

[0063] Furthermore, the specific configuration of the drone and AI-based building facade three-dimensional mapping system also includes an anomaly recognition module, which is used to record the collection anomalies of the drone group and perform anomaly level judgment, match the response schemes in the response scheme library according to the anomaly level judgment results, and perform anomaly management according to the response scheme matching results. The response scheme library includes anomaly marking response schemes and temporary task creation response schemes.

[0064] Preferably, the anomaly recognition module is used to continuously monitor the collection work of the drone, record various possible collection anomalies, such as incomplete collection data, blurred images, data errors caused by sensor failure, abnormal distance to the target building, etc., and then classify the recorded collection anomalies according to their severity and the degree of impact on the surveying and mapping task. For example, slight image blur may be judged as a lower-level anomaly, while a complete sensor failure resulting in a large amount of data loss may be judged as a higher-level anomaly; then, for different anomaly levels, response plans are matched in the anomaly marking response plan library, where the response plan library contains anomaly marking response plans and temporary task creation response plans. Specifically, the anomaly marking response plan is mainly for abnormal situations. Mark and record the time, location, type, and other information of the anomaly, so that technicians can conduct detailed inspection and analysis after the task is completed; create a temporary task response plan based on the specific circumstances of the anomaly to solve the anomaly problem. For example, if the data collection of a certain area is missing due to a drone failure, a new task is created based on the anomaly level judgment result, and other drones are arranged to re-collect the area to ensure the integrity of the surveying and mapping data; finally, the anomaly is managed according to the matching results. If the anomaly marking response plan is matched, the anomaly is marked and recorded according to the plan; if the temporary task creation response plan is matched, the corresponding temporary task is automatically created and executed to solve the anomaly and ensure that the surveying and mapping task can continue smoothly. Through the anomaly identification module, various collection anomalies can be effectively managed and handled, improving the reliability and accuracy of drone group mapping.

[0065] In the above, refer to Figure 1 The following describes in detail the building facade 3D mapping system based on drone and AI according to the embodiment of the present invention. Figure 2 A method for three-dimensional mapping of building facades based on drones and AI according to an embodiment of the present invention is described.

[0066] 3D mapping method of building facade based on drone and AI, such as Figure 2As shown, the method includes: reading the calibration scene information of the scene to be surveyed, performing information analysis of the calibration scene information, establishing N surveying sub-areas, and marking each surveying sub-area as a node to be collected; after distributing the collection starting point of the drone group, allocating an initial pheromone value on each surveying path, configuring an inspiration factor, and calculating the node transfer probability of each node to be collected; configuring a virtual drone to start from the collection starting point, calculating the pheromone concentration according to the initial pheromone value, and using the pheromone concentration and the inspiration factor to iteratively select the surveying sub-area to establish an optimization result, wherein, after each global path planning is completed, the initial pheromone value is accumulated and updated; based on the optimization result, the drone group is controlled to perform three-dimensional surveying of the building facade, and a shared information pool is established according to the real-time surveying position clustering of the drone group, and the real-time surveying result is uploaded to the shared information pool; collaborative decision optimization of the local path is performed through the shared information pool and the optimization result, and the collection path is updated according to the collaborative decision optimization result.

[0067] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades also includes: a drawing reading module, used to read the scene drawings of the scene to be mapped, and use the scene drawings as calibration scene information; a grid segmentation module, used to establish a median grid granularity, perform grid division of the building facade of the scene to be mapped based on the median grid granularity, and generate an initial grid segmentation result; a grid update module, used to identify the integrity of elements within the grid on the initial grid segmentation result, perform grid migration clustering based on the integrity identification result, generate a grid migration clustering result, and establish N mapping sub-areas based on the grid migration clustering result.

[0068] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades further includes: obtaining a set of features not covered by the initial grid segmentation result and marking it as an updated feature set; obtaining dimension data of each feature in the updated feature set, and selecting clustering features based on the matching degree between the dimension data and the median grid granularity; and using the clustering features to complete grid migration clustering.

[0069] In one possible implementation, the drone- and AI-based 3D mapping method for building facades further includes: establishing a collection node transfer cost function as follows:

[0070] ;

[0071] in, Characterization from the acquisition node Transfer to the collection node The acquisition node transfer cost function, Characterization acquisition node and collection nodes distance, and Characterize the acquisition nodes separately and collection nodes risk factors, and Characterize the acquisition nodes separately and collection nodes The node priority, 、 、 They are distance weight factor, risk weight factor, and node priority weight factor respectively; according to the acquisition node transfer cost function, the heuristic factor is configured, and the node transfer probability is calculated using the heuristic factor as follows:

[0072] ;

[0073] in, Characterization from the acquisition node To the collection node The node transition probability, Characterization acquisition node To the collection node The pheromone concentration, Is the heuristic factor, specifically representing the collection node To the collection node Heuristic information, , Characterizes the set of optional collection nodes that the drone has not visited, Represents any optional acquisition node, Characterization acquisition node To the collection node The pheromone concentration, Characterization acquisition node To the collection node Heuristic information, Characterize pheromone influencing factors, Characterize the heuristic factors affecting the factors.

[0074] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades also includes: obtaining the spatial position coordinates of each drone in the drone swarm at the current time node; obtaining the next-round mapping sub-area of ​​each drone in the drone swarm; performing real-time mapping position clustering based on the next-round mapping sub-area and the spatial position coordinates to generate a real-time mapping position clustering result; identifying the drone with the largest storage space in each cluster in the real-time mapping position clustering result, creating a shared information pool for the identified drones, and receiving the real-time mapping results of other drones in the real-time mapping position clustering result.

[0075] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades further includes: obtaining environmental shared information in a shared information pool, and reconstructing local environmental mutation constraints based on the environmental shared information; obtaining mapping anomalies based on real-time mapping results in the shared information pool, and creating additional mapping tasks using the mapping anomalies; and performing collaborative decision-making optimization of local paths based on the local environmental mutation constraints, the additional mapping tasks, and the optimization results.

[0076] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades also includes: dynamically acquiring a real-time microscopic image of a target magnetic floc in the target water body; performing feature evaluation and analysis on the real-time microscopic image according to a microscopic state evaluation mechanism to obtain a real-time microscopic state; obtaining an arbitrary information value coefficient of the real-time microscopic state for the arbitrary water quality index; predicting and analyzing the real-time water quality parameters in combination with the arbitrary information value coefficient to obtain predicted real-time water quality parameters; and calibrating the real-time water quality parameters with the predicted real-time water quality parameters.

[0077] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades further includes: collecting real-time image feature information of the real-time microscopic image; normalizing and weighting the real-time image color parameters and the real-time image structure parameters in the real-time image feature information to obtain a real-time image feature value; and performing standardized correction on the real-time image feature value in combination with the microscopic state coefficient stored in the microscopic state evaluation mechanism to obtain the real-time microscopic state.

[0078] In one possible implementation, the drone- and AI-based three-dimensional mapping method for building facades also includes: extracting a first historical data set from a magnetic coagulation sedimentation database; performing mutual information analysis on a first historical microscopic state of a first magnetic floc in the first historical data set and a first historical indicator parameter corresponding to a first historical arbitrary water quality indicator to obtain a first historical mutual information coefficient; and using the first historical mutual information coefficient as the arbitrary information value coefficient.

[0079] The drone- and AI-based three-dimensional building facade surveying and mapping system provided in an embodiment of the present invention can execute the drone- and AI-based three-dimensional building facade surveying and mapping method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0080] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0081] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. The 3D mapping system for building facades based on drones and AI is characterized by: The system comprises: A scene analysis module is used to read the calibration scene information of the scene to be surveyed, perform information analysis on the calibration scene information, establish N survey sub-areas, and mark each survey sub-area as a node to be collected; The collection planning module is used to distribute the initial pheromone value on each mapping path after distributing the collection starting point of the drone group, configure the heuristic factor, and calculate the node transition probability of each node to be collected. The calculation of the node transition probability of each node to be collected includes: Establish the acquisition node transfer cost function as follows: ； in, Characterization from the acquisition node Transfer to the collection node The acquisition node transfer cost function, Characterization acquisition node and collection nodes distance, and Characterize the acquisition nodes separately and collection nodes risk factors, and Characterize the acquisition nodes separately and collection nodes The node priority, 、 、 They are distance weight factor, risk weight factor, and node priority weight factor respectively; The heuristic factor is configured according to the acquisition node transfer cost function, and the node transfer probability is calculated using the heuristic factor as follows: ； in, Characterization from the acquisition node To the collection node The node transition probability, Characterization acquisition node To the collection node The pheromone concentration, Is the heuristic factor, specifically representing the collection node To the collection node Heuristic information, , Characterizes the set of optional collection nodes that the drone has not visited, Represents any optional acquisition node, Characterization acquisition node To the collection node The pheromone concentration, Characterization acquisition node To the collection node Heuristic information, Characterize pheromone influencing factors, Characterize the heuristic factors affecting factors; a surveying and searching module configured to configure a virtual drone to depart from the acquisition starting point, calculate the pheromone concentration based on the initial pheromone value, and iteratively select and map sub-regions using the pheromone concentration and the heuristic factor to establish an optimal search result, wherein the initial pheromone value is cumulatively updated after each completion of global path planning; A clustering and sharing module is used to control the drone swarm to perform three-dimensional mapping of the building facade based on the optimization result, establish a shared information pool based on the real-time mapping position clusters of the drone swarm, and upload the real-time mapping results to the shared information pool; The surveying and mapping optimization module is used to perform collaborative decision-making optimization of the local path through the shared information pool and the optimization results, and update the acquisition path according to the collaborative decision-making optimization results.

2. The drone-based AI-based 3D mapping system for building facades according to claim 1, characterized in that: The scene parsing module includes: A drawing reading module is used to read the scene drawing of the scene to be measured and mapped, and use the scene drawing as calibration scene information; A grid segmentation module is used to establish a median grid granularity, perform grid division of the building facade of the scene to be measured based on the median grid granularity, and generate an initial grid segmentation result; The grid updating module is used to identify the integrity of elements within the grid based on the initial grid segmentation result, perform grid migration clustering based on the integrity identification result, generate grid migration clustering results, and establish N surveying and mapping sub-areas based on the grid migration clustering results.

3. The drone-based AI-based 3D mapping system for building facades according to claim 2, wherein: In the grid update module, grid migration clustering is performed according to the integrity identification result, including: Obtain the feature set that is not covered by the initial grid segmentation result and mark it as the updated feature set; Obtaining size data of each element in the update element set, and selecting cluster elements according to a matching degree between the size data and the median grid granularity; The clustering elements are used to complete grid migration clustering.

4. The drone-based AI-based 3D building facade mapping system according to claim 1, wherein: In the cluster sharing module, a shared information pool is established based on the real-time mapping position clustering of the drone group, including: Get the spatial position coordinates of each drone in the drone swarm at the current time node; Get the next round of mapping sub-area for each drone in the drone swarm; Performing real-time mapping position clustering according to the next round of mapping sub-areas and the spatial position coordinates to generate a real-time mapping position clustering result; Identify the drone with the largest storage space in each cluster in the real-time mapping location clustering result, create a shared information pool among the identified drones, and receive the real-time mapping results of other drones in the real-time mapping location clustering result.

5. The drone and AI-based building facade 3D mapping system according to claim 1, wherein: In the surveying and mapping optimization module, collaborative decision-making optimization of local paths is performed by sharing the information pool and the optimization results, including: Acquire environmental shared information in a shared information pool, and reconstruct local environmental mutation constraints based on the environmental shared information; Acquire surveying and mapping anomalies based on the real-time surveying and mapping results in the shared information pool, and create additional surveying and mapping tasks using the surveying and mapping anomalies; Collaborative decision optimization of the local path is performed according to the local environmental mutation constraint, the additional surveying and mapping task, and the optimization result.

6. The drone and AI-based building facade 3D mapping system according to claim 1, wherein: The system further comprises: The anomaly recognition module is used to record the collection anomalies of the drone group and perform anomaly level judgment. According to the anomaly level judgment results, the response plan library is matched, and the anomaly management is performed according to the response plan matching results. The response plan library includes anomaly marking response plans and temporary task creation response plans.

7. The three-dimensional mapping method of building facades based on drones and AI is characterized by: The method is applied to the drone- and AI-based three-dimensional mapping system for building facades according to any one of claims 1 to 6, and the method comprises: Read the calibration scene information of the scene to be surveyed, perform information analysis on the calibration scene information, establish N survey sub-areas, and mark each survey sub-area as a node to be collected; After distributing the collection starting points of the drone swarm, the initial pheromone value is assigned on each mapping path, and the heuristic factor is configured to calculate the node transfer probability of each node to be collected; configuring a virtual drone to depart from the collection starting point, calculating the pheromone concentration based on the initial pheromone value, and iteratively selecting and mapping sub-areas using the pheromone concentration and the heuristic factor to establish an optimization result, wherein the initial pheromone value is cumulatively updated after each completion of global path planning; Based on the optimization result, the drone swarm is controlled to perform three-dimensional mapping of the building facade, and a shared information pool is established according to the real-time mapping position clustering of the drone swarm, and the real-time mapping results are uploaded to the shared information pool; The collaborative decision-making optimization of the local path is performed by sharing the information pool and the optimization result, and the collection path is updated according to the collaborative decision-making optimization result.

Images