Intelligent design method for exhibition hall display based on augmented reality recognition and space reconstruction

Abstract

The application discloses an exhibition hall display intelligent design method based on augmented reality recognition and space reconstruction, and comprises the following steps: acquiring augmented reality scanning data of an exhibition hall physical space, generating scene point cloud data and an object semantic feature set; constructing a scene object relationship graph, generating a space topology initial graph structure; based on the space topology initial graph structure, constructing a double-genome NEAT network, generating a space adjacency matrix code; based on the space adjacency matrix code, generating a continuous space occupancy field expression function; based on the scene object relationship graph, generating a space region label set, updating the space adjacency matrix code and the continuous space occupancy field expression function; based on the continuous space occupancy field expression function, generating an exhibition hall space reconstruction model; and based on the exhibition hall space reconstruction model, generating an exhibition hall display design scheme. The application improves space utilization and display effect and enhances the experience of visitors.

Description

Technical Field

[0001] This invention relates to the field of spatial reconstruction and 3D modeling technology, and in particular to an intelligent design method for exhibition hall displays based on augmented reality recognition and spatial reconstruction. Background Technology

[0002] With the increasingly diverse demands of modern exhibition hall design, traditional exhibition design methods still have some limitations. Traditional methods primarily rely on manually drawn floor plans and manual adjustments to exhibit positions. While these methods can accomplish basic display tasks, they lack flexibility and interactivity, making it difficult to adjust in real-time according to visitor needs. Furthermore, traditional designs often neglect space utilization and visual optimization, resulting in irrational exhibition space layouts that negatively impact exhibit display effectiveness and visitor experience.

[0003] While existing augmented reality (AR) technologies have been applied to exhibitions in some fields, they are mostly limited to a combination of virtual displays and static interactions, failing to effectively integrate into the reconstruction and layout optimization of the physical exhibition space. In these technologies, the combination of virtual displays and physical space relies on simple models and partial design schemes, failing to fully consider the coordination of functionality and aesthetics, and lacking an effective fusion of interactivity and experiential elements in the exhibits. At the same time, real-time interaction and dynamic design based on physical space are not effectively realized in existing technologies, failing to meet the complex needs of exhibitions.

[0004] Furthermore, existing spatial reconstruction technologies are mostly applied in the architectural field and have not been optimized for the specific needs of exhibition hall design, lacking comprehensive consideration of exhibit display, spatial function division, and visitor path optimization. These shortcomings make it difficult for traditional design methods to meet the personalized and diversified needs of modern exhibition hall displays in practical applications.

[0005] Therefore, how to provide intelligent design methods for exhibition hall displays based on augmented reality recognition and spatial reconstruction is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction. This invention fully utilizes augmented reality technology, spatial reconstruction technology, and dual-genome NEAT networks, innovatively combining spatial topology and regional species evolution mechanisms to optimize the exhibition hall design process. By calculating based on spatial adjacency matrix encoding and continuous spatial occupancy field expression functions, a three-dimensional spatial grid model is generated. Furthermore, through species division driven by regional tag genes and cross-regional migration and recombination, the layout of exhibits and space utilization are precisely optimized. This invention offers advantages such as efficient spatial layout optimization, improved display effects, and enhanced visitor experience, meeting the diverse and personalized needs of modern exhibition hall design.

[0007] The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to embodiments of the present invention includes the following steps: Acquire augmented reality scanning data of the physical space of the exhibition hall, perform unified spatial coordinate processing and feature extraction processing, and generate scene point cloud data and object semantic feature set; Based on scene point cloud data and object semantic feature set, a scene object relationship graph is constructed, structured encoding processing is performed, and an initial spatial topology graph structure is generated; A dual-genome NEAT network is constructed based on the initial spatial topology graph structure, including a perceptual topology genome and a spatial structure-generated genome, which generates a spatial adjacency matrix encoding. Based on spatial adjacency matrix encoding, spatial function fusion operation is performed to generate continuous spatial occupancy field expression function; Based on the scene object relationship graph, spatial region division is performed, a set of spatial region labels is generated, the region label genes are written into the dual-genome NEAT network, evolutionary iteration and cross-regional structural migration and recombination are performed, and the spatial adjacency matrix encoding and continuous spatial occupancy field expression function are updated. A model for reconstructing exhibition space is generated based on the continuous spatial occupancy field expression function. The overall fitness is calculated based on the exhibition space reconstruction model, and the feedback is fed into the dual-genome NEAT network to perform topology updates and regional species set adjustments until the overall fitness reaches the fitness threshold, generating an exhibition display design scheme.

[0008] Optionally, the generated scene point cloud data and object semantic feature set include: Augmented reality scanning data of the physical space of the exhibition hall were collected in chronological order to form a multi-frame data sequence. Each frame contains color image data, depth image data, and corresponding pose data in the device pose sequence. Based on the device pose sequence, rigid registration, point-by-point cumulative fusion, and weighted average update are performed on multi-frame depth image data to generate scene point cloud data. For scene point cloud data, perform occlusion detection processing and mark missing areas, interpolate to generate complete point cloud data, and merge it into scene point cloud data; Based on color image data, pixel-level category discrimination processing is performed to generate a pixel-by-pixel category probability distribution, which is mapped to the corresponding spatial points in the scene point cloud data, and the category with the highest probability value is selected as the object category identifier. Based on the object category identifier, the scene point cloud data is subjected to 3D bounding box fitting to generate object size parameters and object pose parameters, forming a set of object semantic features; Perform consistency verification processing on scene point cloud data and object semantic feature set.

[0009] Optionally, the generated spatial topology initial graph structure includes: Construct an object node set based on scene point cloud data and object semantic feature set; Construct a set of spatial relationship edges based on a set of object nodes, including edges for establishing adjacency relationships, occlusion relationships, and functional association relationships; Assign an edge type identifier to each edge in the spatial relation edge set, and calculate the edge weight parameters; A scene object relationship graph is constructed based on the object node set and the spatial relationship edge set, which is composed of the object node set and the spatial relationship edge set with bound edge type identifier and edge weight parameter; The scene object relationship graph is subjected to structured encoding processing to generate a topology node index table. A topology edge index table is constructed based on the spatial relationship edge set, and the two are combined and encapsulated to generate the initial spatial topology graph structure.

[0010] Optionally, the generation of the spatial adjacency matrix encoding includes: A dual-genome NEAT network was established based on the initial spatial topology graph structure to construct an initial population. Each individual is composed of a perceptual topology genome and a spatial structure-generated genome, both of which contain node genes and connection genes. A shared connection weight parameter is established between the perceptual topological genome and the spatial structure-generating genome. The connection weight parameters of the corresponding connection genes are managed using the same parameter index table and the connection weight parameters are updated synchronously. Configure spatial mapping tags in the connective genes of the spatial structure-generating genome; The initial spatial topology graph structure is converted into a structure encoding vector and input into the perceptual topology genome to perform topology feature propagation calculation and generate topology feature vectors; The output layer structure of the genome is generated by constructing a spatial structure, and the structure generation calculation is performed based on the topological feature vector to generate the spatial adjacency matrix encoding; After generating the spatial adjacency matrix encoding, consistency verification and weight update processing are performed on the spatial adjacency matrix encoding based on the spatial mapping labels.

[0011] Optionally, the function for generating a continuous spatial occupancy field includes: Based on spatial adjacency matrix encoding, the expression form of node genes in the spatial structure-generated genome is expanded and replaced with implicit spatial function nodes; Summation is performed on each row of the spatial adjacency matrix encoding, and the summation result is used as the fusion weight of the corresponding implicit spatial function node. At any point in three-dimensional space, the continuous space occupation field expression function is determined by a weighted linear combination of the outputs of all implicit space function nodes; Perform point-by-point sampling calculations on the continuous spatial occupancy field expression function within a preset three-dimensional space range to generate spatial distribution data of the continuous spatial occupancy field expression function, and output the continuous spatial occupancy field expression function.

[0012] Optionally, the updated spatial adjacency matrix encoding and continuous spatial occupancy field representation function include: Based on the set of spatial relationship edges in the scene object relationship graph, spatial region division is performed, and a set of spatial region labels is generated. Write the set of spatial region tags into the region tag genes of the perceptual topology genome and the spatial structure generation genome; Based on the regional tag genes, a regional species evolution mechanism is constructed to determine the individual affiliation rules. Individuals are classified into corresponding regional species according to the spatial regional tag set, and all individuals are classified into a regional species set. Within each regional species set, evolutionary iterations are performed to update the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the corresponding individuals; Perform cross-regional structural migration and reorganization between regional species assemblies, and update the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the target individuals.

[0013] Optionally, the generated exhibition hall space reconstruction model includes: Based on the spatial distribution data of the continuous spatial occupancy field expression function, a three-dimensional voxel mesh space is constructed. The preset three-dimensional spatial range is divided into regular voxel units according to a fixed voxel resolution. For the vertex coordinates of each regular voxel unit, the spatial distribution data of the continuous spatial occupancy field expression function is called to obtain the corresponding occupancy value, and a voxel occupancy value field is generated. A spatial continuum model is generated based on the voxel occupancy value field. Based on the spatial continuum model, perform mesh reconstruction processing to construct a spatial three-dimensional mesh model; Perform mesh smoothing on the spatial 3D mesh model; Based on the spatial three-dimensional mesh model, a reconstruction model of the exhibition hall space is generated, including three-dimensional geometric structure data and spatial connectivity data.

[0014] Optionally, the generated exhibition hall display design scheme includes: Based on the exhibition hall space reconstruction model, connectivity component analysis is performed on the three-dimensional mesh model of the space to count the number of connected components in the three-dimensional mesh model of the space, calculate the average shortest path length between any two connected components, and generate a spatial structure connectivity index. Based on the exhibition hall space reconstruction model, the set of spatial area labels is mapped to the grid partition boundary of the spatial three-dimensional mesh model. The set of mesh patches corresponding to each spatial area label is calculated, and the proportion of overlapping area between the set of mesh patches and the corresponding area boundary in the set of spatial area labels is statistically analyzed to generate the regional function matching index. Based on the exhibition hall space reconstruction model, calculate the space utilization rate index; The spatial structure connectivity index, regional function matching index, and spatial utilization index are weighted and summed according to preset weights to generate a comprehensive fitness score. The comprehensive fitness feedback is input into the dual-genome NEAT network to perform topology updates and individual weight distribution adjustments to the regional species set. After each round of updates, the spatial adjacency matrix encoding and continuous spatial occupancy field expression function are regenerated, and spatial reconstruction and index calculation are re-executed. During the iteration process, the iteration terminates when the change in the comprehensive fitness is less than the set threshold for consecutive preset rounds. Based on the current exhibition hall space reconstruction model, an exhibition hall display design scheme is generated, including spatial structure files, area mapping parameter tables, object layout parameter tables, and circulation parameter tables.

[0015] The beneficial effects of this invention are: First, by combining augmented reality and spatial reconstruction technologies, this invention can accurately acquire three-dimensional information of the physical space of an exhibition hall and generate a spatial continuum model and a three-dimensional mesh model based on this information. This technology enables precise optimization of spatial layout, enhances the display effect of exhibits, ensures the effective use of exhibition space, and avoids the problems of space waste and unreasonable layout in traditional designs.

[0016] Secondly, this invention innovatively employs a dual-genome NEAT network for exhibition hall design. Through the coupled evolution of the perceptual topological genome and the spatial structure-generated genome, the layout of exhibits can be dynamically adjusted. Furthermore, through species division and cross-regional migration and recombination driven by region-tag genes, efficient division and optimization of spatial functional areas can be achieved. This method makes exhibition hall design more flexible and intelligent, automatically adapting to different exhibition needs, and improving the scalability and real-time performance of the design.

[0017] Furthermore, this invention ensures the accuracy and stability of the design scheme by utilizing the fusion calculation of spatial adjacency matrix encoding and continuous spatial occupancy field expression function. During the spatial planning of exhibit displays, the functional matching degree and spatial connectivity of each exhibition area can be accurately calculated and adjusted, thereby enhancing the interactive experience of visitors and improving the overall exhibition effect and visitor comfort. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of the intelligent exhibition hall display design method based on augmented reality recognition and spatial reconstruction proposed in this invention; Figure 2 This is a schematic diagram of constructing a dual-genome NEAT network based on an initial spatial topology graph structure in this invention; Figure 3 This is a flowchart illustrating the generation of a spatial three-dimensional mesh model based on a spatial continuum model in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 The intelligent design method for exhibition hall displays based on augmented reality recognition and spatial reconstruction includes the following steps: The system acquires augmented reality scanning data of the physical space of the exhibition hall, performs unified spatial coordinate processing and feature extraction processing on the augmented reality scanning data, and generates scene point cloud data and object semantic feature set. The augmented reality scanning data includes color image data, depth image data, and device pose sequence. The object semantic feature set includes object category identifier, object size parameter, and object pose parameter. Based on scene point cloud data and object semantic feature set, a scene object relationship graph is constructed. The scene object relationship graph includes a set of object nodes and a set of spatial relationship edges. Each edge in the set of spatial relationship edges is bound with an edge type identifier and an edge weight parameter. The scene object relationship graph is subjected to structured encoding processing to generate an initial spatial topology graph structure. The initial spatial topology graph structure includes a topology node index table and a topology edge index table. The topology edge index table is determined by the edge type identifier and the edge weight parameter. A dual-genome NEAT network is constructed based on the initial spatial topology graph structure. The dual-genome NEAT network is established based on the NeuroEvolution of Augmenting Topologies algorithm. The dual-genome NEAT network includes a perceptual topology genome and a spatial structure generation genome. The perceptual topology genome and the spatial structure generation genome perform coupled evolution by sharing connection weight parameters. The perceptual topology genome receives the topology node index table and the topology edge index table. The spatial structure generation genome generates a spatial adjacency matrix encoding and configures spatial mapping labels for the connecting genes. The spatial mapping labels are used to bind the matrix elements of the spatial adjacency matrix encoding and the edge identifiers of the topology edge index table. Based on spatial adjacency matrix encoding, the node expression in the spatial structure-generated genome is extended from numerical neurons to implicit spatial function nodes. The implicit spatial function nodes perform spatial function fusion operations with spatial adjacency matrix encoding as fusion weights to generate continuous spatial occupation field expression functions. Spatial region division is performed based on the scene object relationship graph, generating a set of spatial region labels. The set of spatial region labels is written into the region label genes of the perceptual topology genome and the spatial structure generation genome. Based on the region label genes, a regional species evolution mechanism is constructed. Individuals in the dual-genome NEAT network are divided into regional species sets that correspond one-to-one with the set of spatial region labels. Evolutionary iteration is performed within the regional species sets, and cross-regional structural migration and recombination are performed between regional species sets. The spatial adjacency matrix encoding and continuous spatial occupancy field expression function are updated. Based on the continuous spatial occupancy field expression function, a spatial continuum model is generated. The spatial continuum model is then subjected to mesh reconstruction processing to generate a spatial three-dimensional mesh model. Based on the spatial three-dimensional mesh model, a reconstructed exhibition hall space model is generated. The overall fitness is calculated based on the exhibition hall space reconstruction model. The overall fitness is determined by the spatial structure connectivity index, the regional function matching index, and the space utilization index. The overall fitness is fed back into the dual-genome NEAT network to drive the perceptual topology genome and the spatial structure generation genome to perform topological updates and regional species set adjustments until the overall fitness reaches the fitness threshold, thus generating the exhibition hall display design scheme.

[0021] In this embodiment, the generation of scene point cloud data and object semantic feature set includes: Augmented reality scanning data of the physical space of the exhibition hall is collected in chronological order. The augmented reality scanning data forms a multi-frame data sequence. Each frame of augmented reality scanning data contains color image data, depth image data, and corresponding pose data in the device pose sequence. The pose data in the device pose sequence is expressed using a unified spatial coordinate system. The pose data consists of a three-dimensional translation vector and a three-dimensional rotation matrix. Based on the device pose sequence, rigid registration processing is performed on multi-frame depth image data. The depth image data of each frame is mapped to a unified spatial coordinate system through a three-dimensional translation vector and a three-dimensional rotation matrix. Point-by-point cumulative fusion processing is performed under the unified spatial coordinate system. Weighted average updates are performed on points with repeated spatial positions to generate scene point cloud data. For scene point cloud data, perform occlusion detection processing and mark missing regions. Establish local curvature continuity constraints based on the rate of change of normal vectors of boundary points of missing regions. Under the condition that the rate of change of normal vectors is lower than the preset curvature threshold, interpolate along the normal vector direction of boundary points to generate complete point cloud data. Incorporate the complete point cloud data into scene point cloud data. Based on color image data, pixel-level category discrimination processing is performed to generate a pixel-by-pixel category probability distribution. The pixel-by-pixel category probability distribution is mapped to the corresponding spatial point in the scene point cloud data according to the pixel correspondence between color image data and depth image data. For each spatial point, the category with the highest probability value in the category probability distribution is selected as the object category identifier. Based on object category identifiers, 3D bounding box fitting is performed on scene point cloud data. The minimum volume bounding box algorithm is used to solve the 3D bounding box for the point cloud subset corresponding to the same object category identifier, generating object size parameters and object pose parameters. The object size parameters are determined by the length, width and height values ​​of the 3D bounding box, and the object pose parameters are determined by the principal axis direction vector of the 3D bounding box, forming a set of object semantic features. Consistency verification is performed on the scene point cloud data and the object semantic feature set. The ratio of the 3D bounding box corresponding to the object size parameter and the object pose parameter in the object semantic feature set to the coverage of the corresponding point cloud subset is used as the coverage rate. When the coverage rate is lower than the preset coverage ratio, the 3D bounding box fitting process is re-executed until the coverage rate reaches the preset coverage ratio. Finally, the scene point cloud data and the object semantic feature set are output.

[0022] In this embodiment, generating the initial spatial topology graph structure includes: An object node set is constructed based on scene point cloud data and object semantic feature set. Each object node in the object node set corresponds one-to-one with a set of object category identifiers, object size parameters, and object pose parameters in the object semantic feature set. The object nodes are numbered according to the object category identifiers in the object semantic feature set to form a unique object number. A spatial relationship edge set is constructed based on the object node set. For any two object nodes, the Euclidean distance between their 3D spatial coordinate center points in the scene point cloud data is calculated. When the Euclidean distance is less than a preset distance threshold, an adjacency relationship edge is established. When there is an occlusion path between object nodes and the occlusion path is determined by the continuously changing point cloud density area in the scene point cloud data, an occlusion relationship edge is established. When there is a predefined function matching rule for the object category identifier, a function association relationship edge is established. Each edge in the spatial relationship edge set contains the starting object number and the ending object number. Assign an edge type identifier to each edge in the spatial relationship edge set. The edge type identifier of the adjacency relationship edge is encoded with a first fixed code value, the edge type identifier of the occlusion relationship edge is encoded with a second fixed code value, and the edge type identifier of the functional association relationship edge is encoded with a third fixed code value. Calculate the edge weight parameter based on the Euclidean distance between object nodes, the occlusion path length, and the functional matching strength. Bind the edge type identifier and edge weight parameter to each edge in the corresponding spatial relationship edge set. A scene object relationship graph is constructed based on the set of object nodes and the set of spatial relationship edges. The scene object relationship graph is composed of the set of object nodes and the set of spatial relationship edges with bound edge type identifiers and edge weight parameters. The scene object relationship graph is subjected to structured encoding processing. The set of object nodes is sorted in ascending order by object number to generate a topology node index table. The index value in the topology node index table corresponds one-to-one with the object number. A topology edge index table is constructed based on the set of spatial relationship edges. The topology edge index table is arranged in the order of starting object number and ending object number. Each record includes starting object number, ending object number, edge type identifier, and edge weight parameter. The topology edge index table is stored in a sparse adjacency structure and only records object node pairs with spatial relationship edges. The topology node index table and the topology edge index table are combined and encapsulated to form the initial spatial topology graph structure.

[0023] In this embodiment, generating the spatial adjacency matrix encoding includes: A dual-genome NEAT network is established based on the initial spatial topology graph structure. The initial population of the dual-genome NEAT network is constructed according to the gene coding rules of the NeuroEvolution of Augmenting Topologies algorithm. Each individual is composed of a perceptual topology genome and a spatial structure-generated genome. Both the perceptual topology genome and the spatial structure-generated genome contain node genes and connection genes. The node genes contain node numbers and node type identifiers, and the connection genes contain input node numbers, output node numbers, connection weight parameters, and activation status identifiers. A shared connection weight parameter is established between the perceptual topology genome and the spatial structure generation genome. The connection weight parameters of the corresponding connection genes in the perceptual topology genome and the spatial structure generation genome are managed by the same parameter index table. When performing topological mutation and connection mutation in the subsequent process, the connection weight parameters of the connection genes with the same parameter index in the two genomes are updated synchronously to maintain the consistency between the topological feature propagation path and the spatial adjacency matrix encoding generation path. Spatial mapping labels are configured in the linker genes of the spatial structure generating genome. The spatial mapping labels are integer index values ​​and correspond one-to-one with the edge identifiers in the topological edge index table. When the spatial structure generating genome performs link mutation, node mutation and crossover operations in the future, the spatial mapping labels are retained and updated along with the linker genes. The topological node index table and topological edge index table in the initial spatial topological graph structure are converted into structural encoding vectors and input into the perceptual topological genome. The perceptual topological genome performs topological feature propagation calculation based on the connection weight calculation rules of the NEAT algorithm combined with node genes and connection genes to generate topological feature vectors. The output layer structure of the spatial structure-generating genome is constructed. The spatial structure-generating genome performs structure generation calculations based on topological feature vectors to generate spatial adjacency matrix encoding. Specifically, the structure generation calculations are as follows: the topological feature vectors are split into node feature vector sequences according to the index order of the topological node index table; pairwise pairing calculations are performed on the node feature vector sequences; edge score values ​​are generated for each pair of node feature vectors to form an edge score matrix; the edge score values ​​in the edge score matrix are located to the corresponding matrix element positions of the spatial adjacency matrix encoding according to the start and end object numbers of the topological edge index table; zero values ​​are written to the corresponding matrix element positions of nodes not recorded in the topological edge index table; symmetric consistency processing is performed on the spatial adjacency matrix encoding to update mutually inverse matrix elements to the same value; numerical range compression processing is performed on the spatial adjacency matrix encoding to map the matrix element values ​​to a preset numerical range; and the spatial adjacency matrix encoding is output. To ensure consistency between the spatial adjacency matrix encoding and the topology edge index table, the binding rules between the spatial mapping label and the topology edge index table are as follows: each spatial mapping label uniquely corresponds to an edge identifier in the topology edge index table, and this edge identifier is only effective in the corresponding position in the topology edge index table; when the topology changes, during the generation of the new topology edge index table, the spatial mapping label will be reconfigured and updated according to the edge identifier in the topology edge index table, thereby ensuring that the one-to-one correspondence between the spatial adjacency matrix encoding and the topology edge index table remains consistent. After generating the spatial adjacency matrix encoding, consistency verification and weight update processing are performed on the spatial adjacency matrix encoding based on the spatial mapping labels. Specifically, the topological edge index table edge identifiers corresponding to the spatial mapping labels of the connecting genes in the genome are located. The matrix element positions in the spatial adjacency matrix encoding are determined based on the topological edge index table edge identifiers. The connection weight parameters are written and updated for the matrix element to maintain the structural consistency between the spatial adjacency matrix encoding and the topological edge index table.

[0024] In this embodiment, generating the continuous spatial occupancy field expression function includes: Based on spatial adjacency matrix encoding, the expression form of node genes in the spatial structure generating genome is expanded, and the numerical neurons corresponding to the node genes are replaced with implicit spatial function nodes. Each implicit spatial function node corresponds one-to-one with a node gene number in the spatial structure generating genome, and a continuous function expression is established in the three-dimensional space. The fusion weight of each implicit spatial function node is calculated based on the spatial adjacency matrix encoding. The elements of each row of the spatial adjacency matrix encoding are summed, and the summation result is used as the fusion weight of the corresponding implicit spatial function node. At any point in three-dimensional space, the continuous spatial occupancy field expression function is determined by a weighted linear combination of the outputs of all implicit spatial function nodes, and its calculation expression is: ; ; in, This represents the fusion weight of the implicit space function node numbered i. Let represent the value of the matrix element in the i-th row and j-th column of the spatial adjacency matrix encoding, and N represent the number of node genes in the spatial structure-generated genome. This represents the function value of the continuous spatial occupancy field expression function at the three-dimensional spatial coordinates (x, y, z). Represents coordinate variables in three-dimensional space. This represents the function value of the implicit spatial function node numbered i at the three-dimensional spatial coordinates (x, y, z); in, ; in, This represents the coordinates of the implicit spatial function node numbered i in three-dimensional space. The function value at a given point represents the occupied response value, where x represents the first coordinate component of the three-dimensional spatial coordinates, y represents the second coordinate component, and z represents the third coordinate component. This represents the first coordinate coefficient of the node gene corresponding to the implicit spatial function node. This coefficient is taken from the connection weight parameter corresponding to the input connection of the first coordinate component in the spatial structure-generated genome. This represents the second coordinate coefficient of the node gene corresponding to the implicit spatial function node. This coefficient is taken from the connection weight parameters corresponding to the input connections of the second coordinate component in the spatial structure-generated genome. This represents the third coordinate coefficient of the node gene corresponding to the implicit spatial function node. This coefficient is taken from the connection weight parameter corresponding to the input connection of the third coordinate component in the spatial structure generating genome. This represents the bias coefficient of the node gene corresponding to the implicit spatial function node. This coefficient is taken from the connection weight parameters corresponding to constant input connections in the spatial structure-generated genome. It represents exponential operations with the natural constant as the base; When the value of a matrix element in the spatial adjacency matrix encoding is zero, the matrix element does not participate in the calculation of the fusion weight of the corresponding implicit spatial function node, thus disabling the corresponding fusion path; Perform point-by-point sampling calculations on the continuous spatial occupancy field expression function within a preset three-dimensional space range to generate spatial distribution data of the continuous spatial occupancy field expression function, and output the continuous spatial occupancy field expression function.

[0025] In this embodiment, updating the spatial adjacency matrix encoding and the continuous spatial occupancy field representation function includes: Spatial region partitioning is performed based on the set of spatial relationship edges in the scene object relationship graph. The spatial region partitioning rules are as follows: the spatial relationship edges corresponding to the functional association relationship are used as the region aggregation constraints, the object nodes in the set of spatial relationship edges that meet the connectivity conditions constitute the spatial connectivity components, each spatial connectivity component is used as a spatial region, and a set of spatial region labels is generated. Each spatial region label in the set of spatial region labels corresponds one-to-one with a spatial connectivity component in the scene object relationship graph. The set of spatial region labels is written into the region label genes of the perceptual topology genome and the spatial structure generation genome. The region label genes correspond one-to-one with the node gene numbers. Each node gene is associated with a spatial region label, which is used to identify the region to which the topology node corresponding to the node gene belongs in the set of spatial region labels. The regional species evolution mechanism is constructed based on regional tag genes. The regional species evolution mechanism includes the division of individuals into regional species sets, the execution of evolutionary iteration within regional species sets, and the execution of cross-regional structural migration and recombination between regional species sets. The rule for determining the individual affiliation is as follows: the connection weight statistics are performed on the genome generated by the spatial structure of each individual in the dual-genome NEAT network, the sum of the absolute values ​​of the connection weight parameters corresponding to each spatial regional tag is calculated, the spatial regional tag with the largest sum of the absolute values ​​of the connection weight parameters is determined as the dominant spatial regional tag of the individual, the individual is assigned to the regional species corresponding to the dominant spatial regional tag according to the set of spatial regional tags, and all individuals are assigned to regional species sets that correspond one-to-one with the set of spatial regional tags. Within each regional species set, evolutionary iterations are performed, including node mutations, connection mutations, and crossover operations. During the evolutionary process, only individuals with the same regional tag genes are allowed to participate in crossover operations, updating the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the corresponding individuals. Cross-regional structural migration and recombination is performed between regional species sets. The cross-regional structural migration and recombination includes: selecting individuals with high connection strength from the regional species with the largest sum of absolute values ​​of connection weight parameters as source regional species; selecting the regional species with the smallest sum of absolute values ​​of connection weight parameters from the remaining regional species as target regional species; selecting a subset of connection genes from the individuals of the source regional species, the spatial mapping label of the connection gene subset corresponding to the spatial regional label of the source regional species; copying the connection gene subset to the individuals of the target regional species; and performing topological validity verification on the copied connection gene subset. The topological validity verification includes: verifying that the spatial mapping label corresponding to the connection gene subset has a corresponding topological edge index table edge identifier in the target regional species; after the verification is passed, the connection gene subset is written into the spatial structure of the target individual to generate the genome; and then the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the target individual are updated. The triggering condition for cross-regional structural migration and recombination is: when the fitness value of an individual in the set of species in a region corresponding to a certain spatial region label does not change significantly over multiple generations, the cross-regional migration and recombination mechanism is triggered, allowing the migration of a subset of the connecting genes of high-fitness individuals to a set of species in a region with lower fitness, and performing topological validity verification.

[0026] In this embodiment, generating the exhibition hall space reconstruction model includes: Based on the spatial distribution data of the continuous spatial occupancy field expression function, a three-dimensional voxel mesh space is constructed. The preset three-dimensional spatial range is divided into regular voxel units according to a fixed voxel resolution. For the vertex coordinates of each regular voxel unit, the spatial distribution data of the continuous spatial occupancy field expression function is called to obtain the corresponding occupancy value, and a voxel occupancy value field is generated. A spatial continuum model is generated based on the voxel occupancy value field. Mesh reconstruction is performed based on a spatial continuum model. The Mesh reconstruction uses the Marching Cubes algorithm to extract isosurfaces. The threshold for isosurface extraction is set as a fixed occupancy boundary value of the continuous spatial occupancy field expression function, which is 0.5. Within each voxel cell, voxel vertices in the spatial distribution data of the continuous spatial occupancy field expression function that are greater than or equal to the fixed occupancy boundary value are marked as internal points, and voxel vertices that are less than the fixed occupancy boundary value are marked as external points. The generation method of triangular patches within the voxel cell is determined according to the lookup table rules of the Marching Cubes algorithm, and a spatial three-dimensional mesh model is constructed. Perform mesh model smoothing on the spatial three-dimensional mesh model. Mesh model smoothing includes: for each triangle vertex, calculate the average coordinates of its adjacent vertices, update the vertex coordinates to the weighted result of the original coordinates and the average coordinates of the adjacent vertices, and the weighting coefficient is a preset fixed value. Based on the spatial 3D mesh model, a reconstructed exhibition hall model is generated. The reconstructed exhibition hall model includes 3D geometric structure data and spatial connectivity data. The 3D geometric structure data comes from the vertex coordinates and triangular facet index data of the spatial 3D mesh model, and the spatial connectivity data is generated based on the facet adjacency relationship of the spatial 3D mesh model.

[0027] In this embodiment, generating the exhibition hall display design scheme includes: Based on the exhibition hall space reconstruction model, the spatial structure connectivity index is calculated. The calculation method of the spatial structure connectivity index is as follows: perform connected component analysis on the spatial three-dimensional mesh model, count the number of connected components in the spatial three-dimensional mesh model, calculate the average shortest path length between any two connected components, normalize the number of connected components and the average shortest path length, and then linearly combine them according to preset weights to generate the spatial structure connectivity index. Based on the exhibition hall space reconstruction model, the regional function matching degree index is calculated. The calculation method of the regional function matching degree index is as follows: map the set of spatial region labels to the grid partition boundary of the spatial three-dimensional mesh model, calculate the set of mesh patches corresponding to each spatial region label, count the proportion of overlapping area between the set of mesh patches and the corresponding regional boundary in the set of spatial region labels, and average the proportion of overlapping area of ​​all regions to generate the regional function matching degree index. Based on the exhibition hall space reconstruction model, the space utilization rate index is calculated. The space utilization rate index is calculated by: the ratio of the number of voxel units inside the three-dimensional mesh model to the total number of voxel units within the preset three-dimensional space range. The spatial structure connectivity index, regional function matching index, and spatial utilization index are weighted and summed according to preset weights to generate a comprehensive fitness score. The comprehensive fitness feedback is input into the dual-genome NEAT network to perform topological updates on the perceptual topological genome and the spatial structure-generated genome, and to perform individual weight distribution adjustments on the regional species set. After each round of updates, the spatial adjacency matrix encoding and continuous spatial occupancy field expression function are regenerated, and spatial reconstruction and index calculation are re-executed. During the iteration process, the iteration terminates when the change in the overall fitness over consecutive preset rounds is less than a set threshold. Based on the current exhibition hall space reconstruction model, an exhibition hall display design scheme is generated. The exhibition hall display design scheme includes a spatial structure file, a region mapping parameter table, an object arrangement parameter table, and a circulation parameter table. The spatial structure file consists of three-dimensional geometric structure data and spatial connectivity data in the exhibition hall space reconstruction model. The region mapping parameter table is generated by mapping the set of spatial region labels to the grid partition boundaries of the three-dimensional spatial grid model. The region mapping parameter table records the set of grid patch indices corresponding to each spatial region label. The object arrangement parameter table is generated based on the set of object nodes in the scene object relationship graph. The object arrangement parameter table records the object posture parameters and object size parameters of each object node in the coordinate system of the three-dimensional spatial grid model and records the spatial region label to which the object node belongs. The circulation parameter table is generated based on the spatial connectivity data. The circulation parameter table records the sequence of connected paths constructed by the adjacency relationship of grid patches in the spatial connectivity data and records the corresponding path length values.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to the exhibition hall design scenario of a large-scale exhibition. The exhibition hall showcased high-tech products such as smart home devices, virtual reality equipment, and robots. The total area of ​​the hall was 600 square meters, divided into multiple exhibition areas, each displaying different categories of products. Traditional exhibition hall design methods typically rely on manual design and static layout, lacking efficient space utilization and real-time adjustments to the display effect. This results in wasted space, unsatisfactory display effects, and an inability to dynamically respond to visitor needs. Traditional methods also suffer from a lack of interactivity between exhibits, obstructed visitor flow, and difficulty in timely adjustments to layout changes.

[0029] To address these challenges, this invention combines augmented reality (AR) technology with spatial reconstruction technology, utilizing a dual-genome NEAT network for optimized exhibition space design. In this scenario, firstly, physical spatial data of the exhibition hall is acquired using AR scanning technology. Spatial coordinate unification and feature extraction are then performed to generate scene point cloud data and object semantic feature sets. This data provides detailed spatial layout information for subsequent exhibition hall design.

[0030] Next, based on scene point cloud data and object semantic feature sets, a scene object relationship graph is constructed, and structured encoding processing is performed to generate an initial spatial topology graph structure. Using this graph structure, a dual-genome NEAT network is constructed, comprising a perceptual topology genome and a spatial structure generation genome. Through coupled evolution using shared connection weight parameters, a spatial adjacency matrix encoding is generated. This encoding not only optimizes the exhibit layout but also incorporates the fusion calculation of spatial functions to generate a continuous spatial occupancy field expression function for further optimization of the exhibition space.

[0031] In the spatial region segmentation stage, a set of spatial region labels was generated based on the scene object relationship graph. Then, driven by region label genes, regional species evolution was performed to execute the functional division and adjustment of the exhibition hall's spatial regions. Through the adjustment of the regional species set using a dual-genome NEAT network, the display areas of the exhibits were further optimized, and space utilization was improved. Finally, a spatial reconstruction model of the exhibition hall was generated using a continuous spatial occupancy field expression function, and an exhibition hall display design scheme was generated based on this model.

[0032]

[0033] The data in Table 1 shows the limitations of traditional methods in exhibition space design. Due to the lack of intelligent spatial layout optimization, the utilization rate of exhibition space using traditional methods is only 70%, the exhibit display effect score is 6.5, the visitor flow path optimization rate is 65%, and the time required to adjust the exhibit layout is relatively long, reaching 4 hours. The exhibit interaction experience score is only 5.4, indicating poor interactivity and visitor experience.

[0034] The method of this invention significantly improves exhibition space design, increasing space utilization to 85%, a 15% improvement. This improvement is mainly attributed to the intelligent optimization of exhibit layout through augmented reality and spatial reconstruction technology, maximizing space utilization efficiency. The display effect score has increased to 8.3 points, a 1.8-point improvement compared to traditional methods, indicating that the display effect of exhibits is more outstanding and the attraction is greatly enhanced. The visitor flow path optimization has increased from 65% to 90%, a 25% improvement. This improvement stems from the application of the regional species evolution mechanism in this invention. Through spatial area division and optimization, visitors can visit exhibits more smoothly, avoiding congestion caused by improper layout in traditional methods.

[0035] The time required for adjusting exhibit layout has been significantly reduced, from 4 hours using traditional methods to 1.5 hours, saving 2.5 hours. This improvement is due to the automated adjustment function in this invention. Utilizing augmented reality and spatial reconstruction technology, exhibit placement adjustments have become more efficient and convenient. The exhibit interaction experience score has also increased from the traditional 5.4 points to 8.2 points, an improvement of 2.8 points. Through intelligent layout and dynamic adjustment, the interactivity of exhibits has been significantly improved, greatly enhancing visitors' sense of participation and experience.

[0036] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent design method for exhibition hall displays based on augmented reality recognition and spatial reconstruction, characterized in that, Includes the following steps: Acquire augmented reality scanning data of the physical space of the exhibition hall, perform unified spatial coordinate processing and feature extraction processing, and generate scene point cloud data and object semantic feature set; Based on scene point cloud data and object semantic feature set, a scene object relationship graph is constructed, structured encoding processing is performed, and an initial spatial topology graph structure is generated; A dual-genome NEAT network is constructed based on the initial spatial topology graph structure, including a perceptual topology genome and a spatial structure-generated genome, which generates a spatial adjacency matrix encoding. Based on spatial adjacency matrix encoding, spatial function fusion operation is performed to generate continuous spatial occupancy field expression function; Based on the scene object relationship graph, spatial region division is performed, a set of spatial region labels is generated, the region label genes are written into the dual-genome NEAT network, evolutionary iteration and cross-regional structural migration and recombination are performed, and the spatial adjacency matrix encoding and continuous spatial occupancy field expression function are updated. A model for reconstructing exhibition space is generated based on the continuous spatial occupancy field expression function. The overall fitness is calculated based on the exhibition space reconstruction model, and the feedback is fed into the dual-genome NEAT network to perform topology updates and regional species set adjustments until the overall fitness reaches the fitness threshold, generating an exhibition display design scheme.

2. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The generated scene point cloud data and object semantic feature set include: Augmented reality scanning data of the physical space of the exhibition hall were collected in chronological order to form a multi-frame data sequence. Each frame contains color image data, depth image data, and corresponding pose data in the device pose sequence. Based on the device pose sequence, rigid registration, point-by-point cumulative fusion, and weighted average update are performed on multi-frame depth image data to generate scene point cloud data. For scene point cloud data, perform occlusion detection processing and mark missing areas, interpolate to generate complete point cloud data, and merge it into scene point cloud data; Based on color image data, pixel-level category discrimination processing is performed to generate a pixel-by-pixel category probability distribution, which is mapped to the corresponding spatial points in the scene point cloud data, and the category with the highest probability value is selected as the object category identifier. Based on the object category identifier, the scene point cloud data is subjected to 3D bounding box fitting to generate object size parameters and object pose parameters, forming a set of object semantic features; Perform consistency verification processing on scene point cloud data and object semantic feature set.

3. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The generated spatial topology initial graph structure includes: Construct an object node set based on scene point cloud data and object semantic feature set; Construct a set of spatial relationship edges based on a set of object nodes, including edges for establishing adjacency relationships, occlusion relationships, and functional association relationships; Assign an edge type identifier to each edge in the spatial relation edge set, and calculate the edge weight parameters; A scene object relationship graph is constructed based on the set of object nodes and the set of spatial relationship edges. It is composed of the set of object nodes and the set of spatial relationship edges bound with edge type identifiers and edge weight parameters. The scene object relationship graph is processed by structured encoding to generate a topology node index table. A topology edge index table is constructed based on the spatial relationship edge set, and the two are combined and encapsulated to generate the initial spatial topology graph structure.

4. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The generated spatial adjacency matrix encoding includes: A dual-genome NEAT network was established based on the initial spatial topology graph structure to construct an initial population. Each individual is composed of a perceptual topology genome and a spatial structure-generated genome, both of which contain node genes and connection genes. A shared connection weight parameter is established between the perceptual topological genome and the spatial structure-generating genome. The connection weight parameters of the corresponding connection genes are managed using the same parameter index table and the connection weight parameters are updated synchronously. Configure spatial mapping tags in the connective genes that generate the spatial structure genome; The initial spatial topology graph structure is converted into a structure encoding vector and input into the perceptual topology genome. Topology feature propagation calculation is then performed to generate topology feature vectors. The output layer structure of the genome is generated by constructing a spatial structure, and the structure generation calculation is performed based on the topological feature vector to generate the spatial adjacency matrix encoding. After generating the spatial adjacency matrix encoding, consistency verification and weight update processing are performed on the spatial adjacency matrix encoding based on the spatial mapping labels.

5. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The function for generating continuous spatial occupancy field includes: Based on spatial adjacency matrix encoding, the expression form of node genes in the spatial structure-generated genome is expanded and replaced with implicit spatial function nodes; The elements of each row of the spatial adjacency matrix are summed, and the summation result is used as the fusion weight of the corresponding implicit spatial function node. At any point in three-dimensional space, the continuous space occupation field expression function is determined by a weighted linear combination of the outputs of all implicit space function nodes; Perform point-by-point sampling calculations on the continuous spatial occupancy field expression function within a preset three-dimensional space range to generate spatial distribution data of the continuous spatial occupancy field expression function, and output the continuous spatial occupancy field expression function.

6. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The updated spatial adjacency matrix encoding and continuous spatial occupancy field expression function include: Based on the set of spatial relationship edges in the scene object relationship graph, spatial region division is performed, and a set of spatial region labels is generated. Write the set of spatial region tags into the region tag genes of the perceptual topology genome and the spatial structure generation genome; Based on the regional tag genes, a regional species evolution mechanism is constructed to determine the individual affiliation rules. Individuals are classified into corresponding regional species according to the spatial regional tag set, and all individuals are classified into a regional species set. Within each regional species set, evolutionary iterations are performed to update the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the corresponding individuals; Perform cross-regional structural migration and reorganization between regional species assemblies, and update the spatial adjacency matrix encoding and continuous spatial occupancy field expression function of the target individuals.

7. The intelligent exhibition hall design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The generated exhibition hall space reconstruction model includes: Based on the spatial distribution data of the continuous spatial occupancy field expression function, a three-dimensional voxel mesh space is constructed. The preset three-dimensional spatial range is divided into regular voxel units according to a fixed voxel resolution. For the vertex coordinates of each regular voxel unit, the spatial distribution data of the continuous spatial occupancy field expression function is called to obtain the corresponding occupancy value, and a voxel occupancy value field is generated. A spatial continuum model is generated based on the voxel occupancy value field. Based on the spatial continuum model, perform mesh reconstruction processing to construct a spatial three-dimensional mesh model; Perform mesh smoothing on the spatial 3D mesh model; Based on the spatial three-dimensional mesh model, a reconstruction model of the exhibition hall space is generated, including three-dimensional geometric structure data and spatial connectivity data.

8. The intelligent exhibition hall display design method based on augmented reality recognition and spatial reconstruction according to claim 1, characterized in that, The generated exhibition hall display design scheme includes: Based on the exhibition hall space reconstruction model, connectivity component analysis is performed on the three-dimensional mesh model of the space to count the number of connected components in the three-dimensional mesh model of the space, calculate the average shortest path length between any two connected components, and generate a spatial structure connectivity index. Based on the exhibition hall space reconstruction model, the set of spatial area labels is mapped to the grid partition boundary of the spatial three-dimensional mesh model. The set of mesh patches corresponding to each spatial area label is calculated, and the proportion of overlapping area between the set of mesh patches and the corresponding area boundary in the set of spatial area labels is statistically analyzed to generate the regional function matching index. Based on the exhibition hall space reconstruction model, calculate the space utilization rate index; The spatial structure connectivity index, regional function matching index, and spatial utilization index are weighted and summed according to preset weights to generate a comprehensive fitness score. The comprehensive fitness feedback is input into the dual-genome NEAT network to perform topology updates and individual weight distribution adjustments to the regional species set. After each round of updates, the spatial adjacency matrix encoding and continuous spatial occupancy field expression function are regenerated, and spatial reconstruction and index calculation are re-executed. During the iteration process, the iteration terminates when the change in the comprehensive fitness is less than the set threshold for consecutive preset rounds. Based on the current exhibition hall space reconstruction model, an exhibition hall display design scheme is generated, including spatial structure files, area mapping parameter tables, object layout parameter tables, and circulation parameter tables.

Images