A method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes

Abstract

The present invention discloses a method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes, comprising the following steps: S1, data collection and preprocessing: collecting target area data based on a collection mechanism based on entity and element type priority, and preprocessing the collected data to obtain model data of the target area; S2, change detection: comparing the newly collected model data with the historical model data, and identifying the changed area and type; S3, entity and element extraction and updating: processing the entities and elements of the changed area respectively to obtain an updated model; S4, model fusion and optimization: fusing the updated entity and element model with the original model. The present application realizes efficient collection and preprocessing of target area data by utilizing multi-source surveying and mapping equipment and a collection mechanism based on entity and element type priority; by dividing small areas according to regional characteristics and correcting the collection frequency, resource allocation is optimized, and the pertinence and efficiency of data collection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional model updating, and in particular to a method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes. Background Art

[0002] In the field of geographic information systems (GIS) and real-world 3D modeling, with the acceleration of urbanization and the continuous changes in urban appearance, there is an increasing demand for dynamic updates of terrain-level real-world 3D entities and elements. Existing technical solutions often use static or periodic data update methods, which cannot meet the requirements of real-time and accuracy. These methods are inefficient when processing large-scale data and have difficulty adapting to rapidly changing environments.

[0003] In addition, traditional data collection and update methods lack consideration of the priority of changes in different entities and elements, resulting in irrational resource allocation and inability to efficiently respond to changes in key areas. Therefore, a technical solution is needed that can respond to environmental changes in real time, process data efficiently, and dynamically adjust update strategies based on the importance of entities and elements. Summary of the invention

[0004] The purpose of the present invention is to provide a method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes, characterized by comprising the following steps:

[0007] S1. Data collection and preprocessing: prioritize the target areas, obtain the collection frequency of the target areas according to the priorities of the target areas, collect data from the target areas according to the collection frequency of the target areas, and preprocess the collected data to obtain the model data of the target areas;

[0008] S2, Change Detection: Compare the newly collected model data with the historical model data to identify the changed areas and types;

[0009] S3, entity and element extraction and update: the entities and elements in the changed area are processed separately to obtain the updated model;

[0010] S4. Model fusion and optimization: merge the updated entity and element models with the original models.

[0011] As a further solution of the present invention: in S1, data collection is performed through multi-source mapping, and the multi-source mapping includes satellite remote sensing images, aerial photogrammetry, ground laser scanning, and mobile surveying vehicles, and the mapping data of the target area includes topography and physical information of land objects;

[0012] Preprocessing includes data format conversion, coordinate system unification, noise removal, data cropping and splicing.

[0013] As a further solution of the present invention: the data collection comprises the following steps:

[0014] S1.1. Prioritize the target areas. The priority is divided by classifying the roles of the entities and elements of the target areas in the entire regional functional system;

[0015] S1.2, obtain the corresponding theoretical acquisition frequency through the priority of the target area;

[0016] S1.3, divide the target area into small areas;

[0017] S1.4. Correct the acquisition frequency of all small areas in the target area to obtain the actual acquisition frequency of all small areas, and collect data for each small area in the target area according to the actual acquisition frequency.

[0018] As a further solution of the present invention: the target area priority division in S1.1 is achieved by dividing the roles of various entities and elements in the target area in the entire regional functional system into levels, including the first level, the second level, and the third level. The first level includes: important transportation facilities and key infrastructure; the second level includes: general buildings, ordinary roads; the third level includes: natural vegetation and small landscape facilities.

[0019] As a further solution of the present invention: the acquisition of the theoretical acquisition frequency of the target area in S1.2 comprises the following steps:

[0020] S1.2.1. The average rating for the region is calculated by summing the ratings of all entities and elements in the region and dividing by the number of entities and elements included in the calculation;

[0021] S1.2.2. Analyze historical data and industry development trends to estimate the frequency of change of different types of entities and elements in the entire region; calculate the mean frequency of change of all types of entities and elements and record it as the trend value;

[0022] S1.2.3. After normalizing the obtained regional average grade score and trend value, the regional average grade score and trend value are used as the radius of the two circles respectively, and the preset weight factor is set as high. A frustum model is established by combining the two established circles, and the volume of the frustum is calculated and recorded as the optimal judgment value. This optimal judgment value is used as the standard for measuring the priority of the entities and elements in the entire region;

[0023] S1.2.4. Compare the obtained optimal judgment value with the preset optimal judgment value interval. Different optimal judgment value intervals correspond to different collection frequencies. When the optimal judgment value interval to which the optimal judgment value belongs is determined, the theoretical collection frequency corresponding to the target area is determined.

[0024] As a further solution of the present invention: the small area division in S1.3 includes division based on geographical features, division based on functional zoning, and division combined with population density distribution;

[0025] Division based on geographical features: Analyze the topography and landforms in the region, including natural geographical elements such as mountains, rivers, and lakes; divide the two sides of the mountains into different small areas; for river areas, divide the river core influence area and transition area into small areas based on the width, flow rate, and whether it is a seasonal river;

[0026] Division based on functional zoning: According to the functional zoning in urban planning, including commercial areas, residential areas, industrial areas, cultural and educational areas; the entities and elements of different functional areas have different changes in emphasis and frequency, and small areas are divided based on this;

[0027] Division based on population density distribution: including densely populated areas and sparsely populated areas.

[0028] As a further solution of the present invention: the acquisition frequency correction of the small area in S1.4 is achieved by obtaining the activity of the entity and element changes in the small area. The higher the activity of the small area, the higher the acquisition frequency.

[0029] When the change activity of entities and elements in a small area is more than 50% higher than the average level of the whole area, the correction coefficient of its collection frequency is set to be greater than 1 to increase the collection frequency. When the change activity of a small area is less than 30% lower than the average level of the whole area, the correction coefficient is set to be less than 1 to reduce the collection frequency.

[0030] The activity of a small area is determined by comparing the historical data of the small area with the average change data of the overall area, counting the physical change data of the number of newly built buildings and the length of road reconstruction in the small area in the past few years, and comparing them with the corresponding data of the overall area; when the growth rate of the number of newly built buildings in a small area exceeds the average level of the overall area, it is judged that the small area is in the development stage, and its physical and element changes are more active.

[0031] As a further solution of the present invention: the change detection in S2 adopts a change detection algorithm based on feature matching, image difference, and semantic segmentation, and weights are assigned to the results of the three algorithms. A voting mechanism is used for comprehensive judgment. If the majority of algorithms determine that the area is a changed area, the change is determined, and then the change type is determined based on the details displayed by each algorithm;

[0032] Preliminary positioning based on feature matching: Assume that the set of feature points extracted from the newly collected data is , the set of feature points extracted from historical data is ; Feature points The total number of feature points is n. The total number of feature points is m. and , calculate the eigenvector distance between them , using Euclidean distance, ;

[0033] in and The feature points are and No. dimensional feature vector components, is the feature vector dimension; set the distance threshold , if exists Make , then it is believed that and Matching, the number of matching feature points is recorded as ; Define the matching ratio ,when When , the area is determined as a change area, where is the matching ratio threshold;

[0034] Image difference detail analysis: Assume that the newly acquired data is , the historical data is , after registration, for the image data, calculate the pixel difference ,in is the pixel coordinate; for point cloud data, calculate the point With corresponding point The distance difference ; Set the difference threshold ,like or , then the pixel or point is marked as a potential change point, and the total number of change points is recorded as ; Define the change ratio ,in is the total number of pixels or points in the data, combined with the feature matching results, when and When further analyzing whether there are interference factors in the area;

[0035] Semantic segmentation category determination: Assume that the classification result of the deep learning semantic segmentation model for the newly collected data is , the classification result of historical data is ,in and Indicates the semantic category of the corresponding position, Index variable used to traverse the classification result set of newly collected data And historical data classification result set Elements in ; define the proportion of category changes ,in ,when When is the threshold of semantic category change ratio, and a comprehensive judgment is made based on the results of the first two algorithms: s is the classification result set of the newly collected data And historical data classification result set The total number of elements in:

[0036] Comprehensive judgment and fusion results: Assign weights to the results of the three algorithms, and use a voting mechanism to make a comprehensive judgment. If the majority of algorithms determine that the area is a change, then the change is confirmed. Then, the type of change is determined based on the details displayed by each algorithm. The change type includes addition, deletion, or modification.

[0037] As a further solution of the present invention: the entity and element extraction and updating in S3, for the newly added entities and elements, the geometric shape, position information, and texture features of the newly added entities and elements are extracted from the newly collected data using 3D modeling technology, including point cloud modeling, texture mapping, and semantic modeling, and the corresponding 3D entity model is constructed and added to the terrain-level real-scene 3D model;

[0038] For deleted entities and elements, the corresponding model data is directly deleted in the terrain-level real-scene 3D model;

[0039] For the modified entities and elements, the geometric shapes, positions, and texture attributes of the corresponding entities and elements in the terrain-level real-scene 3D model are updated and modified according to the change information in the newly collected data;

[0040] When adding or modifying entities and elements, an incremental modeling mechanism is used. The process is as follows:

[0041] When new collected data arrives, local modeling and updates are only performed on the newly added or modified parts; by establishing a data index and a changed area identification mechanism, the entities and elements that need to be updated are located, and then the existing model data is used as a reference and constraint to build a new local model and seamlessly integrate it with the original model.

[0042] As a further solution of the present invention: S4 fuses the updated entities and elements with the original terrain-level real-scene 3D model, and determines whether the fused model is qualified by comprehensively analyzing the evaluation parameters of the fused model, and the evaluation parameters include:

[0043] Average distance error: Calculate the average distance error between the updated entities and features and the corresponding parts of the original model. For each point in the model , s is the number of points in the model; find the nearest point in the original model , calculate the distance , where represents the Euclidean distance, then the average distance error ;

[0044] Shape similarity index: The shape similarity is calculated using a shape descriptor-based method and recorded as ;

[0045] Texture mean square error: For the fused texture image, sample points are selected at the boundary and inside of the update area and the original area, and the updated texture pixel value is set to , the original texture pixel value is ,in For pixel coordinates, calculate the texture mean square error , where is the total number of sample points;

[0046] Texture correlation coefficient: Calculate the correlation coefficient between the texture of the updated area and the original area. Suppose the texture pixel value sequence of the updated area is , the original area texture pixel value sequence is , correlation coefficient , where and are the means of the two sequences, texture correlation coefficient The value range is ;

[0047] Semantic category consistency ratio: Count the number of entities and elements with consistent semantic categories in the fused model and record it as ;

[0048] Then the average distance error is obtained , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio After normalization, the following formula is entered: To obtain the overall judgment value , where The average distance error , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio The preset weight coefficient; and the obtained comprehensive judgment value As a criterion for measuring whether the fused model is qualified;

[0049] The obtained comprehensive judgment value Compare with the preset comprehensive judgment threshold. If it is greater than the preset comprehensive judgment threshold, the fused model is judged to be qualified, otherwise, it is judged to be unqualified and the fused model is optimized, including model simplification, compression, and topological relationship repair.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. This application realizes efficient data collection and preprocessing of target area data by utilizing multi-source surveying and mapping equipment and a collection mechanism based on entity and feature type priority. It not only ensures data consistency and availability, but also divides small areas according to regional characteristics and corrects the collection frequency, optimizes resource allocation, and improves the pertinence and efficiency of data collection. Priority collection of high-level entities such as important transportation facilities and key infrastructure ensures real-time updating of key data, thereby improving the accuracy and practicality of overall data collection.

[0052] 2. This application uses feature matching, image difference, and semantic segmentation algorithms to automatically identify the changed areas and types, and significantly improves the accuracy of change detection by combining fusion judgment with manual verification; the comprehensive use of multiple algorithms reduces the misjudgment that may be caused by a single algorithm, and further ensures the reliability of change detection results through manual verification. In addition, the automated change detection process reduces manual intervention, improves work efficiency, and reduces costs;

[0053] 3. This application improves the efficiency of entity and feature extraction and updating through incremental mechanism modeling and real-time rendering and modeling collaborative technology; this method only performs local modeling and updating for the changed area, greatly reducing the amount of data processing and modeling time, and is particularly suitable for frequent update scenarios of large-scale terrain-level real-scene 3D models; at the same time, the fusion model is judged to be qualified through comprehensive evaluation parameters, and necessary optimization operations are performed to ensure the quality of the fusion model. This not only improves the storage efficiency, rendering efficiency and visualization effect of the model, but also ensures the accuracy and integrity of the model, providing users with high-quality real-scene 3D models. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1A flow chart of the method of this application; DETAILED DESCRIPTION

[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] See also Figure 1 In an embodiment of the present invention, a method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes includes the following steps:

[0057] S1. Data collection and preprocessing: Multi-source surveying and mapping equipment is used to collect target area data according to the collection mechanism based on entity and feature type priority. Preprocessing ensures data consistency and availability. Small areas are divided according to regional characteristics and the collection frequency is corrected.

[0058] Through multi-source surveying and mapping data acquisition including satellite remote sensing images, aerial photogrammetry, ground laser scanning, and mobile surveying vehicles, the surveying and mapping data of the target area include topography, landforms, and ground entity information. The collected multi-source surveying and mapping data are pre-processed, including data format conversion, coordinate system unification, noise removal, data cropping and splicing operations, to ensure data consistency and availability.

[0059] The priority collection mechanism includes the following steps:

[0060] S1.1. Prioritize the target areas. The priority is divided by classifying the roles of the entities and elements of the target areas in the entire regional functional system;

[0061] The priority division of target areas is achieved by dividing the roles of various entities and elements in the target areas in the entire regional functional system into different levels. Each level corresponds to a different level score. The first level includes: important transportation facilities (highways, railways, main urban roads, airport runways, etc.), key infrastructure (nuclear power plants, large reservoirs and dams, important communication hubs, etc.); the second level includes: general buildings (ordinary residences, commercial shops, etc.), ordinary roads (community roads, secondary branches, etc.); the third level includes: natural vegetation (ordinary forests, grasslands, etc.) and small landscape facilities (park sketches, street greening decorations, etc.).

[0062] S1.2. Obtaining the corresponding theoretical acquisition frequency through the priority of the target area. Obtaining the theoretical acquisition frequency of the target area includes the following steps:

[0063] S1.2.1. The average rating for the region is calculated by summing the ratings of all entities and elements in the region and dividing by the number of entities and elements included in the calculation;

[0064] S1.2.2. Analyze historical data and industry development trends to estimate the frequency of change of different types of entities and elements in the entire region; calculate the mean frequency of change of all types of entities and elements and record it as the trend value;

[0065] S1.2.3. After normalizing the obtained regional average grade score and trend value, the regional average grade score and trend value are used as the radius of the two circles respectively, and the preset weight factor is set as the high value. The weight factor is the weight factor for calculating the optimal value. Combine the two established circles to establish a frustum model, calculate the volume of the frustum, and record it as the optimal value. This optimal value is used as the standard for measuring the priority of the entire regional entities and elements;

[0066] S1.2.4. Compare the obtained optimal judgment value with the preset optimal judgment value interval. Different optimal judgment value intervals correspond to different collection frequencies. When the optimal judgment value interval to which the optimal judgment value belongs is determined, the theoretical collection frequency corresponding to the target area is determined.

[0067] S1.3. Divide the target area into small areas based on geographical features, functional zoning, and population density distribution:

[0068] Division based on geographical features: Analyze the topography and landforms in the region, such as mountains, rivers, lakes and other natural geographical elements. Use the mountains as the boundary to divide the two sides of the mountains into different small areas, because mountains may affect the direction of transportation routes, the degree of construction and development, and the distribution of vegetation and other entities and elements. For the areas around the river, divide the river core influence area (such as a certain range on both sides of the river bank), transition area and other small areas according to factors such as the width of the river, the flow rate, and whether it is a seasonal river. For example, in the core influence area of ​​the river, due to the possible existence of water conservancy facilities construction, riverbank erosion protection projects, etc., the changes in related entities and elements are quite different from those in areas far away from the river. Consider geological conditions, such as soft soil geology and rocky geological distribution areas. Soft soil geology areas may be more prone to problems such as ground subsidence, which affects the stability of buildings and road conditions. They are divided into independent small areas for focused attention;

[0069] Division based on functional zoning: Division is based on the functional zoning in urban planning, such as commercial area, residential area, industrial area, cultural and educational area, etc. Due to frequent commercial activities, the commercial area has large changes in store decoration, sign replacement, and personnel and logistics facilities; the residential area is mainly for residents' lives, and the changes are relatively concentrated in house decoration, community internal facilities renewal, etc.; the industrial area has a large number of factory construction, production line transformation, logistics channel adjustment, etc.; the cultural and educational area focuses on school building expansion, campus facility renewal, etc. The entities and elements of different functional areas have different changes in focus and frequency, and small areas are divided based on this.

[0070] S1.4. Correct the acquisition frequency of all small areas in the target area to obtain the actual acquisition frequency of all small areas, and collect data for each small area in the target area according to the actual acquisition frequency:

[0071] For each divided small area, its historical data is compared with the average change data of the whole area. The physical change data of the number of newly built buildings and the length of road reconstruction in the small area in the past few years are counted and compared with the corresponding data of the whole area. When the growth rate of the number of newly built buildings in a small area exceeds the average level of the whole area and is greater than the threshold of the difference, it is judged that the small area is in a rapid development stage, and its physical and element changes are more active.

[0072] Analyze the difference between the monitoring data in the small area (such as the ground subsidence rate and building tilt data monitored by sensors) and the monitoring data of the whole area. If the average ground subsidence rate of a small area is several times that of the whole area, it indicates that the geological stability of the area is poor, and the related infrastructure and buildings face a greater risk of change. In other words, the entities and elements in the small area are more active, and the correction coefficient needs to be determined based on the comparison results of the acquired data;

[0073] When the change activity of entities and elements in a small area is more than 50% higher than the average level of the whole area, the correction coefficient of its collection frequency is set to be greater than 1, and 1.5-1.8 is used, that is, the collection frequency is increased to 1.5-1.8 times the original. Conversely, if the change activity of a small area is less than 30% lower than the average level of the whole area, the correction coefficient is set to less than 1, and 0.5-0.6 is used to reduce the collection frequency.

[0074] S2. Change detection: Compare the newly collected model data with the historical model data to identify the changed area and type. Change detection uses a change detection algorithm based on feature matching, image difference, and semantic segmentation. The three algorithm results are weighted and a voting mechanism is used for comprehensive judgment. If the majority of algorithms determine that the area is a changed area, the change is determined. Then the change type is determined based on the details displayed by each algorithm.

[0075] The pre-processed newly collected data is compared and analyzed with the historical terrain-level real-scene 3D model data. The change detection algorithm based on feature matching, image difference, and semantic segmentation is used to automatically identify the changed areas of entities and elements, including changes of addition, deletion, and modification types; the results of the three change detection algorithms are fused. The specific process is as follows:

[0076] Preliminary positioning based on feature matching: Assume that the set of feature points extracted from the newly collected data is , the set of feature points extracted from historical data is , feature points The total number of feature points is n. The total number of feature points is m. and , n is the number of feature points extracted from the newly collected data, that is, the feature point set The number of elements in, m is the number of feature points extracted from historical data, that is, the feature point set The number of elements in ; calculate the eigenvector distance between them (For example, Euclidean distance, in and They are feature points and No. dimensional feature vector components, is the feature vector dimension). Set the distance threshold , if exists Make , then it is believed that and Matching, the number of matching feature points is recorded as . Define the matching ratio ,when ( is the matching ratio threshold), the area is initially determined to be a changed area.

[0077] Image difference detail analysis: It is used to accurately determine the changed area and eliminate the misjudgment caused by errors or interference that may occur during data acquisition or processing. Suppose the newly acquired data is , the historical data is , after registration, for the image data, calculate the pixel difference ,in is the pixel coordinate; for point cloud data, calculate the point With corresponding point The distance difference ; Set the difference threshold ,like or , then the pixel or point is marked as a potential change point, and the total number of change points is recorded as ; Define the change ratio ,in is the total number of pixels or points in the data, combined with the feature matching results, when and When , it means that although there are fewer feature matches in this area (indicating possible changes), and the image difference analysis also shows that there are many potential change points, it is necessary to further analyze whether there are interference factors in this area.

[0078] Semantic segmentation category determination: Assume that the classification result of the deep learning semantic segmentation model for the newly collected data is , the classification result of historical data is ,in and Indicates the semantic category of the corresponding position, Index variable used to traverse the classification result set of newly collected data And historical data classification result set Elements in ; define the proportion of category changes ,in ,in is the threshold of semantic category change ratio, and a comprehensive judgment is made based on the results of the first two algorithms: s is the classification result set of the newly collected data And historical data classification result set The total number of elements in , it indicates that there is a significant difference in the semantic categories of the newly collected data and the historical data in the area. At this time, it is necessary to combine the results of the first two algorithms (preliminary positioning based on feature matching and image difference detail analysis) for comprehensive judgment. This comprehensive judgment helps to more accurately determine the type of changed area (such as addition, deletion or modification), avoid misjudgment caused by a single algorithm, and improve the accuracy and reliability of change detection;

[0079] Comprehensive judgment and fusion results: Assign weights to the three algorithm results and use a voting mechanism to make a comprehensive judgment. The process is as follows:

[0080] Let the result of judging the changed area based on feature matching be F, the result of judging the changed area based on image difference be D, and the result of judging the changed area based on semantic segmentation be S; when the algorithm judges that the area is a changed area, the corresponding value is 1, otherwise it is 0;

[0081] The comprehensive judgment result C can be calculated by the following formula: ,in, are the weight coefficients of feature matching, image difference, and semantic segmentation results in comprehensive judgment, and ;like (T is the set comprehensive judgment threshold, the value range is between 0 and 1), then the area is judged as a change area;

[0082] When determining the type of change, the type of change is determined according to each algorithm;

[0083] In this embodiment, the specific process of determining the change type is as follows:

[0084] Feature matching algorithm:

[0085] Newly added: When the newly collected data has no matching feature points and is clustered and continuous, with independent spatial relationships and no interference factors, such as new building outline feature points on an open space, it is judged as newly added;

[0086] Deletion: If the feature points of historical data have no matches in the new data, the corresponding area is missing or empty, the surrounding spatial relationship has changed, and other algorithms have also changed accordingly, such as the missing of road feature points and the changes in image and semantics, it is considered deleted;

[0087] Modification: Some feature points can be matched but the relative position relationship has changed, or the distribution density of feature points has changed. At the same time, the image difference has changed and the semantic segmentation category has not changed significantly. For example, the corner points of buildings are offset or the number of road feature points increases and other algorithms meet the requirements, which is considered as modification.

[0088] Image difference algorithm:

[0089] Newly added: If the pixel values ​​in a certain area of ​​the image data change regularly with clear boundaries, or if bulges or new clusters appear in the point cloud data, and if they are confirmed by other algorithms, such as features and semantics, they are considered newly added.

[0090] Deletion: If the pixel difference in the image data area is low in brightness or simplistic with a clear boundary, or the point cloud in the point cloud data area is missing or has a low density, and if the feature points of other algorithms disappear or the semantic category changes, it will be deleted;

[0091] Modification: The pixels in the image data area change locally and are related to the physical structure, the point cloud range of the corresponding area of ​​the point cloud data is adjusted, the positions of some feature matching points change, and the semantic segmentation categories remain unchanged, which is judged as modification;

[0092] Semantic Segmentation Algorithm:

[0093] Newly added: The semantic category of the newly collected data area has been added from scratch, and the feature matching and image difference also show the newly added features, which is judged as newly added;

[0094] Deletion: If the semantic category of the historical data area disappears and the feature matching and image difference change accordingly, it is considered deleted;

[0095] Modification: If the semantic category changes, the feature matching changes, and the image difference is adjusted accordingly, it is judged as modification;

[0096] Finally, the change detection results are manually checked and corrected to remove misjudged change information and ensure the accuracy of change detection.

[0097] S3, Entity and feature extraction and update: Process the entities and features in the changed area separately to obtain the updated model; Entity and feature extraction and update, process the newly added, deleted, and modified entities and features in the changed area separately, and use incremental mechanism modeling and real-time rendering and modeling collaborative technology to improve the update efficiency and effect;

[0098] For the newly added entities and elements in the changed area, 3D modeling technology, including point cloud modeling, texture mapping, and semantic modeling, is used to extract the geometric shape, position information, and texture features of the newly added entities and elements from the newly collected data, and build the corresponding 3D entity model, which is added to the terrain-level real-scene 3D model; for the deleted entities and elements in the changed area, the corresponding model data is directly deleted in the terrain-level real-scene 3D model; for the modified entities and elements, the geometric shape, position, and texture attributes of the corresponding entities and elements in the terrain-level real-scene 3D model are updated and modified according to the change information in the newly collected data.

[0099] When adding or modifying entities and elements, an incremental modeling mechanism is used. The process is as follows:

[0100] When new collected data arrives, instead of remodeling the entire changed area, only local modeling and updating are performed for the newly added or modified parts; by establishing a data index and a change area identification mechanism, the entities and elements that need to be updated are located, and then the existing model data around them is used as a reference and constraint condition to build a new local model and seamlessly integrate it with the original model; this method greatly reduces the amount of data processing and modeling time, and is especially suitable for frequent update scenarios of large-scale terrain-level real-life 3D models, and can achieve near-real-time dynamic update effects; using real-time rendering and modeling collaborative technology, a preliminary rendering preview is performed during the modeling process, allowing modelers or automated algorithms to promptly discover problems in model construction (such as texture stretching, geometric shape anomalies, etc.) and make adjustments. At the same time, using cloud computing or edge computing resources, the modeling and rendering tasks are distributed to multiple computing nodes for parallel processing, further improving the speed of modeling and updating, and ensuring that the updated model is provided to users for visualization and analysis in the shortest time.

[0101] S4. Model fusion and optimization: merge the updated entity and element models with the original models.

[0102] Model fusion and optimization: the updated entities and elements are integrated with the original model, and the evaluation parameters are calculated to determine whether the integrated model is qualified. If it is not qualified, optimization operations are performed;

[0103] The updated entities and elements are integrated with the original terrain-level real-scene 3D model to ensure that the integrated model is qualified. Specifically, the evaluation parameters of the integrated model are comprehensively analyzed, including:

[0104] Average distance error: Calculate the average distance error between the updated entities and features and the corresponding parts of the original model. For each point in the model , s is the number of points in the model; find the nearest point in the original model , calculate the distance ( represents the Euclidean distance), then the average distance error ;

[0105] Shape similarity index: The shape similarity is calculated using a shape descriptor-based method and recorded as ;

[0106] Texture mean square error: For the fused texture image, sample points are selected at the boundary and inside of the update area and the original area, and the updated texture pixel value is set to , the original texture pixel value is ,in For pixel coordinates, calculate the texture mean square error , where is the total number of sample points;

[0107] Texture correlation coefficient: Calculate the correlation coefficient between the texture of the updated area and the original area. Suppose the texture pixel value sequence of the updated area is , the original area texture pixel value sequence is , correlation coefficient ( and are the means of the two sequences respectively). Texture correlation coefficient The value range is ; Semantic category consistency ratio: Count the number of entities and elements with consistent semantic categories in the fused model and record it as ;

[0108] Then the average distance error is obtained , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio After normalization, the following formula is entered: To obtain the overall judgment value , where The average distance error , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio The preset weight coefficient; and the obtained comprehensive judgment value As a criterion for measuring whether the fused model is qualified; the comprehensive judgment value Compare with the preset comprehensive judgment threshold. If it is greater than the preset comprehensive judgment threshold, the fused model is judged to be qualified, otherwise, it is judged to be unqualified and the fused model is optimized; including model simplification, compression, and topological relationship repair operations to improve the storage efficiency, rendering efficiency and visualization effect of the model, while ensuring the accuracy and integrity of the model.

[0109] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0110] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes, characterized in that: The following steps are involved: S1. Data collection and preprocessing: Prioritize the target areas, obtain the collection frequency of the target areas according to the priority of the target areas, collect data on the topography and physical information of the target areas according to the collection frequency of the target areas, and preprocess the collected data to obtain the model data of the target areas. The preprocessing includes data format conversion, coordinate system unification, noise removal, data clipping and splicing; Data collection includes the following steps: S1.

1. Prioritize the target areas. The priority is determined by classifying the roles of the entities and elements in the target areas in the entire regional functional system; S1.2, obtain the corresponding theoretical acquisition frequency through the priority of the target area; S1.3, divide the target area into small areas; S1.

4. Correct the acquisition frequency of all small areas in the target area to obtain the actual acquisition frequency of all small areas. Collect data from each small area in the target area according to the actual acquisition frequency. The acquisition frequency of the small area is corrected by obtaining the activity of the entities and elements in the small area. The higher the activity of the small area, the higher the acquisition frequency. S2. Change detection: Compare the newly collected model data with the historical model data to identify the changed areas and types. The change types include addition, deletion or modification. S3, entity and element extraction and update: add, delete, and modify the entities and elements in the changed area to obtain the updated model; When adding or modifying entities and elements, an incremental modeling mechanism is used. The process is as follows: When new collected data arrives, only the newly added or modified parts are locally modeled and updated; by establishing a data index and a change area identification mechanism, the entities and elements that need to be updated are located, and then the existing model data is used as a reference and constraint to build a new local model and seamlessly integrate it with the original model; S4. Model fusion and optimization: Fuse the updated entity and element models with the original model and optimize the fused model.

2. A method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: In S1, data is collected through multi-source mapping, which includes satellite remote sensing images, aerial photogrammetry, ground laser scanning, and mobile surveying vehicles.

3. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: The target area priority division in S1.1 is achieved by dividing the roles of various entities and elements in the target area in the entire regional functional system into levels, including the first level, the second level, and the third level. The first level includes: important transportation facilities and key infrastructure; the second level includes: general buildings, ordinary roads; the third level includes: natural vegetation and small landscape facilities.

4. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: The acquisition of the theoretical acquisition frequency of the target area in S1.2 comprises the following steps: S1.2.

1. The average rating for the region is calculated by summing the ratings of all entities and elements in the region and dividing by the number of entities and elements included in the calculation; S1.2.

2. Analyze historical data and industry development trends to estimate the frequency of changes of different types of entities and elements in the entire region; calculate the mean frequency of changes of all types of entities and elements and record it as the trend value; S1.2.

3. After normalizing the obtained regional average grade score and trend value, the regional average grade score and trend value are used as the radius of the two circles respectively, and the preset weight factor is set as high. A frustum model is established by combining the two established circles, and the volume of the frustum is calculated and recorded as the optimal judgment value. This optimal judgment value is used as the standard for measuring the priority of the entities and elements in the entire region; S1.2.

4. Compare the obtained optimal judgment value with the preset optimal judgment value interval. Different optimal judgment value intervals correspond to different collection frequencies. When the optimal judgment value interval to which the optimal judgment value belongs is determined, the theoretical collection frequency corresponding to the target area is determined.

5. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: The small area division in S1.3 includes division based on geographical features, division based on functional zoning, and division combined with population density distribution; Division based on geographical features: Analyze the topography and landforms in the region, including natural geographical elements such as mountains, rivers, and lakes; divide the two sides of the mountains into different small areas; for river areas, divide the river core influence area and transition area into small areas based on the width, flow rate, and whether it is a seasonal river; Division based on functional zoning: According to the functional zoning in urban planning, including commercial areas, residential areas, industrial areas, cultural and educational areas; the entities and elements of different functional areas have different changes in emphasis and frequency, and small areas are divided based on this; Division based on population density distribution: including densely populated areas and sparsely populated areas.

6. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: S1.4 When the change activity of entities and elements in a small area is more than 50% higher than the average level of the whole area, the correction coefficient of its collection frequency is set to be greater than 1 to increase the collection frequency. When the change activity of a small area is less than 30% lower than the average level of the whole area, the correction coefficient is set to be less than 1 to reduce the collection frequency. The activity of a small area is determined by comparing the historical data of the small area with the average change data of the overall area, counting the physical change data of the number of newly built buildings and the length of road reconstruction in the small area in the past few years, and comparing them with the corresponding data of the overall area; when the growth rate of the number of newly built buildings in a small area exceeds the average level of the overall area, it is judged that the small area is in the development stage, and its physical and element changes are more active.

7. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: The change detection in S2 adopts the change detection algorithm based on feature matching, image difference and semantic segmentation. The weights are assigned to the results of the three algorithms, and a voting mechanism is used for comprehensive judgment. If the majority of algorithms determine that the area is a changed area, the change is determined, and then the change type is determined based on the details displayed by each algorithm; Preliminary positioning based on feature matching: Assume that the set of feature points extracted from the newly collected data is , the set of feature points extracted from historical data is ; Feature points The total number of feature points is n. The total number of feature points is m. and , calculate the eigenvector distance between them , using Euclidean distance, ; in and They are feature points and No. dimensional feature vector components, is the feature vector dimension; Setting distance threshold , if exists Make , then it is believed that and Matching, the number of matching feature points is recorded as ; Define the matching ratio ,when When , the area is determined as a change area, where is the matching ratio threshold; Image difference detail analysis: Assume that the newly acquired data is , the historical data is , after registration, for the image data, calculate the pixel difference ,in is the pixel coordinate; for point cloud data, calculate the point With corresponding point The distance difference ; Set the difference threshold ,like or , then the pixel or point is marked as a potential change point, and the total number of change points is recorded as ; Define the change ratio ,in is the total number of pixels or points in the data, combined with the feature matching results, when and When further analyzing whether there are interference factors in the area; Semantic segmentation category determination: Assume that the classification result of the deep learning semantic segmentation model for the newly collected data is , the classification result of historical data is ,in and Indicates the semantic category of the corresponding position, Index variable used to traverse the classification result set of newly collected data And historical data classification result set Elements in ; define the category change ratio ,in ,when When is the threshold of semantic category change ratio, and a comprehensive judgment is made based on the results of the first two algorithms: s is the classification result set of the newly collected data And historical data classification result set The total number of elements in: Comprehensive judgment and fusion results: Assign weights to the results of the three algorithms, and use a voting mechanism to make a comprehensive judgment. If the majority of algorithms determine that the area is a change, then the change is confirmed. Then, the type of change is determined based on the details displayed by each algorithm. The change type includes addition, deletion, or modification.

8. A method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1 or 7, characterized in that: Extraction and update of entities and elements in the S3; For newly added entities and elements, 3D modeling technologies, including point cloud modeling, texture mapping, and semantic modeling, are used to extract the geometric shapes, location information, and texture features of the newly added entities and elements from the newly collected data, and the corresponding 3D entity models are constructed and added to the terrain-level real-scene 3D model; For deleted entities and elements, the corresponding model data is directly deleted in the terrain-level real-scene 3D model; For the modified entities and elements, the geometric shapes, positions, and texture attributes of the corresponding entities and elements in the terrain-level real-scene 3D model are updated and modified according to the change information in the newly collected data.

9. The method for dynamically updating three-dimensional entities and elements based on terrain-level real scenes according to claim 1, characterized in that: The S4 fuses the updated entities and elements with the original terrain-level real-scene 3D model, and determines whether the fused model is qualified by comprehensively analyzing the evaluation parameters of the fused model. The evaluation parameters include: Average distance error: Calculate the average distance error between the updated entities and features and the corresponding parts of the original model. For each point in the model , s is the number of points in the model; find the nearest point in the original model , calculate the distance , where represents the Euclidean distance, then the average distance error ; Shape similarity index: The shape similarity is calculated using a shape descriptor-based method and recorded as ; Texture mean square error: For the fused texture image, sample points are selected at the boundary and inside of the update area and the original area, and the updated texture pixel value is set to , the original texture pixel value is ,in For pixel coordinates, calculate the texture mean square error , where is the total number of sample points; Texture correlation coefficient: Calculate the correlation coefficient between the texture of the updated area and the original area. Suppose the texture pixel value sequence of the updated area is , the original area texture pixel value sequence is , correlation coefficient , where and are the means of the two sequences, texture correlation coefficient The value range is ; Semantic category consistency ratio: Count the number of entities and elements with consistent semantic categories in the fused model and record it as ; Then the average distance error is obtained , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio After normalization, the following formula is entered: To obtain the overall judgment value , where The average distance error , shape similarity index , texture mean square error , texture correlation coefficient and semantic category consistency ratio The preset weight coefficient; and the obtained comprehensive judgment value As a criterion for measuring whether the fused model is qualified; The obtained comprehensive judgment value Compare with the preset comprehensive judgment threshold. If it is greater than the preset comprehensive judgment threshold, the fused model is judged to be qualified, otherwise, it is judged to be unqualified and the fused model is optimized, including model simplification, compression, and topological relationship repair.

Images