A cross-scene spatial intelligent semantic topology generation method and system

Abstract

The application relates to the field of space intelligence, and provides a cross-scene space intelligent semantic topology generation method and system. In the process of constructing a cross-scene space semantic topology, a transition region modeling mechanism with a transition node as a core is introduced, regions where mutations in geometric structures, semantic attributes and topological scales simultaneously occur between different space scenes are identified, expanded and processed as independent objects, and the problem of semantic rupture, topological distortion and space incoherence caused by simple splicing at the scene boundary level in the prior art is avoided. The transition node is structurally expanded based on a topological connection relationship, a transition region reflecting the connectivity of different space scenes is constructed, and semantic transition consistency constraints and topological structure deformation compensation are implemented in the transition region. In addition, the application combines a multi-scale topological self-adaptive adjustment mechanism, so that the cross-scene semantic topology can dynamically adjust node granularity and topological levels under different space scales.

Description

Technical Field

[0001] This invention belongs to the field of spatial intelligence, specifically relating to a cross-scenario spatial intelligent semantic topology generation method and system. Background Technology

[0002] With the development of spatial perception and intelligent computing technologies, the acquisition and representation of spatial information has been widely applied in fields such as intelligent transportation, indoor navigation, smart cities, augmented reality, and intelligent building management. Existing spatial intelligence systems typically model the geometric structure, spatial relationships, and semantic attributes of the environment to understand and utilize the spatial environment, thereby providing support for path planning, spatial positioning, and intelligent decision-making.

[0003] Currently, spatial semantic modeling and topological representation technologies mostly revolve around specific types of spatial environments, such as constructing rooms for indoor environments or topological networks of roads and regional units for outdoor environments. These technologies can effectively describe spatial structures in their respective independent application scenarios, but they are usually based on the assumption of a single spatial domain and lack the ability to uniformly describe the continuous relationships between different types of spaces.

[0004] In practical applications, indoor and outdoor spaces are often connected through areas such as building entrances, lobbies, corridors, staircases, or underground passages. These areas exhibit distinct transitional characteristics in terms of spatial form, semantic attributes, and structural relationships, differing from both typical open outdoor spaces and enclosed indoor spaces. Existing technologies often fail to model these transitional areas effectively due to significant differences in spatial structure or inconsistent semantic definitions, resulting in disjointed spatial representations and impacting subsequent navigation and spatial analysis.

[0005] Furthermore, different spatial environments exhibit significant differences in scale, node density, and spatial hierarchy. For example, urban road networks typically have a large spatial scale and low semantic granularity, while building interiors have more detailed spatial divisions and higher semantic complexity. Existing spatial topology modeling methods often employ fixed structures or static rules when dealing with spatial transitions at different scales, making it difficult to balance the representation of the overall structure with the description of local details, and easily leading to problems such as redundant spatial representations or missing key information.

[0006] On the other hand, due to differences in lighting conditions, sensing methods, and spatial feature distribution between indoor and outdoor environments, spatial models built based on multi-source data exhibit inconsistent performance under different environments. Although existing technologies can process multi-source spatial data to some extent, they still have shortcomings in terms of spatial relationships and semantic consistency, making it difficult to guarantee the stability and reliability of cross-environment spatial representations.

[0007] Therefore, existing spatial semantic modeling and topological representation technologies still have limitations in dealing with the continuous representation of different spatial environments, such as indoor and outdoor spaces. In particular, they are still unable to meet the stability and accuracy requirements of complex spatial intelligent applications in terms of structural coherence, semantic consistency, and unified description of spaces at different scales in spatial transition areas. Summary of the Invention

[0008] This invention provides a cross-scene spatial intelligent semantic topology generation method and system to solve the problem in the prior art that the spatial structure expression is inconsistent and the modeling accuracy is reduced due to semantic discontinuity and topological relationship distortion during scene switching or transition of indoor and outdoor spaces.

[0009] To address the aforementioned technical problems, the present invention discloses the following technical solutions:

[0010] One aspect of the present invention provides a cross-scenario spatial intelligent semantic topology generation method, comprising:

[0011] Acquire spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies for each;

[0012] Identify transition nodes connecting different spatial scenes in spatial semantic topology;

[0013] Based on the transition nodes, the spatial semantic topology of different spatial scenarios is fused to generate an initial semantic topology across scenarios.

[0014] Perform multi-scale topology adaptive adjustment on the initial semantic topology across scenarios;

[0015] The transition region in the initial semantic topology across scenes after adjustment is determined based on the transition node;

[0016] Semantic transition consistency constraints are applied to nodes in the transition region, and topological deformation in the transition region is compensated to obtain the final cross-scene spatial semantic topology.

[0017] Optionally, the step of acquiring spatial semantic data from different spatial scenarios and constructing corresponding spatial semantic topologies includes:

[0018] For each spatial scenario, perform the following steps:

[0019] Collect multi-source spatial data in the aforementioned spatial scene, and extract the spatial geometric information and semantic attribute information of each spatial object;

[0020] Based on the aforementioned spatial geometric information, the spatial positional relationships between different spatial objects are determined;

[0021] Each spatial object is treated as a topological node and associated with semantic attribute information. The corresponding spatial location relationships are used as topological connection relationships to construct the spatial semantic topology of the spatial scene.

[0022] Optionally, the spatial data includes at least point cloud data, image data, and depth data;

[0023] The extraction of spatial geometric information and semantic attribute information for each spatial object includes:

[0024] Perform semantic segmentation on image data, identify spatial objects, and extract semantic labels;

[0025] The point cloud data, image data, and depth data are fused with the semantic label information of each spatial object;

[0026] Spatial geometric information of each spatial object is obtained based on point cloud data and depth data.

[0027] Optionally, identifying transition nodes connecting different spatial scenes in the spatial semantic topology includes:

[0028] For any spatial semantic topology of a spatial scene, the following steps are performed:

[0029] Based on the boundary shape, surface normal angle, and depth difference of spatial blocks in the spatial semantic topology, boundary nodes whose spatial changes exceed a preset change threshold are selected.

[0030] Spatial aggregation algorithms are used to aggregate nearby boundary nodes into boundary clusters;

[0031] When the boundary cluster overlaps spatially with any boundary cluster in the adjacent spatial semantic topology and the semantic category matches, the boundary node in the boundary cluster is determined as a candidate node.

[0032] Determine whether there exists a candidate node whose spatial location overlaps with any candidate node in the adjacent spatial semantic topology and whose semantic category matches.

[0033] If so, the candidate node and the matching candidate node in the adjacent spatial semantic topology are determined as a pair of transition nodes.

[0034] Optionally, the step of fusing the spatial semantic topology of different spatial scenarios based on the transition node to generate a cross-scenario initial semantic topology includes:

[0035] Using transition nodes as anchor points for cross-scene topology fusion, a corresponding relationship is established between different spatial semantic topologies;

[0036] Weighted fusion processing is performed on corresponding nodes from different spatial semantic topologies. The weighted fusion processing introduces semantic similarity, geometric position constraints and spatial reachability constraints between nodes.

[0037] The topological connections of the fused spatial semantic topology are integrated, and the consistency verification and merging of conflicting or redundant topological connections are performed to obtain the initial semantic topology across scenarios.

[0038] Optionally, the multi-scale topology adaptive adjustment of the initial semantic topology across scenes includes:

[0039] Based on the spatial location relationships and semantic attribute information of nodes in the cross-scene initial semantic topology, the cross-scene initial semantic topology is divided into multiple spatial regions;

[0040] For each spatial region, the corresponding initial topological level is determined based on the number of nodes, spatial range, and semantic diversity;

[0041] When the node density in a spatial region exceeds a preset density threshold, or when the spatial range exceeds a preset range threshold, the spatial region is merged and the topology level is raised by one level.

[0042] When a spatial region meets the preset condition of containing complex channels or multiple semantic nodes, the spatial region is split into nodes and the topology level is reduced by one level.

[0043] Optionally, determining the transition region in the adjusted cross-scene initial semantic topology based on transition nodes includes:

[0044] By taking the transition nodes in the initial semantic topology of cross-scene as core nodes, and expanding the structure of adjacent nodes according to the topological connection relationship, a transition region reflecting the connectivity between different spatial scenes is constructed.

[0045] Optionally, the semantic transition consistency constraint on nodes in the transition region includes:

[0046] Compare the semantic feature similarity and spatial position relationship of adjacent nodes in the transition region to evaluate the semantic continuity between adjacent nodes;

[0047] Intermediate semantic interpolation is performed on adjacent nodes whose semantic continuity is lower than a preset continuity threshold.

[0048] Optionally, the compensation for topological deformation in the transition region includes:

[0049] Continuously monitor changes in the relative distance and connection direction between nodes in the transition region to determine whether geometric deformation has occurred in the topology.

[0050] If so, compensation is performed at the geometric deformation points, and the compensation includes at least adjusting the node position coordinates, resetting the connection weights, or regenerating the local connection relationships.

[0051] Another aspect of the present invention provides a cross-scenario spatial intelligent semantic topology generation system, comprising:

[0052] The spatial semantic topology module is configured to acquire spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies respectively.

[0053] The transition node identification module is configured to identify transition nodes connecting different spatial scenes in the spatial semantic topology;

[0054] The fusion module is configured to fuse the spatial semantic topology of different spatial scenes based on the transition node to generate an initial semantic topology across scenes.

[0055] The scaling module is configured to perform multi-scale topology adaptive scaling on the initial semantic topology across scenes;

[0056] The transition region determination module is configured to determine the transition region in the adjusted cross-scene initial semantic topology based on the transition node;

[0057] The transition region adjustment module is configured to impose semantic transition consistency constraints on nodes in the transition region and compensate for topological deformation in the transition region between different spatial scenes to obtain the final cross-scene spatial semantic topology.

[0058] This invention discloses a cross-scene spatial intelligent semantic topology generation method and system. By introducing a transition region modeling mechanism centered on transition nodes during the cross-scene spatial semantic topology construction process, regions where abrupt changes occur simultaneously in geometric structure, semantic attributes, and topological scale between different spatial scenes are identified, expanded, and processed as independent objects. This avoids the semantic breaks, topological distortions, and spatial incoherence problems caused by simply splicing together scenes at the scene boundary level in existing technologies. By structurally expanding transition nodes based on topological connectivity, a transition region reflecting the connectivity relationships between different spatial scenes is constructed. Semantic transition consistency constraints and topological deformation compensation are implemented within this transition region, ensuring that the cross-scene semantic topology remains continuous, smooth, and structurally stable at key connectivity points.

[0059] Furthermore, this invention incorporates a multi-scale topology adaptive adjustment mechanism, enabling cross-scene semantic topology to dynamically adjust node granularity and topological hierarchy at different spatial scales, effectively reducing redundant computation while ensuring spatial representation accuracy. The final cross-scene spatial semantic topology generated by this invention significantly outperforms existing technologies in terms of geometric continuity, semantic coherence, and topological stability, making it suitable for various application scenarios such as spatial understanding, navigation planning, and intelligent spatial modeling.

[0060] The summary section is provided to present the chosen concepts in a simplified form, which will be further described in the detailed description below. The summary section is not intended to identify essential or indispensable features of this disclosure, nor is it intended to limit the scope of this disclosure. Attached Figure Description

[0061] The above and other objects, features and advantages of this disclosure will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0062] Figure 1 A flowchart illustrating a cross-scenario spatial intelligent semantic topology generation method provided in an embodiment of the present invention;

[0063] Figure 2 An implementation provided by an embodiment of the present invention Figure 1 A flowchart illustrating step S100;

[0064] Figure 3 An implementation provided by an embodiment of the present invention Figure 2 A flowchart illustrating step S101;

[0065] Figure 4 An implementation provided by an embodiment of the present invention Figure 1 A flowchart illustrating step S200;

[0066] Figure 5 An implementation provided by an embodiment of the present invention Figure 1 A flowchart illustrating step S300;

[0067] Figure 6 An implementation provided by an embodiment of the present invention Figure 1 A flowchart illustrating step S400;

[0068] Figure 7 This is a flowchart illustrating the steps for implementing semantic transition consistency constraints on nodes in a transition region, as provided in an embodiment of the present invention.

[0069] Figure 8This is a flowchart illustrating another step in compensating for topological deformation in a transition region, provided by an embodiment of the present invention.

[0070] Figure 9 This is a schematic diagram of a cross-scenario spatial intelligent semantic topology generation system provided in an embodiment of the present invention. Detailed Implementation

[0071] Embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thorough and complete, and will fully convey the scope of the present disclosure to those skilled in the art.

[0072] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0073] Figure 1 This is a flowchart illustrating a cross-scenario spatial intelligent semantic topology generation method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0074] Step S100: Obtain spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies respectively.

[0075] A spatial scene is used to describe a spatial region that differs significantly in its spatial form, functional attributes, and structural organization, such as an outdoor scene and an indoor scene. Since different spatial scenes differ significantly in spatial scale, structural complexity, and object composition, this embodiment of the invention uses the spatial scene as the basic unit to process and model data from different spatial scenes.

[0076] In the embodiments disclosed in this invention, the spatial scene includes at least outdoor scenes and indoor scenes, and may also include other types of scenes, such as semi-open or semi-indoor spaces, and special environmental spaces (such as factory workshops, subway stations, airport terminals, etc.). For ease of understanding, the following embodiments will use indoor scenes and outdoor scenes as examples for illustration.

[0077] In one embodiment of the present invention, such as Figure 2As shown, the following steps are performed for each spatial scenario:

[0078] Step S101: Collect multi-source spatial data in the spatial scene and extract the spatial geometric information and semantic attribute information of each spatial object.

[0079] For different spatial scenarios, multi-source spatial data that can comprehensively reflect the spatial structure and distribution characteristics of spatial objects is collected. Multi-source spatial data includes at least satellite remote sensing imagery, urban road network data, building BIM model data, indoor point cloud map data, and visual imagery data, which are used to describe the spatial environment at multiple scales from macro to micro and from outdoor to indoor.

[0080] Specifically, for outdoor spaces, the collected data is used to characterize at least the building's external structure, road morphology, street boundaries, and their spatial distribution. For indoor spaces, the collected data is used to characterize at least the geometric form and organizational structure of spatial units such as rooms, corridors, staircases, and lobbies. By acquiring this multi-source spatial data, different spatial objects possess a sufficient information foundation at both the geometric and semantic levels.

[0081] After data acquisition, the multi-source spatial data is processed to extract spatial geometric and semantic attribute information of different spatial objects. Spatial geometric information describes the three-dimensional location, scale, morphological contour, and spatial distribution characteristics of spatial objects; semantic attribute information describes the category and functional attributes of spatial objects.

[0082] For outdoor scenes, at least point cloud data, image data, and depth data of the building's external structure, roads, and street boundaries should be collected;

[0083] For indoor scenes, at least point cloud data, image data, and depth data of rooms, corridors, staircases, and lobbies should be collected.

[0084] In one embodiment of the present invention, such as Figure 3 As shown, the spatial geometric information and semantic attribute information of each spatial object are extracted, including the following sub-steps:

[0085] Step S1011: Perform semantic segmentation on the image data, identify spatial objects, and extract semantic labels.

[0086] First, semantic recognition processing is performed on the acquired image data. Through image segmentation and semantic parsing methods, different spatial regions in the image data are identified and divided into spatial object regions with clear semantic meanings. Simultaneously, a corresponding semantic label is assigned to each spatial object region to identify its semantic category.

[0087] Semantic tags include at least roads, building facades, street boundaries, entrances, rooms, corridors, doorways, and stairs, used to characterize the semantic attributes of spatial objects.

[0088] Step S1012: Fuse point cloud data, image data, and depth data with the semantic label information of each spatial object to obtain spatial data with semantic attribute information.

[0089] After semantic label extraction, the semantic label information is fused with the corresponding spatial geometric data. Specifically, through spatial correspondence, the semantic recognition results are matched with spatial geometric information in point cloud data, image data, depth data, or BIM models, so that each spatial object in the spatial data not only contains geometric description information, but also is associated with corresponding semantic attribute information.

[0090] This fusion process creates a spatial dataset with semantic attribute information, thereby achieving a unified expression of spatial geometric features and semantic category features in the same data structure.

[0091] Step S1013: Obtain the spatial geometric information of each spatial object based on point cloud data and depth data.

[0092] Based on the fused point cloud data and depth data, the spatial geometric information of each spatial object is extracted and calculated. This spatial geometric information includes at least the three-dimensional position, outline, scale parameters, and relative spatial distribution characteristics of the spatial object. By comprehensively utilizing the high-precision spatial coordinate information of the point cloud data and the depth constraints of the depth image, the geometric shape of the spatial object can be more accurately depicted.

[0093] Step S102: Determine the spatial positional relationship between different spatial objects based on spatial geometric information.

[0094] Spatial positional relationships describe the relative relationships between spatial objects, such as adjacency, containment, connectivity, or hierarchical relationships. By analyzing the geometric position, boundary range, and relative distance of spatial objects, it is possible to determine whether there are direct or indirect spatial connections between different spatial objects, thus providing a basis for establishing topological connections.

[0095] Step S103: Treat each spatial object as a topological node and associate it with semantic attribute information, and treat the corresponding spatial location relationship as a topological connection relationship to construct the spatial semantic topology of the spatial scene.

[0096] Each spatial object identified in step S101 is mapped to a topological node, and its corresponding semantic attribute information is associated with that topological node. Simultaneously, the spatial positional relationships between spatial objects determined in step S102 are mapped to topological connection relationships between topological nodes. Through this method, a spatial semantic topology that simultaneously reflects spatial structural relationships and semantic information is constructed.

[0097] The following is a general overview of step S100:

[0098] In the embodiments disclosed in this invention, to improve the alignment efficiency and matching accuracy of spatial semantic data between different spatial scenes, a spatial index structure based on geographic coordinates and semantic labels is constructed. This spatial index structure is used to hierarchically organize and quickly retrieve various spatial objects in spatial semantic data, enabling efficient location of spatial nodes with potential correspondences under cross-scene and cross-scale conditions. By simultaneously introducing spatial location constraints and semantic attribute constraints, the mismatch problem caused by relying solely on geometric proximity is avoided, thus providing a stable and reliable index foundation for subsequent cross-spatial semantic topology fusion and alignment. The final output of this spatial index module is a multi-domain feature vector set containing spatial geometric features, semantic category features, and topological association features, which serves as the input basis for subsequent topology fusion processing.

[0099] In terms of data acquisition, a combination of outdoor and indoor spatial data is used to comprehensively describe the spatial environment. For outdoor spaces, spatial data reflecting the external structure of buildings, road morphology, and the distribution characteristics of street boundaries are acquired to characterize city-level or street-level spatial organization. For indoor spaces, the focus is on acquiring geometric morphological data of spatial units such as rooms, corridors, staircases, and lobbies to characterize the spatial structural features inside buildings. Through the collaborative acquisition of indoor and outdoor spatial data, spatial semantic data can cover multi-level structural features from the macro-urban environment to the micro-building interior space.

[0100] To ensure consistency of data across different sources and spatial scenarios in terms of time, timestamp information is uniformly introduced into all spatial data during the data acquisition process. This ensures that indoor and outdoor spatial data are aligned on the same time reference. This time consistency processing method effectively reduces spatial misalignment caused by differences in data acquisition time, providing a reliable guarantee for subsequent spatial coordinate unification and cross-scenario topology construction.

[0101] In terms of data preprocessing, the first step is to perform quality control on the collected point cloud spatial data. Through noise filtering and density adjustment, isolated points, anomalous reflection points, and invalid data points in the spatial point cloud are removed, thereby improving the overall quality and stability of the spatial geometric data and providing a reliable data foundation for subsequent spatial analysis and feature extraction.

[0102] Subsequently, semantic recognition processing is performed on the image data. Image segmentation algorithms are used to analyze different spatial regions within the images, identifying object regions with clear spatial semantic meanings and assigning them corresponding semantic labels. These semantic labels are used to distinguish different types of spatial objects, such as roads, building facades, entrances, doors, stairs, and corridors, ensuring that the spatial data not only contains geometric structural information but also clear semantic hierarchical information.

[0103] After semantic recognition is completed, the obtained semantic label information is fused with the corresponding depth data, so that each spatial object in the point cloud data is associated with a corresponding semantic attribute. Through this fusion process, a spatial point cloud dataset with semantic labels is formed, realizing a unified expression of spatial geometric information and semantic information.

[0104] Finally, the fused spatial data undergoes spatial coordinate transformation, converting the geographic coordinate system used for the outdoor spatial data and the local coordinate system used for the indoor spatial model to the same reference coordinate system, thereby achieving spatial alignment of data from different spatial scenarios. This data preprocessing workflow ensures the uniformity of indoor and outdoor spatial semantic data in terms of geometric accuracy, semantic consistency, and spatial alignment.

[0105] Step S200: Identify transition nodes connecting different spatial scenes in the spatial semantic topology.

[0106] After constructing the spatial semantic topology for different spatial scenarios, this step is used to automatically identify key nodes, or transition nodes, that enable cross-scenario connections between multiple spatial semantic topologies. Transition nodes represent the continuous relationship between different spatial scenarios in terms of spatial location and semantic function, such as spatial units like entrances, exits, lobbies, and corridors between outdoor and indoor scenarios. By identifying transition nodes, stable anchor points can be provided for subsequent fusion of cross-scenario semantic topologies, avoiding structural mismatches or semantic breaks during the topology fusion process.

[0107] In one embodiment of the present invention, such as Figure 4 As shown, for any spatial semantic topology of a spatial scene, the following steps are performed:

[0108] Step S201: Based on the boundary shape, surface normal angle and depth difference of the spatial blocks in the spatial semantic topology, filter out the boundary nodes whose spatial change exceeds the preset change threshold.

[0109] Based on the geometric description information of adjacent spatial blocks in the spatial semantic topology, the boundary morphology features of nodes located at the boundaries of spatial blocks are extracted, including but not limited to changes in boundary contour curvature and the degree of boundary continuity disruption. Simultaneously, the normal angle distribution of the corresponding surface of the boundary node is calculated, and the difference in normal angles between adjacent spatial blocks at the boundary is compared. Furthermore, combined with depth information or spatial height information, the depth difference or height abrupt change between the spatial blocks on both sides of the boundary node is calculated. When at least one or more of the above features exceed a preset change threshold, it is determined that there is a significant change in spatial structure or abrupt change in spatial attributes at the boundary node, thus filtering it as a boundary node with significant spatial changes.

[0110] Step S202: Use a spatial aggregation algorithm to aggregate nearby boundary nodes into a boundary cluster.

[0111] Based on the geometric distance, topological adjacency, and similarity of variation characteristics of boundary nodes in three-dimensional space, a spatial aggregation algorithm is used to automatically group adjacent boundary nodes with consistent variation characteristics into the same boundary cluster. This aggregation process integrates discrete and scattered boundary nodes into continuous boundary structure units, reducing the interference of noisy nodes on the judgment results.

[0112] Step S203: When the boundary cluster overlaps spatially with any boundary cluster in the adjacent spatial semantic topology and the semantic category matches, the boundary node in the boundary cluster is determined as a candidate node.

[0113] The system determines whether the current boundary cluster spatially overlaps, partially coincides with, or satisfies a preset spatial proximity condition with any boundary cluster in the adjacent spatial semantic topology, and further compares the semantic category information of the corresponding spatial blocks of the two. When two boundary clusters from different spatial semantic topologies have a corresponding relationship in spatial location and their semantic categories are consistent or conform to predefined semantic matching rules (e.g., belonging to the same channel class, entrance class, or passable structure class spatial objects), the system determines that the boundary cluster is in a potential connected region between different spatial scenes, and marks the boundary nodes contained in the boundary cluster as a candidate node set for subsequent confirmation of transition nodes.

[0114] Step S204: Determine whether there exists a candidate node whose spatial location overlaps with any candidate node in the adjacent spatial semantic topology and whose semantic category matches.

[0115] Candidate nodes undergo cross-spatial consistency verification to further confirm their connectivity validity. Specifically, for any candidate node in the current spatial semantic topology, it is determined whether there exists another candidate node in an adjacent spatial semantic topology that satisfies an overlap relationship or positional consistency constraint with it in spatial location, and maintains consistency or high correlation in semantic category. This consistency verification process is used to exclude invalid nodes that only exhibit boundary features in a single spatial scene but do not actually participate in cross-spatial connectivity, thereby ensuring that the identified nodes have cross-scene connectivity significance at both the geometric and semantic levels.

[0116] If such a candidate node exists, proceed to step S205.

[0117] If no such candidate node exists, it is determined that there is no transition node in the spatial semantic topology of the current spatial scene.

[0118] Step S205: Determine the candidate node and the matching candidate node in the adjacent spatial semantic topology as a pair of transition nodes.

[0119] Candidate nodes in the current spatial semantic topology and candidate nodes in adjacent spatial semantic topologies that pass consistency verification are identified as a pair of transition nodes. These transition node pairs represent the actual connectivity positions of different spatial scenes in terms of geometric structure, semantic attributes, and topological relationships, and are assigned unique node identifiers to serve as key anchor points in the cross-scene semantic topology fusion process.

[0120] The following is a general overview of step S200:

[0121] The purpose of transition node detection and pairing is to automatically identify spatial nodes that can characterize scene transition relationships from both outdoor and indoor spatial semantic data, and to construct stable and reliable cross-spatial semantic connection anchors. Since outdoor and indoor scenes differ significantly in spatial scale, structural form, and semantic expression, the lack of clear transition nodes as a connection basis can easily lead to breaks in the topology or semantic inconsistencies at the fusion boundary. Therefore, this step achieves accurate detection and effective pairing of transition nodes through joint analysis of spatial geometric continuity and semantic consistency.

[0122] In terms of target identification, the focus is on areas that are geometrically located at the boundary between indoor and outdoor spaces and semantically serve as spatial transition areas. These areas typically possess obvious spatial connectivity attributes, such as building entrances, lobbies, stairwells, corridor turning areas, or semi-open transition spaces. These areas often simultaneously possess geometric adjacency and semantic complementarity in spatial semantic topology, naturally serving as connecting nodes between outdoor and indoor topologies, and are therefore defined as candidate targets for transition nodes.

[0123] In terms of detection methods, spatial data is divided into several spatial blocks, and the boundary morphology between adjacent spatial blocks is analyzed. By comparing the changes in geometric features at the boundaries, it is determined whether there is obvious spatial segmentation or scene transition. Geometric feature changes include, but are not limited to, differences in the morphology of the boundary surface, abrupt changes in the surface normal angle, and significant changes in depth values. Through this method, a set of boundary points with obvious spatial continuity characteristics can be selected from continuous space.

[0124] Subsequently, spatial clustering algorithms are used to aggregate boundary points that are spatially close and have similar geometric features, forming several boundary clusters. Each boundary cluster corresponds to a potential spatial transition region, which can relatively stably describe the overall change characteristics of the spatial structure. Based on this, boundary clusters from outdoor spatial semantic data and indoor spatial semantic data are further compared. When an outdoor boundary cluster and an indoor boundary cluster are detected to have spatial overlap or close adjacency, and their semantic categories are functionally matched or complementary, the pair of boundary clusters is marked as a candidate transition node.

[0125] During the node pairing confirmation phase, each candidate transition node must undergo dual verification for both location consistency and semantic consistency. Location consistency determines whether outdoor and indoor nodes correspond to the same actual physical location or the same spatial transition area within a unified spatial reference frame. Semantic consistency verifies the rationality of their spatial functions or semantic attributes, such as whether they both point to an entrance, passageway connection point, or vertical transportation node. Only candidate nodes that simultaneously meet both of these constraints are confirmed as valid transition nodes, assigned a unique identifier, and their corresponding node pairing information is uniformly stored in a cross-spatial anchor point set for subsequent cross-scene topology fusion and structural alignment processes.

[0126] In terms of output, this step ultimately forms a set of transition node pairs that explicitly describes the corresponding connections between the semantic topology of outdoor and indoor spaces. Each pair of nodes in the set contains node identifiers, spatial relationships, and semantic matching information from different spatial scenes, thus providing clear structural anchors for the subsequent topology fusion stage.

[0127] Step S300: Based on the transition node, the spatial semantic topology of different spatial scenes is fused to generate an initial semantic topology across scenes.

[0128] This step aims to utilize the identified transition node pairs to perform unified modeling and structural fusion of spatial semantic topologies from different spatial scenes, thereby generating a cross-scene initial semantic topology capable of simultaneously describing multiple spatial scenes and their connections. Since different spatial scenes differ in acquisition methods, spatial scales, and semantic expressions, directly splicing topological structures can easily lead to node misalignment, semantic conflicts, or inconsistent connections. Therefore, this step achieves ordered alignment and smooth fusion between different spatial semantic topologies through a fusion method using transition nodes as structural anchors.

[0129] In one embodiment of the present invention, such as Figure 5 As shown, this step can be achieved in the following ways:

[0130] Step S301: Using transition nodes as anchor points for cross-scene topology fusion, establish corresponding relationships between different spatial semantic topologies.

[0131] For each pair of transition nodes, it is considered as a corresponding node representing the same actual spatial transition position, and this serves as the starting point for topological alignment. In the process of establishing correspondence, the process expands gradually to its neighboring nodes, centered on the transition node. By analyzing the spatial positional relationships, topological connectivity relationships, and semantic attribute information of the neighboring nodes, the corresponding mapping relationship between nodes in different spatial semantic topologies is determined, thereby forming a stable cross-scene node correspondence structure.

[0132] Step S302: Perform weighted fusion processing on corresponding nodes from different spatial semantic topologies.

[0133] Weighted fusion processing is performed on corresponding nodes from different spatial semantic topologies to generate a unified cross-scene node representation. This weighted fusion is not a simple averaging process, but rather a comprehensive calculation incorporating semantic similarity, geometric position constraints, and spatial reachability constraints to weight the attribute information of each node. Semantic similarity measures the consistency of different nodes in spatial function and category; geometric position constraints ensure that the positions of the fused nodes conform to the actual spatial layout; and spatial reachability constraints evaluate the connectivity rationality between nodes within the topological structure. Through this multi-factor weighted fusion processing, a more consistent and stable cross-scene node representation can be generated while preserving the original spatial structure characteristics.

[0134] Step S303: Integrate the topological connection relationships of the fused spatial semantic topology, and perform consistency verification and merging processing on the topological connection relationships that have conflicts or redundancies to obtain the initial semantic topology across scenarios.

[0135] The connections associated with corresponding nodes in different spatial semantic topologies are uniformly incorporated into the fusion process, and consistency checks are performed on any conflicting or redundant connections. When multiple semantically or geometrically identical connections are detected between the same pair of nodes, they are merged; when conflicts are detected in the directionality, weight, or semantic attributes of the connections, they are corrected or reconstructed based on the fused node attributes and topological constraints. Through the above consistency checks and merging processes, a structurally coherent and semantically consistent initial semantic topology across scenarios is finally formed.

[0136] The following is a general overview of step S300:

[0137] The goal of cross-spatial topology fusion is to eliminate the topological fragmentation caused by differences in data sources, spatial scales, and semantic representations between different spatial scenes. Through this step, a cross-scene spatial semantic topology that maintains geometric continuity and semantic consistency can be constructed while preserving the spatial structural characteristics of each scene.

[0138] At the input level, this step uses the outdoor semantic topology map, the indoor semantic topology map, and the transition node pairing information obtained in the previous step as the basis for fusion. The outdoor semantic topology map describes the macroscopic spatial structure of building exteriors, roads, and blocks, while the indoor semantic topology map describes the fine spatial structure of rooms, passageways, and staircases. The transition node pairing information clarifies the actual connection positions of different spatial scenes in the real space. By introducing transition nodes as prior constraints, blindly performing global matching of the two types of topologies is avoided, thereby reducing structural ambiguity during the fusion process.

[0139] In terms of fusion strategy, a cross-spatial topology fusion algorithm is employed to orderly merge indoor and outdoor topologies. First, semantic features, geometric location information, and inter-node connections are extracted from both indoor and outdoor nodes, forming a structured set of node features. This step provides a unified feature representation foundation for subsequent node alignment and fusion. Then, using the identified transition nodes as fusion anchors, a clear node correspondence is established between different spatial semantic topologies. Based on this correspondence, neighboring nodes are progressively expanded and matched. During this process, the semantic category similarity and spatial proximity of nodes are comprehensively considered, and semantic comparison and geometric correction are performed on neighboring nodes to ensure reasonable correspondence between nodes across different scenes during the fusion process.

[0140] When two nodes from different spatial semantic topologies are detected to be highly similar in semantic category and satisfy distance constraints in spatial location, a cross-domain connection edge is automatically established between them. This cross-domain connection edge not only represents the structural connectivity between indoor and outdoor spaces, but also clarifies the functional associations between different spatial objects at the semantic level, thereby achieving deep integration of indoor and outdoor topologies at the semantic level.

[0141] During the structural adjustment phase, to prevent redundant connections or overly dense networks in the merged topology, the connections generated during the fusion process are screened and optimized. When a node has too many connected edges or highly overlapping connections after fusion, the connection priority is recalculated based on the spatial distance between nodes, the degree of semantic difference, and the importance of the connections, retaining only those edges that represent the main spatial connectivity. Through this structural optimization mechanism, the merged topology maintains accessibility while avoiding topological imbalance and unnecessary increases in computational complexity.

[0142] Through the above cross-space topology fusion process, a unified cross-scene semantic topology map is finally generated. This topology achieves continuous connection between indoor and outdoor spaces in terms of spatial geometry and forms a unified spatial expression form in terms of semantic logic. It can fully reflect the overall connectivity structure of outdoor space and retain the fine topological features of indoor space.

[0143] Step S400: Perform multi-scale topology adaptive adjustment on the initial semantic topology across scenes.

[0144] The purpose of this step is to address the topological imbalance and local deformation issues introduced by differences in spatial scale, uneven node distribution, and changes in semantic complexity after cross-scene fusion. By performing multi-scale adaptive adjustment of the initial cross-scene semantic topology and combining semantic transition consistency constraints and topological deformation compensation mechanisms, the final output cross-scene spatial semantic topology can maintain structural rationality between different spatial levels and maintain geometric continuity and semantic coherence in the transition regions between different spatial scenes.

[0145] In one embodiment of the present invention, multi-scale topology adaptive adjustment of the initial semantic topology across scenes is performed in the following manner: Figure 6 As shown, it includes the following steps:

[0146] Step S4011: Based on the spatial location relationship and semantic attribute information of nodes in the cross-scene initial semantic topology, divide the cross-scene initial semantic topology into multiple spatial regions.

[0147] Based on the spatial relationships between nodes in the topology and their associated semantic attributes, the entire topology is regionalized, dividing sets of nodes with high spatial proximity and semantic relevance into the same spatial region. This approach allows large-scale cross-scene topologies to be broken down into multiple relatively independent local regions, providing operational units for subsequent multi-scale adjustments to different regions.

[0148] Step S4012: For each spatial region, determine the corresponding initial topology level based on the number of nodes, spatial range, and semantic diversity.

[0149] By comprehensively analyzing the number of nodes within a spatial region, the size of the spatial area it covers, and the diversity of the nodes' semantic attributes, we can determine whether the spatial region is more suitable for modeling at a macro, meso, or micro level. When a region has a large number of nodes, a large spatial span, and relatively general semantic types, it is classified as a higher-level topological structure; conversely, when a region has fewer nodes, a concentrated spatial range, and finer-grained semantic information, it is classified as a lower-level topological structure.

[0150] Step S4013: When the node density in a spatial region exceeds a preset density threshold, or when the spatial range exceeds a preset range threshold, perform node merging processing on the spatial region and upgrade the topology level by one level.

[0151] During the subsequent adaptive adjustment process, the node distribution in each spatial region is continuously evaluated. When the node density in a spatial region exceeds a preset density threshold, or the spatial range covered by the region exceeds a preset range threshold, it is considered that the structural representation at the current topology level is too refined, which is detrimental to the overall stability of the topology and computational efficiency. In this case, nodes that are spatially close and semantically similar within the spatial region are merged. By reducing the number of nodes and simplifying the connection relationships, a higher level of node representation is generated, and the topology level of the region is raised by one level.

[0152] Step S4014: When a spatial region meets the preset condition of containing complex channels or multiple semantic nodes, perform node splitting on the spatial region and reduce the topology level by one level.

[0153] Corresponding to node merging, when a spatial region is detected to contain complex channel structures or multiple semantic nodes, and the existing topological hierarchy cannot fully express the spatial structure and semantic differences within the region, it is determined that the region needs refined modeling. In this case, node splitting is performed on the spatial region, splitting the original abstract nodes into multiple child nodes to describe different spatial units or semantic objects, and refining the connection relationships between nodes accordingly. At the same time, the topological hierarchy of the region is reduced by one level.

[0154] The following is a general overview of the multi-scale topology adaptive adjustment process:

[0155] Multi-scale adaptive topology construction comprehensively addresses the challenge of unified semantic topology modeling across spatial and scale scenarios. Its core idea lies in maintaining an adaptive balance between representational accuracy and computational efficiency through dynamic hierarchical adjustment. After initial fusion of indoor and outdoor semantic topologies, different spatial regions often exhibit significant differences in geometric scale, semantic granularity, and structural complexity. Using a fixed-level or uniform-granularity topological representation can easily lead to local information redundancy, global structural distortion, or excessive computational burden. Therefore, by introducing a multi-scale construction mechanism, the topology can adaptively evolve according to the characteristics of the spatial scene.

[0156] At the level of construction objectives, this step addresses the issue of "scale inconsistency" after cross-scene fusion. The system no longer relies on manually preset hierarchical division rules, but instead automatically adjusts the abstraction level of the topology based on node distribution characteristics, spatial coverage, and semantic complexity. This ensures that when switching between city-level, building-level, and fine-grained interior spaces, the topological structure maintains sufficient expressiveness while avoiding ineffective detail stacking. This adaptive adjustment provides a stable data foundation for subsequent path planning, semantic understanding, and spatial reasoning.

[0157] In terms of structural layering, the macro, meso, and micro layers correspond to different spatial cognitive granularities. The macro layer emphasizes global connectivity and overall layout, suitable for describing the relationships between urban roads, building complexes, or functional areas; the meso layer focuses on the organizational structure of individual buildings and their internal floors; and the micro layer is used to depict fine spatial units such as rooms, doors, and passageways. During the initialization phase, by comprehensively analyzing the number of nodes, spatial span, and semantic diversity, an appropriate topological level is automatically assigned to each region, and dynamically adjusted according to actual conditions during operation.

[0158] In the adaptive adjustment mechanism, a space complexity index is introduced. This index quantifies the complexity of the regional structure by comprehensively considering the degree of node clustering, the number of semantic types, and spatial overlap. When the complexity of a region is high, the topological resolution is improved by increasing the number of nodes and refining edge connections to accurately characterize the space where complex channels or multiple semantic objects coexist. Conversely, in regions with simple structures and simple semantics, nodes are merged and connections are simplified to reduce computational complexity and improve overall processing efficiency.

[0159] In terms of consistency maintenance, the multi-scale topology does not exist in isolation, but rather forms a clear hierarchical correspondence through mapping tables and indexing mechanisms. Strict membership and mapping logic are maintained between macroscopic, mesoscopic, and microscopic nodes, ensuring semantic and structural consistency in the description of the same spatial entity across different levels. When switching between different levels or performing cross-level reasoning, the corresponding spatial region can be accurately traced and located, avoiding semantic breaks or path inconsistencies.

[0160] Through this multi-scale adaptive topology construction process, the output spatial semantic topology not only possesses a unified expressive capability across scenarios but also allows for flexible switching at different scales. In practical applications, whether for macro-level navigation planning or fine-grained indoor path guidance, the topology structure can automatically adjust its expression granularity according to requirements, achieving continuous, stable, and efficient spatial semantic modeling results.

[0161] To facilitate understanding, the following specific examples will be used to illustrate the above content:

[0162] In a three-story office building and its surrounding courtyard and parking lot, the collected spatial semantic data was first analyzed. Based on the number of nodes, spatial range, and semantic diversity, the topological level of each area was initially determined. The courtyard and parking lot areas have fewer nodes, larger spatial range, and relatively simple semantic objects, so they were initially classified as the macro level. The external structural information of the entire building is more complex, belonging to the meso level. The corridors, offices, lobbies, and stairwells inside the buildings are dense and semantically rich, belonging to the micro level.

[0163] For areas with excessively high node density or excessively large spatial extent, such as plazas or courtyards outside buildings, multiple adjacent nodes are merged to generate a higher-level topology. By merging these areas into a single macroscopic node, redundant nodes are reduced, simplifying the macroscopic topology while preserving spatial connectivity between areas, ensuring the entire topology remains clear and efficient at a global scale.

[0164] For areas with complex local spatial structures, such as corridors and offices within a floor, the nodes of the original macro or meso level are split to generate the next level of micro topology. Each micro node corresponds to a specific room, corridor segment, or stairwell, enabling the system to represent the geometric relationships and semantic information of complex spaces with fine detail, thereby improving the accuracy of path planning and spatial analysis.

[0165] Through the aforementioned merging and splitting mechanisms, adaptive adjustment can be achieved between different spatial levels. The macro-level has fewer nodes and represents the global structure; the meso-level nodes represent floors and major building units; and the micro-level nodes describe rooms, passageways, and complex local spatial structures in detail.

[0166] Step S500: Determine the transition region in the initial semantic topology across scenes after adjustment based on the transition nodes.

[0167] This step is used to explicitly characterize the continuous connection regions between different spatial scenes from the initial semantic topology across scenes after completing multi-scale topology adaptive adjustment, so as to avoid the structural incompleteness problem caused by describing cross-scene relationships only through isolated nodes.

[0168] In one embodiment of the present invention, this step can be implemented in the following manner:

[0169] By taking the transition nodes in the initial semantic topology of cross-scene as core nodes, and expanding the structure of adjacent nodes according to the topological connection relationship, a transition region reflecting the connectivity between different spatial scenes is constructed.

[0170] First, the confirmed transition nodes are used as the core nodes for constructing the transition region. These transition nodes are determined after dual verification of spatial location consistency and semantic consistency in the spatial semantic topology of different spatial scenarios. They are naturally located at the connection positions of different spatial scenarios in the topological structure and can stably represent the actual connectivity relationship across scenarios.

[0171] Based on this, and according to the topological connections established in the initial semantic topology across scenes, the structure of directly connected adjacent nodes is expanded along the topological connection direction, starting from the transition node. The structural expansion process includes: traversing nodes directly or indirectly connected to the transition node layer by layer according to the topological edge connections, and determining whether the adjacent nodes are still within the transition range between different spatial scenes by combining the spatial distance, connection weight, and semantic similarity between nodes. In this way, multiple nodes that are geometrically continuous and smoothly change semantically around the transition node can be incorporated into the same structural unit, rather than using the transition node itself as the sole representation across scenes.

[0172] Furthermore, during the structural expansion process, the expansion range can be constrained according to preset expansion conditions, such as limiting the maximum number of topological hops, spatial distance thresholds, or semantic change amplitude thresholds, to prevent the transition region from unboundedly expanding into purely indoor or purely outdoor areas. When a neighboring node is significantly deviated from the core position of the transition region in spatial location, or when the difference between its semantic category and the semantic type corresponding to the transition node exceeds a preset threshold, the expansion along that topological direction will be terminated. Through this constraint mechanism, it is ensured that the finally constructed transition region is structurally compact, semantically continuous, and truly reflects the connection segments between different spatial scenes.

[0173] Through the aforementioned structural expansion process, centered on transition nodes and incorporating topological connectivity, one or more transition regions are ultimately constructed within the initial semantic topology across different scenes. Each transition region is composed of multiple spatially continuous, topologically connected, and semantically progressively related nodes, serving as a holistic representation of the connections between different spatial scenes and their surrounding structural environment. This transition region acts as the target for subsequent semantic transition consistency constraints and topological deformation compensation, providing a clear structural carrier for the stable fusion and refined modeling of cross-scene spatial semantic topology.

[0174] Step S600: Apply semantic transition consistency constraints to the nodes in the transition region, and compensate for the topological deformation in the transition region to obtain the final cross-scene spatial semantic topology.

[0175] In one embodiment of the present invention, semantic transition consistency constraints are applied to nodes in the transition region in the following manner: Figure 7 As shown, it includes the following steps:

[0176] Step S601: Compare the semantic feature similarity and spatial position relationship of adjacent nodes in the transition region to evaluate the semantic continuity between adjacent nodes.

[0177] The similarity of semantic feature vectors of each pair of adjacent nodes is compared, and the geometric continuity index is calculated by combining the spatial positional relationship of the nodes. If adjacent nodes have similar semantic categories and are spatially closely connected, the continuity is high; conversely, if the semantic categories are significantly different or there is a spatial offset, the continuity is low.

[0178] Step S602: Perform intermediate semantic interpolation on adjacent nodes whose semantic continuity is lower than a preset continuity threshold.

[0179] For adjacent nodes with semantic continuity below a preset threshold, intermediate semantic interpolation is performed. This process generates one or more transition nodes between the two nodes and assigns them semantic attribute values, making the semantic changes between nodes smoother. For example, in the transition area from an outdoor road to a building entrance, if the road node and the indoor lobby node have significant semantic differences, a transition node is inserted between them, giving it the semantic characteristics of half road and half lobby, and adjusting its spatial position to achieve geometric continuity. Through interpolation, the cross-scene topology not only maintains geometric connectivity but also achieves a smooth transition in semantic expression.

[0180] The following is a general overview of the semantic transition consistency constraint process:

[0181] The purpose of semantic transition consistency constraints is to ensure logical coherence between different spatial semantic regions, especially preventing semantic jumps or topological inconsistencies at cross-scene boundaries. In real urban and architectural environments, there are significant semantic differences between urban roads and building interior spaces. For example, if one enters a "lobby" or "staircase" directly from a "road," abrupt semantic changes may occur in the topological structure without processing, leading to anomalies in path planning, navigation, or semantic analysis. To overcome this problem, embodiments of this invention introduce a semantic transition constraint mechanism into the spatial semantic topology, making semantic changes between different scenes smoother and more natural.

[0182] Regarding the constraint mechanism, the semantic feature similarity and spatial position relationship of adjacent nodes are compared. For node pairs with low continuity, intermediate semantic interpolation is performed between the two nodes. This interpolation not only considers the smooth transition of semantic categories but also incorporates geometric adjustments based on spatial position to ensure that newly generated nodes maintain reasonable spatial connectivity with adjacent nodes. For example, in the transition from "outdoor road" to "building lobby," the system automatically inserts "entrance area" or "porch" nodes as an intermediate layer, allowing the semantic label to smoothly transition from road to lobby, conforming to the real spatial structure while ensuring topological logical continuity.

[0183] Dynamic correction is another crucial aspect of this mechanism. During system operation, environmental factors such as changes in lighting, occlusion, or sensor errors can cause shifts in node semantic recognition; for example, an entrance region might be misidentified as another type. To address this issue, time-series data is used to monitor changes in node semantics, and the node semantic categories are dynamically adjusted based on the latest observations. This mechanism maintains the semantic stability of the topology map under environmental changes, preventing breaks or errors during path planning or navigation.

[0184] After semantic transition consistency constraint processing, the transition area nodes of cross-scene topology form a smooth transition in semantic space, ensuring semantic coherence from the outside to the inside, or between different building modules.

[0185] In one embodiment of the present invention, the topological deformation in the transition region is compensated using the following method: Figure 8 As shown, it includes the following steps:

[0186] Step S603: Continuously monitor the changes in the relative distance and connection direction between nodes in the transition region to determine whether the topology has undergone geometric deformation.

[0187] The system acquires the spatial coordinates of nodes in the transition area and their connecting edge information in real time. By comparing the node positions and connection directions across consecutive time frames, it calculates changes in distance between nodes and angular offsets of edges. When significant changes occur in the distance between nodes or the connection direction deviates from the original topological logic, potential geometric deformations are identified. This monitoring relies not only on spatial geometric information but also on node semantic attributes and topological relationships to determine the severity of the deformation. For example, if the distance between outdoor road nodes at the building entrance and lobby nodes abnormally increases or decreases, this area is marked as a deformed node zone, and compensation is prepared.

[0188] When geometric deformation is detected, step S604 is executed.

[0189] Step S604: Compensate for the geometric deformation.

[0190] The compensation operation involves multi-level processing, including: adjusting node coordinates to restore deformed nodes to reasonable spatial positions, ensuring the distances between nodes match the expected topology; resetting connection weights, recalculating the weights of edges affected by deformation to guarantee the rationality of path selection or network analysis; and regenerating local connectivity relationships, re-establishing reachability connections between nodes for edge breaks, overlaps, or redundancies caused by deformation, ensuring the continuity and integrity of the local topology. This compensation process combines geometric constraints, semantic consistency, and topology optimization algorithms to maintain visual and computational stability in the topology of the transition region.

[0191] The following is a general overview of the topological deformation compensation process:

[0192] In practical applications, outdoor environments typically include roads, blocks, and building facades, characterized by large spatial scales and complex textures; while indoor environments consist of rooms, passageways, staircases, and lobbies, with relatively smaller scales and variable lighting conditions. Due to these differences, directly fusing indoor and outdoor topologies can lead to problems such as misaligned node positions, abnormal edge directions, and discontinuous semantic labels. To avoid these issues, this invention introduces an environment domain alignment mechanism, ensuring that nodes from different environments tend to be consistent in the high-dimensional semantic representation space, thereby achieving effective fusion of indoor and outdoor topologies.

[0193] The alignment method achieves cross-domain node correspondence through feature space mapping. First, multi-dimensional features of indoor and outdoor nodes are extracted, including structural feature vectors (describing the spatial geometric relationships and adjacency patterns of nodes) and semantic vectors (describing the category labels, functional attributes, etc. of nodes). Then, these features are mapped to a shared high-dimensional feature space. By minimizing feature distribution differences and optimizing category consistency, nodes of the same semantic type overlap as much as possible in the feature space. For example, outdoor building entrance nodes and indoor lobby nodes form a tight cluster in the feature space, enabling automatic establishment of cross-environment semantic connections during subsequent topology fusion.

[0194] Deformation detection and compensation ensure the geometric stability of the topology during the fusion process. After fusion, the relative distance and connection direction changes between nodes are continuously monitored. When geometric deformation is detected in a region (e.g., a stretched staircase, offset corridor nodes, or distorted doorways), compensation is automatically performed. Compensation includes adjusting the spatial coordinates of nodes to restore reasonable spacing, resetting node connection weights to optimize path connectivity, and, if necessary, regenerating local connections to eliminate broken links or overlapping edges. This mechanism effectively eliminates topological distortions caused by scale differences, point cloud noise, or environmental changes.

[0195] The environment domain alignment and deformation compensation processes are executed asynchronously in the system background, ensuring that current navigation or task operations are unaffected. After compensation is completed, a global consistency check is automatically performed to ensure that the entire cross-scene topology structure is unbroken, non-overlapping, and semantically coherent, and outputs a final stable spatial semantic topology map. This mechanism not only improves the accuracy of topology fusion but also ensures the reliability of navigation paths and spatial analysis tasks.

[0196] Through the above mechanism, node feature alignment, semantic continuity, and topological stability can be achieved in complex indoor and outdoor environments, ensuring that cross-scene spatial intelligent semantic topology retains both local details and global connectivity.

[0197] Figure 9 This is a schematic diagram of the structure of a cross-scene spatial intelligent semantic topology generation system provided in an embodiment of the present invention, as shown below. Figure 9 As shown, the system includes the following modules:

[0198] Spatial semantic topology module 1 is configured to acquire spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies respectively.

[0199] The transition node identification module 2 is configured to identify transition nodes connecting different spatial scenes in the spatial semantic topology.

[0200] The fusion module 3 is configured to fuse the spatial semantic topology of different spatial scenes based on the transition node to generate an initial semantic topology across scenes.

[0201] The scale adjustment module 4 is configured to perform multi-scale topology adaptive adjustment on the cross-scene initial semantic topology.

[0202] Transition region determination module 5 is configured to determine the transition region in the initial semantic topology across scenes after adjustment based on the transition node.

[0203] The transition region adjustment module 6 is configured to impose semantic transition consistency constraints on nodes in the transition region and compensate for topological deformation in the transition region between different spatial scenes to obtain the final cross-scene spatial semantic topology.

[0204] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A cross-scenario spatial intelligent semantic topology generation method, characterized in that, include: Acquire spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies for each, including: For each spatial scenario, multi-source spatial data under the spatial scenario are collected, including at least point cloud data, image data and depth data; the spatial scenario includes at least outdoor scene and indoor scene, wherein, for outdoor scene, the collected data is used to characterize the external structure of buildings, road form and street boundary; for indoor scene, the collected data is used to characterize the geometry of rooms, corridors, stairs and lobbies. Extract the spatial geometric information and semantic attribute information of each spatial object, including: Perform semantic segmentation on image data, identify spatial objects, and extract semantic labels; The point cloud data, image data, and depth data are fused with the semantic label information of each spatial object; Spatial geometric information of each spatial object is obtained based on point cloud data and depth data; Identify transition nodes connecting different spatial scenes in spatial semantic topology; Based on the transition nodes, the spatial semantic topology of different spatial scenarios is fused to generate an initial semantic topology across scenarios. Multi-scale topology adaptive adjustment of the initial semantic topology across scenes includes: Based on the spatial location relationships and semantic attribute information of nodes in the cross-scene initial semantic topology, the cross-scene initial semantic topology is divided into multiple spatial regions; For each spatial region, the corresponding initial topological level is determined based on the number of nodes, spatial range, and semantic diversity; When the node density in a spatial region exceeds a preset density threshold, or when the spatial range exceeds a preset range threshold, the spatial region is merged and the topology level is raised by one level. When a spatial region meets the preset condition of containing complex channels or multiple semantic nodes, the spatial region is split into nodes and the topology level is reduced by one level. The transition region in the initial semantic topology across scenes after adjustment is determined based on the transition node; Semantic transition consistency constraints are applied to nodes in the transition region, and topological deformation in the transition region is compensated to obtain the final cross-scene spatial semantic topology, including: Compare the semantic feature similarity and spatial position relationship of adjacent nodes in the transition region to evaluate the semantic continuity between adjacent nodes; Intermediate semantic interpolation is performed on adjacent nodes whose semantic continuity is lower than a preset continuity threshold. Continuously monitor changes in the relative distance and connection direction between nodes in the transition region to determine whether geometric deformation has occurred in the topology. If so, compensation is performed at the geometric deformation points, and the compensation includes at least adjusting the node position coordinates, resetting the connection weights, or regenerating the local connection relationships.

2. The method according to claim 1, characterized in that, The acquisition of spatial semantic data from different spatial scenarios and the construction of corresponding spatial semantic topologies include: For each spatial scenario, perform the following steps: Based on the aforementioned spatial geometric information, the spatial positional relationships between different spatial objects are determined; Each spatial object is treated as a topological node and associated with semantic attribute information. The corresponding spatial location relationships are used as topological connection relationships to construct the spatial semantic topology of the spatial scene.

3. The method according to claim 1, characterized in that, The identification of transition nodes connecting different spatial scenes in spatial semantic topology includes: For any spatial semantic topology of a spatial scene, the following steps are performed: Based on the boundary shape, surface normal angle, and depth difference of spatial blocks in the spatial semantic topology, boundary nodes whose spatial changes exceed a preset change threshold are selected. Spatial aggregation algorithms are used to aggregate nearby boundary nodes into boundary clusters; When the boundary cluster overlaps spatially with any boundary cluster in the adjacent spatial semantic topology and the semantic category matches, the boundary node in the boundary cluster is determined as a candidate node. Determine whether there exists a candidate node whose spatial location overlaps with any candidate node in the adjacent spatial semantic topology and whose semantic category matches. If so, the candidate node and the matching candidate node in the adjacent spatial semantic topology are determined as a pair of transition nodes.

4. The method according to claim 1, characterized in that, The process of fusing spatial semantic topologies of different spatial scenarios based on the transition nodes to generate an initial cross-scenario semantic topology includes: Using transition nodes as anchor points for cross-scene topology fusion, a corresponding relationship is established between different spatial semantic topologies; Weighted fusion processing is performed on corresponding nodes from different spatial semantic topologies. The weighted fusion processing introduces semantic similarity, geometric position constraints and spatial reachability constraints between nodes. The topological connections of the fused spatial semantic topology are integrated, and the consistency verification and merging of conflicting or redundant topological connections are performed to obtain the initial semantic topology across scenarios.

5. The method according to claim 1, characterized in that, The determination of transition regions in the adjusted cross-scene initial semantic topology based on transition nodes includes: By taking the transition nodes in the initial semantic topology of cross-scene as core nodes, and expanding the structure of adjacent nodes according to the topological connection relationship, a transition region reflecting the connectivity between different spatial scenes is constructed.

6. A cross-scenario spatial intelligent semantic topology generation system, characterized in that, include: The spatial semantic topology module is configured to acquire spatial semantic data from different spatial scenarios and construct corresponding spatial semantic topologies, including: For each spatial scenario, multi-source spatial data under the spatial scenario are collected, including at least point cloud data, image data and depth data; the spatial scenario includes at least outdoor scene and indoor scene, wherein, for outdoor scene, the collected data is used to characterize the external structure of buildings, road form and street boundary; for indoor scene, the collected data is used to characterize the geometry of rooms, corridors, stairs and lobbies. Extract the spatial geometric information and semantic attribute information of each spatial object, including: Perform semantic segmentation on image data, identify spatial objects, and extract semantic labels; The point cloud data, image data, and depth data are fused with the semantic label information of each spatial object; Spatial geometric information of each spatial object is obtained based on point cloud data and depth data; The transition node identification module is configured to identify transition nodes connecting different spatial scenes in the spatial semantic topology; The fusion module is configured to fuse the spatial semantic topology of different spatial scenes based on the transition node to generate an initial semantic topology across scenes. The scaling module is configured to perform multi-scale topology adaptive scaling on the initial semantic topology across scenes, including: Based on the spatial location relationships and semantic attribute information of nodes in the cross-scene initial semantic topology, the cross-scene initial semantic topology is divided into multiple spatial regions; For each spatial region, the corresponding initial topological level is determined based on the number of nodes, spatial range, and semantic diversity; When the node density in a spatial region exceeds a preset density threshold, or when the spatial range exceeds a preset range threshold, the spatial region is merged and the topology level is increased by one level. When a spatial region meets the preset condition of containing complex channels or multiple semantic nodes, the spatial region is split into nodes and the topology level is reduced by one level. The transition region determination module is configured to determine the transition region in the adjusted cross-scene initial semantic topology based on the transition node; The transition region adjustment module is configured to impose semantic transition consistency constraints on nodes in the transition region and compensate for topological deformations in the transition regions between different spatial scenes, resulting in the final cross-scene spatial semantic topology, including: Compare the semantic feature similarity and spatial position relationship of adjacent nodes in the transition region to evaluate the semantic continuity between adjacent nodes; Intermediate semantic interpolation is performed on adjacent nodes whose semantic continuity is lower than a preset continuity threshold. Continuously monitor changes in the relative distance and connection direction between nodes in the transition region to determine whether geometric deformation has occurred in the topology. If so, compensation is performed at the geometric deformation points, and the compensation includes at least adjusting the node position coordinates, resetting the connection weights, or regenerating the local connection relationships.

Images