A Spatial Intelligent Semantic Scene Graph Consistency Maintenance Method Based on 3D Verification
By performing 3D verification of entity boundaries and updating hierarchical semantic relationships, the problem of judgment errors caused by abnormal boundaries of semantic scene graphs in existing technologies has been solved, achieving consistency maintenance of relationships in semantic scene graphs and improving data accuracy and reliability.
Patent Information
- Application Number
- CN202611619024.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for maintaining the consistency of spatial intelligent semantic scene graph relationships fail to effectively filter and process boundary anomalies in 3D scene semantic records, leading to errors in semantic relationship determination. They also lack a hierarchical verification mechanism, making it difficult to guarantee the basic reliability and overall relationship consistency of entity spatial data.
Semantic records are categorized into self-consistent and non-self-consistent records through 3D verification of entity boundaries. The bounding box is updated and semantic relationships are verified. Boundary and relationship verification is performed in a hierarchical manner to ensure the accuracy and consistency of semantic relationships.
It improves the accuracy of underlying entity data and overall logical reliability of semantic scene graphs, eliminates semantic deviations caused by boundary anomalies, and achieves consistent relationship maintenance of spatial intelligent semantic scene graphs.
Smart Images

Figure CN123118932A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of spatial intelligence and relates to scene recognition technology, specifically a spatial intelligent semantic scene based on three-dimensional verification. Figure 1 Consistent maintenance methods. Background Technology
[0002] Existing methods for maintaining consistency of spatial intelligent semantic scene graph relationships have the following drawbacks when performing spatial 3D scene verification: 1. Existing maintenance methods do not perform prior 3D verification of entity boundaries for 3D scene semantic records. They often directly perform semantic relationship discrimination based on the original acquired entity bounding boxes. In actual scenarios, due to factors such as acquisition noise and coordinate offset, the boundary parameters of some semantic entities may be abnormal. If they are not filtered and processed in advance, such records with inconsistent boundaries will directly participate in subsequent relationship calculations, which can easily introduce geometric deviations and cause errors in semantic relationship judgment. It is difficult to ensure the basic reliability of entity space data from the source. 2. Existing maintenance methods rarely adopt a layered verification approach. They often perform entity boundary verification and semantic relationship verification together. When there are records with abnormal entity boundaries, they will directly participate in the semantic relationship verification process. Errors caused by the boundaries will interfere with the semantic relationship judgment results. At the same time, there is a lack of a mechanism to distinguish and handle the two types of abnormal records. It is impossible to perform bounding box correction and semantic relationship update separately. It is easy for erroneous records to be not effectively corrected, which is not conducive to achieving consistent maintenance of the overall relationship of the semantic scene graph.
[0003] Therefore, this application proposes a spatial intelligent semantic scene based on three-dimensional verification. Figure 1 Consistent maintenance methods. Summary of the Invention
[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a spatial intelligent semantic scene based on 3D verification. Figure 1 The present invention provides a consistency maintenance method that enables the maintenance of the overall relationships in a semantic scene graph.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: a spatial intelligent semantic scene based on three-dimensional verification. Figure 1 The consistency maintenance method includes the following steps: Step S1: Obtain the semantic scene graph to be maintained, perform 3D entity boundary verification on the 3D scene semantic records contained in the semantic scene graph to be maintained, and screen the 3D scene semantic records into entity boundary self-consistent records or entity boundary non-self-consistent records according to the verification results to obtain preliminary 3D spatial verification data. Step S2: Update the bounding boxes of non-self-consistent entity boundary records based on the preliminary three-dimensional verification data in space, verify the semantic relationship of entity entities for self-consistent entity boundary records, and categorize the self-consistent entity boundary records into semantically consistent records or semantically non-self-consistent records based on the verification results to obtain semantic relationship verification data. Step S3: Update the semantic relationships of records with non-self-consistent semantic relationships based on the semantic relationship verification data.
[0006] Furthermore, in step S1, the specific steps are as follows: Step S11: Obtain the semantic scene graph that needs to maintain the consistency of semantic scene relationships, obtain the semantic scene graph to be maintained, collect the scene semantic records existing in the semantic scene graph to be maintained, and obtain multiple three-dimensional scene semantic records. Step S12: Collect the actual 3D scene corresponding to the semantic scene graph to be maintained to obtain the semantic maintenance 3D scene, and perform spatial 3D modeling of the semantic maintenance 3D scene using a 3D scanning device to obtain the semantic scene 3D model; Step S13: Select any P1 scene semantic record from the acquired 3D scene semantic records, and perform 3D verification of the entity boundary of the P1 scene semantic record based on the 3D model of the semantic scene. According to the verification results, the P1 scene semantic records are screened into entity boundary self-consistent records or entity boundary non-self-consistent records. Step S14: Perform 3D entity boundary verification on each 3D scene semantic record to obtain preliminary 3D spatial verification data.
[0007] Furthermore, in step S13, the specific steps are as follows: The semantic objects contained in the semantic record of scene P1 are collected to obtain semantic entity A and semantic entity B; The bounding box coordinate system corresponding to the semantic scene graph to be maintained is obtained to obtain the scene bounding coordinate system, and the obtained scene bounding coordinate system is mapped onto the semantic scene 3D model to obtain the semantic scene 3D coordinate system. The spatial sampling points of semantic entity A covered by the three-dimensional model of the semantic scene are collected to obtain entity A sampling points. The three-dimensional coordinates of each entity A sampling point in the three-dimensional coordinate system of the semantic scene are obtained to obtain multiple entity A three-dimensional coordinates.
[0008] Furthermore, step S13 also includes: The 3D coordinates of each entity A are resolved into X-axis coordinates, Y-axis coordinates, and Z-axis coordinates. The obtained X-axis coordinates are compared numerically, and the largest X-axis coordinate is marked as the largest X-axis boundary, and the smallest X-axis coordinate is marked as the smallest X-axis boundary. The obtained Y-axis coordinates are compared numerically, and the largest Y-axis coordinate is marked as the largest Y-axis boundary, and the smallest Y-axis coordinate is marked as the smallest Y-axis boundary. The obtained Z-axis coordinates are compared numerically, and the largest Z-axis coordinate is marked as the largest Z-axis boundary, and the smallest Z-axis coordinate is marked as the smallest Z-axis boundary. The actual bounding box of entity A is obtained. The bounding box of semantic entity A in the semantic record of scene P1 is collected to obtain the semantic bounding box of entity A. If the semantic bounding box of entity A is completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically self-consistent entity. If the semantic bounding box of entity A is not completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically non-self-consistent entity.
[0009] Furthermore, step S13 also includes: Perform geometric position analysis of sampling points on semantic entity B, and obtain the bounding box of entity B based on the analysis results; The bounding box of semantic entity B in the semantic record of scene P1 is collected to obtain the semantic bounding box of entity B. If the semantic bounding box of entity B is completely consistent with the actual bounding box of entity B, then semantic entity B is marked as a geometrically self-consistent entity. If the semantic bounding box of entity B is not completely consistent with the actual bounding box of entity B, then semantic entity B is marked as a geometrically non-self-consistent entity. If both semantic entity A and semantic entity B are geometrically self-consistent entities, then the semantic records of scene P1 are screened into entity boundary self-consistent records. If semantic entity A is a geometrically non-self-consistent entity, or semantic entity B is a geometrically non-self-consistent entity, then the semantic records of scene P1 are screened into entity boundary non-self-consistent records.
[0010] Furthermore, in step S2, the specific steps are as follows: Step S21: Obtain preliminary three-dimensional verification data in space, and obtain self-consistent records and non-self-consistent records of entity boundaries based on the preliminary three-dimensional verification data in space. Step S22: If the 3D scene semantic record is a non-self-consistent entity boundary record, then obtain the semantic entity bounding box and the actual entity bounding box to which the 3D scene semantic record belongs. If the semantic entity bounding box and the actual entity bounding box are consistent, then retain the semantic entity bounding box in the 3D scene semantic record. If the semantic entity bounding box and the actual entity bounding box are inconsistent, then use the actual entity bounding box to update the semantic entity bounding box in the 3D scene semantic record until the non-self-consistent entity boundary record is verified as a self-consistent entity boundary record. Step S23: If the semantic record of the 3D scene is a self-consistent record of entity boundaries, then the semantic relationship of the entity boundary self-consistent record is verified. Based on the verification result, the entity boundary self-consistent record is screened into semantic relationship self-consistent record or semantic relationship non-self-consistent record to obtain semantic relationship verification data.
[0011] Furthermore, in step S23, the specific steps are as follows: Step S231: Randomly select one W1 entity boundary self-consistent record from the acquired entity boundary self-consistent records, and obtain the semantic relationship corresponding to the W1 entity boundary self-consistent record to obtain the W1 entity semantic relationship. Step S232: If the semantic relationship of entity W1 is an inclusion semantic relationship, then perform bounding box coordinate analysis on the semantic entities contained in the self-consistent record of entity W1 boundary, and screen the self-consistent record of entity W1 boundary into self-consistent record of semantic relationship or non-self-consistent record of semantic relationship according to the analysis results. Step S233: Filter the self-consistent records of each entity boundary into self-consistent records of semantic relations or non-self-consistent records of semantic relations to obtain semantic relation verification data.
[0012] Furthermore, in step S232, the specific steps are as follows: Semantic entity bounding box acquisition is performed on semantic entity A to obtain the bounding box of semantic entity A. Based on the bounding box of semantic entity A, the maximum boundary Axmax, minimum boundary Axmin, maximum boundary Aymax, minimum boundary Aymin, maximum boundary Azmax, and minimum boundary Azmin of semantic entity A on the X-axis, the corresponding maximum boundary Azmax and minimum boundary Azmin on the Y-axis, and the corresponding maximum boundary Azmax and minimum boundary Azmin on the Z-axis are obtained. Semantic entity bounding box acquisition is performed on semantic entity B to obtain the bounding box of semantic entity B. Based on the bounding box of semantic entity B, the maximum boundary Bxmax, minimum boundary Bxmin, maximum boundary Bymax, minimum boundary Bymin, maximum boundary Bzmax, and minimum boundary Bzmin of semantic entity B are obtained. If Axmax ≥ Bxmax and Axmin ≤ Bxmin, then the X-axis is marked as the axis where the relationship is satisfied; if Axmax < Bxmax or Axmin > Bxmin, then the X-axis is marked as the axis where the relationship is not satisfied. If Aymax ≥ Bymax and Aymin ≤ Bymin, then the Y-axis is marked as the axis where the relationship is satisfied; if Aymax < Bymax or Aymin > Bymin, then the Y-axis is marked as the axis where the relationship is not satisfied. If Azmax ≥ Bzmax and Azmin ≤ Bzmin, then the Z-axis is marked as the axis that satisfies the relationship; if Azmax < Bzmax or Azmin > Bzmin, then the Z-axis is marked as the axis that does not satisfy the relationship. If the X-axis, Y-axis, and Z-axis are all coordinate axes that satisfy the relation, then the W1 entity boundary self-consistent record will be screened into a semantic relation self-consistent record. If there are relationships in the X-axis, Y-axis, and Z-axis that do not satisfy the coordinate axes, then the self-consistent records of the W1 entity boundary will be sieved into records with non-self-consistent semantic relationships.
[0013] Furthermore, in step S3, the specific steps are as follows: Obtain semantic relation verification data, obtain semantic relation non-consistent records based on semantic relation verification data, and arbitrarily select one R1 non-consistent record from the obtained semantic relation non-consistent records. Semantic entity bounding boxes are collected for semantic entities A and B in the non-self-consistent record R1, resulting in the bounding boxes of semantic entities A and B. Obtain the maximum X-axis boundary Axmax, minimum X-axis boundary Axmin, maximum Y-axis boundary Aymax, minimum Y-axis boundary Aymin, maximum Z-axis boundary Azmax, and minimum Z-axis boundary Azmin corresponding to semantic entity A based on the bounding box of semantic entity A.
[0014] Furthermore, in step S3, the specific steps are as follows: Based on the coordinate arrangement and combination of the maximum boundary Axmax of the X-axis, the minimum boundary Axmin of the X-axis, the maximum boundary Aymax of the Y-axis, the minimum boundary Aymin of the Y-axis, the maximum boundary Azmax of the Z-axis, and the minimum boundary Azmin of the Z-axis, we obtain the points enclosed by A1 to A8. Create a geometric volume with points A1 to A8 as vertices to obtain semantic volume A. Based on the bounding box markers B1 to B8 of the semantic entity B, a geometric volume is created with the bounding points B1 to B8 as the vertices of the volume, resulting in the semantic volume B. If semantic space volume A and semantic space volume B do not have any spatial overlap, then update the semantic relation corresponding to the non-self-consistent record of R1 to a disjoint semantic relation. If semantic space A and semantic space B have spatial overlap, then the semantic relation corresponding to the non-self-consistent record of R1 will be updated to the overlapping semantic relation.
[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention performs 3D entity boundary verification, dividing the 3D scene semantic records into self-consistent and non-self-consistent entity boundary records, and outputs preliminary 3D spatial verification data. This allows for early verification of the rationality of the bounding box boundary parameters of each semantic entity, filtering out abnormal semantic records caused by acquisition errors and coordinate offsets. For non-self-consistent boundary records, bounding box updates can correct errors in entity spatial range, eliminating semantic deviations caused by entity geometric boundaries at the source. This ensures a reliable spatial geometric foundation for subsequent semantic relationship verification and improves the accuracy of the underlying entity data in the semantic scene graph. 2. Based on the completion of entity boundary verification, this invention further verifies the semantic relationship of entities on the boundary self-consistent records, distinguishes between semantically self-consistent and non-self-consistent records, obtains semantic relationship verification data, and performs semantic relationship updates on semantically non-self-consistent records. This enables layered processing of boundary verification and relationship verification, avoids boundary errors interfering with semantic relationship determination, achieves relationship consistency maintenance of spatial intelligent semantic scene graph, and improves the overall logical reliability of 3D scene semantic records. Attached Figure Description
[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.
[0017] Figure 1 This is a diagram illustrating the implementation steps of the present invention; Figure 2 This is a schematic diagram of the enclosing point of the present invention; Figure 3 This is a schematic diagram of the spatial volume overlap of the present invention. Detailed Implementation
[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0019] Please see Figure 1 This invention provides a technical solution: a spatial intelligent semantic scene based on three-dimensional verification. Figure 1 The consistency maintenance method includes the following steps: Step S1: Obtain the semantic scene graph to be maintained, perform 3D entity boundary verification on the 3D scene semantic records contained in the semantic scene graph to be maintained, and screen the 3D scene semantic records into entity boundary self-consistent records or entity boundary non-self-consistent records according to the verification results to obtain preliminary 3D spatial verification data. In step S1, the specific steps are as follows: Obtain the semantic scene graph that needs to maintain the consistency of semantic scene relationships, and obtain the semantic scene graph to be maintained. Collect the scene semantic records existing in the semantic scene graph to be maintained to obtain multiple three-dimensional scene semantic records. It should be noted here that: In this application, the scene semantic record is a single structured record unit extracted from the semantic scene graph to be maintained. Each record corresponds to a semantic relationship between a pair of objects. It integrates semantic relationship labels, associated objects, observation time and corresponding three-dimensional geometric data (bounding box / pose). It is the basic data unit for consistency maintenance and geometric verification, which is different from the natural language scene description statement of pure text. The scene semantic records in the semantic scene graph can include the following records: Record 1: Object (Carton 01, Pallet 01), semantic relationship: above, timestamp: t1, carton surrounding box, pallet surrounding box; Record 2: Objects (Carton 01, Pallet 01), semantic relationship: away, timestamp: t2, carton surrounding box, pallet surrounding box.
[0020] The actual 3D scene corresponding to the semantic scene graph to be maintained is collected to obtain the semantic maintenance 3D scene. The 3D scene is then spatially 3D modeled using a 3D scanning device to obtain the semantic scene 3D model. Arbitrarily select a P1 scene semantic record from the acquired 3D scene semantic records, and perform 3D entity boundary verification on the P1 scene semantic record based on the 3D semantic scene model. According to the verification results, the P1 scene semantic records are screened into entity boundary self-consistent records or entity boundary non-self-consistent records.
[0021] Specifically as follows: The semantic objects contained in the semantic record of scene P1 are collected to obtain semantic entity A and semantic entity B; It should be noted here that: In this application, if the semantic record of scene P1 is a semantic object (semantic entity A, semantic entity B), the semantic relationship is: far away, the timestamp is 9:13:43, the bounding box of semantic entity A is (12,23,56,87,27,87), and the bounding box of semantic entity B is (32,43,32,87,23,87), the coordinate correspondence of the bounding boxes is (x min x max ,y min ,y max ,z min ,z max ), The bounding box coordinate system corresponding to the semantic scene graph to be maintained is obtained to obtain the scene bounding coordinate system, and the obtained scene bounding coordinate system is mapped onto the semantic scene 3D model to obtain the semantic scene 3D coordinate system. The spatial sampling points of semantic entity A covered by the three-dimensional model of the semantic scene are collected to obtain entity A sampling points. The three-dimensional coordinates of each entity A sampling point in the three-dimensional coordinate system of the semantic scene are obtained to obtain multiple entity A three-dimensional coordinates.
[0022] The 3D coordinates of each entity A are resolved into X-axis coordinates, Y-axis coordinates, and Z-axis coordinates. The obtained X-axis coordinates are compared numerically, and the largest X-axis coordinate is marked as the largest X-axis boundary, and the smallest X-axis coordinate is marked as the smallest X-axis boundary. The obtained Y-axis coordinates are compared numerically, and the largest Y-axis coordinate is marked as the largest Y-axis boundary, and the smallest Y-axis coordinate is marked as the smallest Y-axis boundary. The obtained Z-axis coordinates are compared numerically, and the largest Z-axis coordinate is marked as the largest Z-axis boundary, and the smallest Z-axis coordinate is marked as the smallest Z-axis boundary. The actual bounding box of entity A is obtained. The bounding box of semantic entity A in the semantic record of scene P1 is collected to obtain the semantic bounding box of entity A. If the semantic bounding box of entity A is completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically self-consistent entity. If the semantic bounding box of entity A is not completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically non-self-consistent entity.
[0023] Repeat the process of obtaining the actual bounding box of entity A, perform geometric position analysis of sampling points on semantic entity B, and obtain the bounding box of entity B based on the analysis results; The bounding box of semantic entity B in the semantic record of scene P1 is collected to obtain the semantic bounding box of entity B. If the semantic bounding box of entity B is completely consistent with the actual bounding box of entity B, then semantic entity B is marked as a geometrically self-consistent entity. If the semantic bounding box of entity B is not completely consistent with the actual bounding box of entity B, then semantic entity B is marked as a geometrically non-self-consistent entity. If both semantic entity A and semantic entity B are geometrically self-consistent entities, then the semantic records of scene P1 are screened into entity boundary self-consistent records. If semantic entity A is a geometrically non-self-consistent entity, or semantic entity B is a geometrically non-self-consistent entity, then the semantic records of scene P1 are screened into entity boundary non-self-consistent records. Repeat the process of performing 3D entity boundary verification on the semantic records of scene P1, and perform 3D entity boundary verification on each 3D scene semantic record to obtain preliminary 3D spatial verification data; Step S2: Update the bounding boxes of non-self-consistent entity boundary records based on the preliminary three-dimensional verification data in space, verify the semantic relationship of entity entities for self-consistent entity boundary records, and categorize the self-consistent entity boundary records into semantically consistent records or semantically non-self-consistent records based on the verification results to obtain semantic relationship verification data. In step S2, the specific steps are as follows: Acquire preliminary three-dimensional verification data in space, and acquire self-consistent records and non-self-consistent records of entity boundaries based on the preliminary three-dimensional verification data in space; If the 3D scene semantic record is a non-self-consistent entity boundary record, then the semantic entity bounding box and the actual entity bounding box to which the 3D scene semantic record belongs are obtained. If the semantic entity bounding box and the actual entity bounding box are consistent, then the semantic entity bounding box in the 3D scene semantic record is retained. If the semantic entity bounding box and the actual entity bounding box are inconsistent, then the semantic entity bounding box in the 3D scene semantic record is updated using the actual entity bounding box, until the non-self-consistent entity boundary record is verified as a self-consistent entity boundary record. If the semantic record of the 3D scene is a self-consistent record of entity boundaries, then the semantic relationship of the entity boundary self-consistent record is verified. Based on the verification results, the entity boundary self-consistent record is screened into semantic relationship self-consistent record or semantic relationship non-self-consistent record, and semantic relationship verification data is obtained.
[0024] Specifically as follows: Arbitrarily select one W1 entity boundary self-consistent record from the obtained entity boundary self-consistent records, and obtain the semantic relationship corresponding to the W1 entity boundary self-consistent record to obtain the W1 entity semantic relationship. It should be noted here that: In this application, the semantic relationships involved are specifically spatial topological relationships, including but not limited to overlap, inclusion, and separation.
[0025] If the semantic relationship of entity W1 is an inclusion semantic relationship, then bounding box coordinate analysis is performed on the semantic entities contained in the self-consistent boundary records of entity W1. Based on the analysis results, the self-consistent boundary records of entity W1 are screened into self-consistent semantic relationship records or non-self-consistent semantic relationship records.
[0026] It should be noted here that: In this application, the aforementioned inclusion relationship specifically means that semantic entity A includes semantic entity B.
[0027] Specifically as follows: Semantic entity bounding box acquisition is performed on semantic entity A to obtain the bounding box of semantic entity A. Based on the bounding box of semantic entity A, the maximum boundary Axmax, minimum boundary Axmin, maximum boundary Aymax, minimum boundary Aymin, maximum boundary Azmax, and minimum boundary Azmin of semantic entity A on the X-axis, the corresponding maximum boundary Azmax and minimum boundary Azmin on the Y-axis, and the corresponding maximum boundary Azmax and minimum boundary Azmin on the Z-axis are obtained. Semantic entity bounding box acquisition is performed on semantic entity B to obtain the bounding box of semantic entity B. Based on the bounding box of semantic entity B, the maximum boundary Bxmax, minimum boundary Bxmin, maximum boundary Bymax, minimum boundary Bymin, maximum boundary Bzmax, and minimum boundary Bzmin of semantic entity B are obtained. If Axmax ≥ Bxmax and Axmin ≤ Bxmin, then the X-axis is marked as the axis where the relationship is satisfied; if Axmax < Bxmax or Axmin > Bxmin, then the X-axis is marked as the axis where the relationship is not satisfied. If Aymax ≥ Bymax and Aymin ≤ Bymin, then the Y-axis is marked as the axis where the relationship is satisfied; if Aymax < Bymax or Aymin > Bymin, then the Y-axis is marked as the axis where the relationship is not satisfied. If Azmax ≥ Bzmax and Azmin ≤ Bzmin, then the Z-axis is marked as the axis that satisfies the relationship; if Azmax < Bzmax or Azmin > Bzmin, then the Z-axis is marked as the axis that does not satisfy the relationship. If the X-axis, Y-axis, and Z-axis are all coordinate axes that satisfy the relation, then the W1 entity boundary self-consistent record will be screened into a semantic relation self-consistent record. If there are relationships in the X-axis, Y-axis, and Z-axis that do not satisfy the coordinate axes, then the self-consistent records of the W1 entity boundary will be sieved into records with non-self-consistent semantic relationships. Repeat the process of type screening for the self-consistent records of entity boundary of W1, and screen each self-consistent record of entity boundary into self-consistent records of semantic relationship or non-self-consistent records of semantic relationship to obtain semantic relationship verification data. Step S3: Update the semantic relationships of records with inconsistent semantic relationships based on the semantic relationship verification data; The specific steps in step S3 are as follows: Obtain semantic relation verification data, obtain semantic relation non-consistent records based on semantic relation verification data, and arbitrarily select one R1 non-consistent record from the obtained semantic relation non-consistent records. Semantic entity bounding boxes are collected for semantic entities A and B in the non-self-consistent record R1, resulting in the bounding boxes of semantic entities A and B. Obtain the maximum X-axis boundary Axmax, minimum X-axis boundary Axmin, maximum Y-axis boundary Aymax, minimum Y-axis boundary Aymin, maximum Z-axis boundary Azmax, and minimum Z-axis boundary Azmin corresponding to semantic entity A based on the bounding box of semantic entity A. Mark the coordinate points by taking the minimum boundary Axmin of the X-axis, the minimum boundary Aymin of the Y-axis, and the minimum boundary Azmin of the Z-axis, and obtain the bounding point (Axmin, Aymin, Azmin) of A1; Mark the coordinate points by taking the maximum boundary Axmax of the X-axis, the minimum boundary Aymin of the Y-axis, and the minimum boundary Azmin of the Z-axis, and obtain the bounding point (Axmax, Aymin, Azmin) of A2; Mark the coordinate points by taking the minimum boundary Axmin of the X-axis, the maximum boundary Aymax of the Y-axis, and the minimum boundary Azmin of the Z-axis, and obtain the bounding point (Axmin, Aymax, Azmin) of A3; Mark the coordinate points by taking the maximum boundary Axmax of the X-axis, the maximum boundary Aymax of the Y-axis, and the minimum boundary Azmin of the Z-axis, and obtain the bounding point (Axmax, Aymax, Azmin) of A4; Mark the coordinate points by taking the minimum boundary Axmin of the X-axis, the minimum boundary Aymin of the Y-axis, and the maximum boundary Azmax of the Z-axis, and obtain the A5 bounding point (Axmin, Aymin, Azmax); Mark the coordinate points by taking the maximum boundary Axmax of the X-axis, the minimum boundary Aymin of the Y-axis, and the maximum boundary Azmax of the Z-axis, and obtain the bounding point (Axmax, Aymin, Azmax) of A6; Mark the coordinate points by taking the minimum boundary Axmin of the X-axis, the maximum boundary Aymax of the Y-axis, and the maximum boundary Azmax of the Z-axis, and obtain the bounding point (Axmin, Aymax, Azmax) of A7; Mark the coordinate points by taking the maximum boundary Axmax of the X-axis, the maximum boundary Aymax of the Y-axis, and the maximum boundary Azmax of the Z-axis, and obtain the bounding point (Axmax, Aymax, Azmax) of A8; Please see Figure 2 Create a geometric space volume with A1 to A8 as its vertices to obtain semantic space volume A. Based on the bounding box markers B1 to B8 of the semantic entity B, a geometric volume is created with the bounding points B1 to B8 as the vertices of the volume, resulting in the semantic volume B. If semantic space volume A and semantic space volume B do not have any spatial overlap, then update the semantic relation corresponding to the non-self-consistent record of R1 to a disjoint semantic relation. If semantic space volume A and semantic space volume B have spatial overlap, then the semantic relation corresponding to the non-self-consistent record of R1 will be updated to the overlapping semantic relation. It should be noted here that: Please see Figure 3 In this application, the existence of spatial overlap between semantic space volume A and semantic space volume B means that the two have a common spatial region in three-dimensional space and their spatial occupancy ranges intersect. Based on the axis-aligned bounding box determination, when the intervals of the two spatial volumes intersect in the X, Y, and Z coordinate axes, it is determined that there is spatial overlap. Semantic space volume A and semantic space volume B do not overlap, meaning they have no common spatial regions in three-dimensional space and their spatial occupancy areas are completely separated. Based on axis-aligned bounding box determination, as long as the intervals of the two spatial volumes are separated along any one of the X, Y, or Z coordinate axes, it can be determined that they do not overlap.
[0028] Compared to the problems described in the background art, this invention, by conducting 3D entity boundary verification, divides the 3D scene semantic records into self-consistent and non-self-consistent entity boundary records, outputting preliminary 3D spatial verification data. This allows for early verification of the rationality of the bounding box boundary parameters of each semantic entity, filtering out abnormal semantic records caused by acquisition errors and coordinate offsets. For non-self-consistent boundary records, bounding box updates can correct errors in entity spatial range, eliminating semantic deviations caused by entity geometric boundaries at the source. This ensures that subsequent semantic relationship verification has a reliable spatial geometric foundation, improving the accuracy of the underlying entity data of the semantic scene graph. Furthermore, based on the completion of entity boundary verification, this invention further verifies the entity semantic relationship of the boundary self-consistent records, distinguishes between semantically self-consistent and non-self-consistent records to obtain semantic relationship verification data, and performs semantic relationship updates on semantically non-self-consistent records. This enables layered processing of boundary verification and relationship verification, avoids boundary errors interfering with semantic relationship determination, achieves relationship consistency maintenance of spatial intelligent semantic scene graph, and improves the overall logical reliability of 3D scene semantic records.
[0029] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only to the technical solutions and their full scope and equivalents.
Claims
1. A method for maintaining consistency of spatial intelligent semantic scene graphs based on 3D verification, characterized in that, include: Step S1: Obtain the semantic scene graph to be maintained. Acquire the 3D scene semantic records existing in the semantic scene graph to be maintained. Collect the actual 3D scene in the semantic scene graph to be maintained and perform spatial 3D modeling to obtain the semantic scene 3D model. Obtain the bounding box coordinate system corresponding to the semantic scene graph to be maintained and map it to the semantic scene 3D model to obtain the semantic scene 3D coordinate system. Collect the semantic objects contained in each 3D scene semantic record to obtain semantic entity A and semantic entity B. Collect the spatial sampling points of semantic entity A covered by the semantic scene 3D model and obtain the entity 3D coordinates of each spatial sampling point in the semantic scene 3D coordinate system. Perform entity boundary 3D verification on the 3D scene semantic records based on the obtained 3D coordinates. According to the verification results, each 3D scene semantic record is screened into entity boundary self-consistent records or entity boundary non-self-consistent records to obtain preliminary spatial 3D verification data. Step S2: Update the bounding boxes of non-self-consistent entity boundary records based on the preliminary 3D verification data in space. Collect semantic entity A bounding boxes and semantic entity B bounding boxes for self-consistent entity boundary records. Based on the coordinate boundary numerical relationship between semantic entity A bounding boxes and semantic entity B bounding boxes, filter the coordinate axes into relation-satisfied coordinate axes or relation-unsatisfied coordinate axes. If all coordinate axes are relation-satisfied coordinate axes, then the W1 entity boundary self-consistent records are filtered into semantic relation self-consistent records. If there are relation-unsatisfied coordinate axes, then the W1 entity boundary self-consistent records are filtered into semantic relation-unconsistent records, thus obtaining semantic relation verification data. Step S3: Perform semantic entity bounding box coordinate boundary analysis on the semantic relationship non-self-consistent records based on the semantic relationship verification data. Obtain the semantic entity bounding points based on the analysis results and create a geometric space volume. Obtain semantic space volume A and semantic space volume B based on the creation results. Update the semantic relationship non-self-consistent records to disjoint semantic relationships or overlapping semantic relationships based on the spatial overlap relationship between semantic space volume A and semantic space volume B.
2. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 1, characterized in that, The specific steps in step S1 are as follows: The 3D coordinates of each entity A are resolved into X-axis coordinates, Y-axis coordinates, and Z-axis coordinates. The maximum X-axis coordinate value is marked as the maximum X-axis boundary, the minimum X-axis coordinate value is marked as the minimum X-axis boundary, the maximum Y-axis coordinate value is marked as the maximum Y-axis boundary, the minimum Y-axis coordinate value is marked as the minimum Y-axis boundary, the maximum Z-axis coordinate value is marked as the maximum Z-axis boundary, and the minimum Z-axis coordinate value is marked as the minimum Z-axis boundary, thus obtaining the actual bounding box of entity A. The bounding box of semantic entity A in the semantic record of the 3D scene is collected to obtain the semantic bounding box of entity A. If the semantic bounding box of entity A is completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically self-consistent entity. If the semantic bounding box of entity A is not completely consistent with the actual bounding box of entity A, then semantic entity A is marked as a geometrically non-self-consistent entity.
3. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 2, characterized in that, The specific steps in step S1 are as follows: Perform geometric position analysis of sampling points on semantic entity B, and obtain the bounding box of entity B based on the analysis results; The bounding box of semantic entity B in the semantic record of the 3D scene is collected to obtain the semantic bounding box of entity B. If the semantic bounding box of entity B is completely consistent with the actual bounding box of entity B, then the semantic entity B is marked as a geometrically self-consistent entity. If the semantic bounding box of entity B is not completely consistent with the actual bounding box of entity B, then the semantic entity B is marked as a geometrically non-self-consistent entity. If both semantic entity A and semantic entity B are geometrically self-consistent entities, then the 3D scene semantic records are sieved into entity boundary self-consistent records. If semantic entity A is a geometrically non-self-consistent entity, or semantic entity B is a geometrically non-self-consistent entity, then the 3D scene semantic records are sieved into entity boundary non-self-consistent records.
4. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 1, characterized in that, In step S2, the specific steps are as follows: Step S21: Obtain preliminary three-dimensional verification data in space, and obtain self-consistent records and non-self-consistent records of entity boundaries based on the preliminary three-dimensional verification data in space. Step S22: If the 3D scene semantic record is a non-self-consistent entity boundary record, then obtain the semantic entity bounding box and the actual entity bounding box to which the 3D scene semantic record belongs. If the semantic entity bounding box and the actual entity bounding box are consistent, then retain the semantic entity bounding box in the 3D scene semantic record. If the semantic entity bounding box and the actual entity bounding box are inconsistent, then use the actual entity bounding box to update the semantic entity bounding box in the 3D scene semantic record until the non-self-consistent entity boundary record is verified as a self-consistent entity boundary record. Step S23: If the semantic record of the 3D scene is a self-consistent record of entity boundaries, then the semantic relationship of the entity boundary self-consistent record is verified. Based on the verification result, the entity boundary self-consistent record is screened into semantic relationship self-consistent record or semantic relationship non-self-consistent record to obtain semantic relationship verification data.
5. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 4, characterized in that, In step S23, the specific steps are as follows: Step S231: Randomly select one W1 entity boundary self-consistent record from the acquired entity boundary self-consistent records, and obtain the semantic relationship corresponding to the W1 entity boundary self-consistent record to obtain the W1 entity semantic relationship. Step S232: If the semantic relationship of entity W1 is an inclusion semantic relationship, then perform bounding box coordinate analysis on the semantic entities contained in the self-consistent record of entity W1 boundary, and screen the self-consistent record of entity W1 boundary into self-consistent record of semantic relationship or non-self-consistent record of semantic relationship according to the analysis results. Step S233: Filter the self-consistent records of each entity boundary into self-consistent records of semantic relations or non-self-consistent records of semantic relations to obtain semantic relation verification data.
6. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 5, characterized in that, In step S232, the specific steps are as follows: Semantic entity bounding box acquisition is performed on semantic entity A to obtain the bounding box of semantic entity A. Based on the bounding box of semantic entity A, the maximum boundary Axmax, minimum boundary Axmin, maximum boundary Aymax, minimum boundary Aymin, maximum boundary Azmax, and minimum boundary Azmin of semantic entity A on the X-axis, the corresponding maximum boundary Azmax and minimum boundary Azmin on the Y-axis, and the corresponding maximum boundary Azmax and minimum boundary Azmin on the Z-axis are obtained. Semantic entity bounding box acquisition is performed on semantic entity B to obtain the bounding box of semantic entity B. Based on the bounding box of semantic entity B, the maximum boundary Bxmax, minimum boundary Bxmin, maximum boundary Bymax, minimum boundary Bymin, maximum boundary Bzmax, and minimum boundary Bzmin of semantic entity B are obtained. If Axmax ≥ Bxmax and Axmin ≤ Bxmin, then the X-axis is marked as the axis where the relationship is satisfied; if Axmax < Bxmax or Axmin > Bxmin, then the X-axis is marked as the axis where the relationship is not satisfied. If Aymax ≥ Bymax and Aymin ≤ Bymin, then the Y-axis is marked as the axis where the relationship is satisfied; if Aymax < Bymax or Aymin > Bymin, then the Y-axis is marked as the axis where the relationship is not satisfied. If Azmax ≥ Bzmax and Azmin ≤ Bzmin, then the Z-axis is marked as the axis that satisfies the relationship; if Azmax < Bzmax or Azmin > Bzmin, then the Z-axis is marked as the axis that does not satisfy the relationship. If the X-axis, Y-axis, and Z-axis are all coordinate axes that satisfy the relation, then the W1 entity boundary self-consistent record will be screened into a semantic relation self-consistent record. If there are relationships in the X-axis, Y-axis, and Z-axis that do not satisfy the coordinate axes, then the self-consistent records of the W1 entity boundary will be sieved into records with non-self-consistent semantic relationships.
7. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 1, characterized in that, The specific steps in step S3 are as follows: Obtain semantic relation verification data, and based on the semantic relation verification data, obtain semantic relation non-consistent records. Randomly select one R1 non-consistent record from the obtained semantic relation non-consistent records. Semantic entity bounding boxes are collected for semantic entities A and B in the non-self-consistent record R1, resulting in the bounding boxes of semantic entities A and B. Obtain the maximum X-axis boundary Axmax, minimum X-axis boundary Axmin, maximum Y-axis boundary Aymax, minimum Y-axis boundary Aymin, maximum Z-axis boundary Azmax, and minimum Z-axis boundary Azmin corresponding to semantic entity A based on the bounding box of semantic entity A.
8. The spatial intelligent semantic scene graph consistency maintenance method based on three-dimensional verification according to claim 7, characterized in that, The specific steps in step S3 are as follows: Based on the coordinate arrangement and combination of the maximum boundary Axmax of the X-axis, the minimum boundary Axmin of the X-axis, the maximum boundary Aymax of the Y-axis, the minimum boundary Aymin of the Y-axis, the maximum boundary Azmax of the Z-axis, and the minimum boundary Azmin of the Z-axis, we obtain the points enclosed by A1 to A8. Create a geometric volume with points A1 to A8 as vertices to obtain semantic volume A. Based on the bounding box markers B1 to B8 of the semantic entity B, a geometric volume is created with the bounding points B1 to B8 as the vertices of the volume, resulting in the semantic volume B. If semantic space volume A and semantic space volume B do not have any spatial overlap, then update the semantic relation corresponding to the non-self-consistent record of R1 to a disjoint semantic relation. If semantic space A and semantic space B have spatial overlap, then the semantic relation corresponding to the non-self-consistent record of R1 will be updated to the overlapping semantic relation.
