Interactive method and system for virtual exhibition hall

Abstract

The present invention relates to the field of virtual exhibition hall interaction technology, and discloses an interaction method and system for a virtual exhibition hall. The method first obtains user interaction data collected by various terminals such as touch, voice, and somatosensory, extracts behavioral features and matches them to obtain interaction pairs, specifically extracting feature descriptors, assigning timing labels, and finding matching nodes to obtain interaction pairs. Then, an interaction model is constructed using user interaction data and interaction pairs, and then scene encapsulation is performed to generate a standard interaction package, including obtaining an interaction model with the interaction pair as an anchor point, and performing layered scene encapsulation through two levels. In addition, the exhibition hall topology data is obtained to generate an interaction path, and various terminals collect data accordingly, and the interaction pairs are also screened. The system includes an interaction controller that can perform the above operations. The present invention realizes the effective processing of multi-terminal interaction data and the construction of interaction models, thereby improving the fluency and accuracy of virtual exhibition hall interactions.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual exhibition hall interaction, and in particular to an interaction method and system for a virtual exhibition hall. Background Art

[0002] With the rapid development of information technology, virtual exhibition halls, as a new form of display, are gradually gaining widespread application in various fields. They can break the limitations of time and space and provide users with a convenient and efficient display experience. However, in the interactive process of virtual exhibition halls, many problems still need to be solved.

[0003] When it comes to processing multi-terminal interaction data, traditional virtual exhibition hall interaction methods are often only able to process interaction data from a single terminal, making it difficult to simultaneously integrate user interaction data collected from multiple terminals, such as touch interaction terminals, voice interaction terminals, and somatosensory interaction terminals. This results in an inability to fully and accurately capture user behavior characteristics, which in turn affects the effectiveness of interaction and user experience. For example, when users interact using both voice and gestures simultaneously, traditional methods are unable to effectively integrate the data from these two interaction methods, resulting in inaccurate and untimely interaction responses.

[0004] When it comes to interaction model construction and scene encapsulation, existing methods lack effective mechanisms for constructing reasonable interaction models based on user interaction data and interaction pairs. Furthermore, it's difficult to effectively encapsulate interaction models into standard interaction packages. This results in incomplete and incoherent interaction scenes in virtual exhibition halls, hindering the user's immersive interactive experience. For example, when constructing interaction models, traditional methods may fail to accurately determine the relationships between interaction pairs, leading to loopholes in the interaction model and compromising the smoothness of interaction.

[0005] When it comes to interactive path planning, traditional methods often fail to fully consider factors such as the target scenario, node status, and terminal load data when generating interactive paths based on exhibition hall topology data. This results in suboptimal interactive paths, potentially leading to path choppy paths, insufficient interactive responses, or high operational complexity. For example, in some complex exhibition hall scenarios, interactive paths generated by traditional methods may experience congestion, impacting the user's interactive experience.

[0006] Existing methods lack effective mechanisms for filtering interaction pairs, making it difficult to remove mismatched interaction pairs, which in turn affects the quality of interaction model construction. This can lead to incorrect interaction pairs in the interaction model, further impacting the accuracy and reliability of virtual exhibition hall interactions.

[0007] Throughout the interactive process of the virtual exhibition hall, poor coordination between various links and a lack of unified management and control lead to inefficient and unstable interactions. For example, data collection, interaction model construction, scenario packaging, and interaction path planning can lead to untimely data transmission or inconsistent processing, impacting the overall interactive effect of the virtual exhibition hall. Summary of the Invention

[0008] The purpose of the present invention is to provide an interactive method and system for a virtual exhibition hall to solve the problems raised in the above background technology.

[0009] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for interacting with a virtual exhibition hall, the method comprising:

[0010] Obtaining user interaction data collected by multiple terminals, extracting behavioral features based on the user interaction data, and matching based on the extracted behavioral features to obtain interaction pairs; the multiple terminals include touch interaction terminals, voice interaction terminals, and somatosensory interaction terminals; matching based on the extracted behavioral features to obtain interaction pairs includes: extracting feature descriptors at the behavioral features, assigning time series labels to the behavioral features, and finding matching nodes based on the feature descriptors to obtain matching interaction pairs; the feature descriptors are attribute sequences that describe the behavioral features;

[0011] Build an interaction model based on user interaction data and interaction pairs, and encapsulate the interaction model into scenarios based on the user interaction data to generate a standard interaction package, including:

[0012] The interaction model is obtained through continuous matching using the interaction pairs obtained by mutual matching as anchor points. A scenario segmentation method is used to expand and filter the sparse scenarios obtained by matching the behavioral characteristics of multiple terminals to obtain a complete or partial interaction path model covering the target area.

[0013] Layered scene encapsulation is adopted, and scene encapsulation is performed through two levels. The first level encapsulates the original interaction space into the intermediate scene that can be expressed by the protocol; the second level encapsulates the data of the intermediate scene into the target space that requires interaction to generate a standard interaction package.

[0014] Preferably, the method further comprises the following steps:

[0015] Obtain the exhibition hall topology data required for interaction and generate the interaction path based on the exhibition hall topology data; multiple terminals collect user interaction data based on the interaction path;

[0016] The exhibition hall topology data includes the target scenario, node status, and terminal load data of the required interaction area. The interaction path is generated based on the exhibition hall topology data, including the following:

[0017] Identify boundaries based on the target scenario to build an interaction area, and plan interaction paths based on the interaction area;

[0018] And establish an interaction benchmark based on node status and terminal load data, and collect user interaction data based on the interaction benchmark.

[0019] Preferably, planning the interaction path according to the interaction area includes:

[0020] Construct a two-dimensional exhibition hall topology map in the target scene to be interacted with, and divide the target scene to be interacted with into a number of grid scenes with preset intervals as the side length according to the two-dimensional exhibition hall topology map;

[0021] Classify the two-dimensional grid scene according to the target scene, node status and terminal load data into three types: core interaction area, general scene area and blocked interaction area, and mark the core interaction area with priority;

[0022] Randomly generate several initial interaction paths to form an initial interaction path set;

[0023] Constructing a fitness evaluation system, wherein the fitness evaluation system includes a path fluency evaluation item, an interactive response quantity evaluation item, and an operation complexity evaluation item;

[0024] The path fluency evaluation item is the cumulative value of the switching delay between scenes in the initial interaction path;

[0025] The interaction response evaluation item is the sum of the interaction traffic that each node in the initial interaction path can carry per unit time;

[0026] The operational complexity evaluation item is the number of terminal configurations and node parameters that need to be adjusted in the initial interaction path;

[0027] Taking the initial interaction path set as the initial population, iterative optimization is performed according to the fitness evaluation system to obtain the optimal interaction path and complete the interaction path planning.

[0028] Preferably, the initial interaction path set is used as the initial population, and iterative optimization is performed according to the fitness evaluation system, including:

[0029] According to the fitness evaluation system, the evaluation value of each initial interaction path is calculated, and the initial interaction paths are sorted from high to low according to the three evaluation items to obtain a sorted set of three initial interaction paths;

[0030] According to the evaluation value, several initial interaction paths are selected from the three sorted sets by probability to form three sub-populations in different directions. The probability of being selected is proportional to the evaluation value.

[0031] Within each subpopulation, genetic recombination, mutation adjustment, and individual learning are performed according to the genetic optimization algorithm, and unreachable path solutions are eliminated. The pre-set number of iterations are repeated to form three subpopulations after iteration.

[0032] Individuals are randomly selected from the three subpopulations based on their evaluation values ​​to form a basic gene pool. Three subpopulations are then cross-pollinated to form three groups of crossover subpopulations. For each crossover subpopulation, the three evaluation values ​​of each individual are calculated. A multi-objective ranking is performed on the two evaluation values ​​involved in the crossover subpopulations. Based on the ranking results, gene recombination, mutation adjustment, and individual learning are performed, and unreachable path solutions are eliminated.

[0033] After the three groups of crossover sub-populations have iterated a preset number of times, the three sub-populations are combined together, and the weighted sum of the three evaluation items is taken to obtain a comprehensive evaluation value. Gene recombination, mutation adjustment and individual learning are performed based on the comprehensive evaluation value. After iterating a preset number of times, the individual with the largest evaluation value is obtained as the optimal interaction path.

[0034] Preferably, the two evaluation values ​​involved in the crossover sub-population are subjected to multi-objective sorting, and gene recombination, variation adjustment and individual learning are performed according to the sorting results. Before the crossover sub-populations cross, a dynamic adjustment mechanism is used to select the crossover object, wherein the selection probability of the crossover object is related to the position of the individual in the sorting, and the later the position of the individual in the sorting, the higher the probability of being selected.

[0035] Preferably, the method further comprises: assigning a value to the selection probability according to the sorting order of the individuals to be crossed, wherein the later the individual is sorted, the greater the selection probability value.

[0036] Preferably, the method further comprises the following steps:

[0037] The matched interaction pairs are screened, and the mismatched interaction pairs are screened. When building the interaction model, it is constructed based on the screened interaction pairs.

[0038] Preferably, the matching interaction pairs are screened by using two-way verification and uniqueness verification to eliminate incorrectly matched interaction pairs.

[0039] Preferably, the present invention further includes an interactive system for a virtual exhibition hall, which is applied to the above-mentioned interactive method for a virtual exhibition hall. The system includes an interactive controller:

[0040] The interaction controller is used to obtain user interaction data collected by various terminals, extract behavioral features based on the user interaction data, and match the extracted behavioral features to obtain interaction pairs;

[0041] The interaction controller is also used to build an interaction model based on user interaction data and interaction pairs, and to perform scenario encapsulation on the interaction model based on the user interaction data to generate a standard interaction package.

[0042] Preferably, the interaction controller is further configured to generate an interaction path according to the exhibition hall topology data, and the various terminals collect and obtain user interaction data according to the interaction path.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The virtual exhibition hall interaction method and system provided by this invention captures user interaction data collected by multiple terminals and extracts and matches behavioral features, enabling comprehensive and accurate capture of user interaction intent. These terminals, including touch, voice, and somatosensory, can capture user behavior information from multiple dimensions. Feature descriptors are extracted, assigned time series labels, and then matching nodes are found to form interaction pairs. This enables the system to more accurately understand user behavior in the virtual exhibition hall, providing a more reliable basis for subsequent interactions.

[0045] When constructing the interaction model and scene encapsulation, we use interaction pairs as anchor points and continuously match them to obtain the interaction model. We then employ a layered scene encapsulation approach, moving from the original interaction space to the intermediate scene, and then to the target space to generate a standard interaction package. This approach builds a complete and coherent interaction model that covers the interaction paths in the target area, enriching and immersing the user experience in the virtual exhibition hall. Based on a scene segmentation approach, we expand and filter sparse scenes to obtain complete or partial interaction path models, ensuring the rationality and coverage of the interaction paths.

[0046] To generate interactive paths from exhibition hall topology data, a two-dimensional exhibition hall topology map is constructed and divided into grid scenes. Core interactive areas and other types are categorized and prioritized. The initial interactive path set is then iteratively optimized using a fitness evaluation system. This fitness evaluation system, which includes evaluation criteria such as path smoothness, interactive responsiveness, and operational complexity, comprehensively considers multiple factors to ultimately determine the optimal interactive path. This results in a smoother and more efficient interactive path, reduces switching delays, increases interactive responsiveness, and reduces operational complexity, ultimately enhancing the user's interactive experience in the virtual exhibition hall.

[0047] Matched interaction pairs are screened using bidirectional validation and uniqueness verification to eliminate incorrectly matched interaction pairs and ensure the accuracy of the interaction pairs used in building the interaction model. This improves the quality of the interaction model, making virtual exhibition hall interactions more accurate and reliable, and reducing the occurrence of misoperations.

[0048] The interactive controller in the interactive system performs the aforementioned operations, processing multi-terminal interaction data, building interaction models, encapsulating scenarios, and generating interaction paths. This makes the entire virtual exhibition hall interaction process more efficient, stable, and coordinated. From data collection to interaction model generation, to interaction path planning and interaction pair screening, each link works closely together, improving the overall performance of virtual exhibition hall interactions and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A diagram showing the working principle of the interactive method of the virtual exhibition hall according to the present invention;

[0050] Figure 2 Design drawings for exhibition hall topology data processing;

[0051] Figure 3 Design diagram for interactive path planning;

[0052] Figure 4 Design diagrams optimized for interaction paths;

[0053] Figure 5 This is a design diagram of the dynamic adjustment mechanism. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] See also Figure 1-Figure 5 The present invention relates to an interactive method for a virtual exhibition hall, and the specific implementation steps are as follows:

[0056] Obtain user interaction data collected by various terminals, including touch interaction terminals, voice interaction terminals, and motion interaction terminals. Perform behavioral feature extraction on each of the acquired user interaction data. Extract feature descriptors from the behavioral features. Feature descriptors are sequences of attributes that describe the behavioral features. Time series labels are assigned to the behavioral features. Matching nodes are found based on the feature descriptors to obtain matching interaction pairs.

[0057] Using the interaction pairs obtained through mutual matching as anchor points, the interaction model is obtained through continuous matching. Using a scenario segmentation approach, based on the sparse scenarios obtained by matching behavioral features between multiple terminals, a complete or partial interaction path model covering the target area is obtained through expansion and screening.

[0058] Layered scene encapsulation is adopted, and scene encapsulation is performed through two levels: the first level encapsulates the original interaction space into the intermediate scene that can be expressed by the protocol; the second level encapsulates the data of the intermediate scene into the target space that requires interaction to generate a standard interaction package. Example 1:

[0059] This embodiment describes in detail how to obtain the required interactive exhibition hall topology data. Exhibition hall topology data includes the target scenario, node status, and terminal load data for the desired interactive area. The target scenario refers to the specific scene in the virtual exhibition hall where interaction is required, such as a specific exhibition area or exhibit display area. Node status includes the operating status of each interactive node, such as availability and load. Terminal load data refers to the load of each interactive terminal, such as processing power and storage capacity.

[0060] The interaction area is constructed based on the boundary identification of the target scene. The boundary identification process can be achieved through image recognition technology, three-dimensional modeling technology, etc. For example, the target scene is modeled using three-dimensional modeling software, and then the boundary of the scene is identified through an algorithm to determine the scope of the interaction area. A two-dimensional exhibition hall topology map is constructed in the target scene to be interacted. The two-dimensional exhibition hall topology map can be obtained by projecting the three-dimensional scene onto a two-dimensional plane, or by drawing it through special drawing software. According to the two-dimensional exhibition hall topology map, the target scene to be interacted is divided into several grid scenes with a preset interval as the side length. The preset interval can be set according to actual needs. For example, it can be set to 1 meter, 2 meters, etc., and the target scene is divided into multiple grid scenes. The size of each grid scene is a square with a preset interval.

[0061] Based on the target scenario, node status, and terminal load data, the two-dimensional grid scenario is classified into three types: core interaction area, common scene area, and blocked interaction area. Core interaction areas are prioritized. Core interaction areas are areas with frequent and high-importance user interactions, such as the core display area for exhibits. Common scene areas are areas with relatively low user interactions. Blocked interaction areas are areas where interactions may be blocked due to factors such as node status or terminal load. Prioritizing core interaction areas allows them to be prioritized in subsequent interaction path planning.

[0062] Randomly generate several initial interaction paths to form an initial interaction path set. Initial interaction paths can be generated using a random algorithm. For example, starting from the starting point of the interaction area, randomly selecting the next grid scene until reaching the end point, forming an initial interaction path. Repeat this process multiple times to generate multiple initial interaction paths to form an initial interaction path set.

[0063] Construct a fitness evaluation system, which includes path fluency evaluation items, interactive response quantity evaluation items, and operation complexity evaluation items. Among them, the path fluency evaluation item is the cumulative value of the switching delay between scenes in the initial interactive path. The delay between scenes refers to the time required for the system to respond when the user moves from one grid scene to another. The smaller the cumulative value, the higher the path fluency. The interactive response quantity evaluation item is the sum of the interactive traffic that each node in the initial interactive path can carry per unit time. The interactive traffic that each node can carry per unit time is determined based on the node status and terminal load data. The larger the sum, the higher the interactive response quantity. The operation complexity evaluation item is the number of terminal configurations and node parameters that need to be adjusted in the initial interactive path. The fewer the number of terminal configurations and node parameters that need to be adjusted, the lower the operation complexity.

[0064] Using the initial interaction path set as the initial population, iterative optimization is performed according to the fitness evaluation system to obtain the optimal interaction path and complete the interaction path planning. The iterative optimization process is as follows: According to the fitness evaluation system, the evaluation value of each initial interaction path is calculated, and the initial interaction paths are sorted from high to low according to the three evaluation items, resulting in three sorted sets of initial interaction paths. For example, for the path fluency evaluation item, the initial interaction paths are sorted from small to large according to the cumulative value of the switching delay to obtain a sorted set; for the interaction response quantity evaluation item, the initial interaction paths are sorted from large to small according to the total interaction traffic to obtain another sorted set; for the operation complexity evaluation item, the terminal configuration and node parameters that need to be adjusted are sorted from small to large to obtain a third sorted set.

[0065] Based on the evaluation values, several initial interaction paths are probabilistically selected from each of the three sorted sets, forming three subpopulations in different directions. The probability of selection is proportional to the evaluation value. For example, in the path smoothness sorted set, initial interaction paths with higher evaluation values ​​(i.e., lower cumulative handover delay values) have a greater probability of selection. In this way, initial interaction paths that perform well in different evaluation criteria are selected to form three subpopulations.

[0066] Within each subpopulation, a genetic optimization algorithm performs genetic recombination, mutation adjustment, and individual learning, eliminating unreachable path solutions. Each iteration is repeated a preset number of times to form three subpopulations. Genetic recombination involves cross-pollinating individuals within a subpopulation to produce new individuals; mutation adjustment involves randomly mutating an individual's genes to increase population diversity; and individual learning involves improving an individual's fitness by emulating the characteristics of other excellent individuals. During the iteration process, unreachable path solutions, such as those containing inaccessible grid locations, must be continuously eliminated.

[0067] Individuals were randomly selected from the three subpopulations based on their evaluation scores to form a base gene pool. Three crossover subpopulations were then formed by pairwise crossover between the three subpopulations. For each crossover subpopulation, the three evaluation scores were calculated for each individual, and a multi-criteria ranking was performed on the two evaluation criteria involved in the crossover subpopulation. For example, the first crossover subpopulation involved path fluency and interactive response. A multi-criteria ranking was performed on these two criteria to determine the order of merit for the individuals.

[0068] Before the crossover subpopulations cross, a dynamic adjustment mechanism is used to select crossover targets. The probability of selection is assigned based on the order of the individuals to be crossed. The later the individual is ranked, the greater the selection probability. The later the individual is ranked, the higher the probability of selection. This dynamic adjustment mechanism can increase population diversity and prevent the algorithm from falling into local optimality.

[0069] Based on the sorting results, gene recombination, mutation adjustment, and individual learning are performed, and unreachable path solutions are eliminated. After the three groups of crossover subpopulations have been iterated a preset number of times, the three subpopulations are combined together, and a weighted sum of the three evaluation items is taken to obtain a comprehensive evaluation value. Based on this comprehensive evaluation value, gene recombination, mutation adjustment, and individual learning are performed. After a preset number of iterations, the individual with the highest evaluation value is determined as the optimal interaction path. The weights of the weighted summation can be set according to actual needs. For example, the weight of the path fluency evaluation item can be set to 0.4, the weight of the interactive response evaluation item to 0.3, and the weight of the operation complexity evaluation item to 0.3, etc.

[0070] Multiple terminals collect user interaction data based on this interaction path. Simultaneously, an interaction benchmark is established based on node status and terminal load data, and user interaction data is collected based on this interaction benchmark. This interaction benchmark can include interaction frequency, data transmission rate, and other factors to ensure that the collected user interaction data meets system requirements. Example 2:

[0071] This embodiment further defines the iterative optimization process based on the embodiment 1.

[0072] The evaluation value of each initial interaction path is calculated based on the fitness evaluation system. The fitness evaluation system includes path fluency evaluation items, interaction response evaluation items, and operation complexity evaluation items. Among them, the path fluency evaluation item corresponds to the cumulative value of the switching delay between each scene in the initial interaction path. The smaller the value, the smoother the path switching. The interaction response evaluation item is the sum of the interaction traffic that each node in the initial interaction path can carry per unit time. The larger the value, the stronger the overall carrying capacity of the node. The operation complexity evaluation item is the number of terminal configurations and node parameters that need to be adjusted in the initial interaction path. The smaller the number, the lower the difficulty of operation.

[0073] Based on these three evaluation criteria, the initial interaction paths are sorted from high to low based on their evaluation values, resulting in a sorted set of three initial interaction paths. For example, for the path fluency evaluation, all initial interaction paths are sorted from small to large by the cumulative handover delay value, with paths with smaller cumulative handover delay values ​​being ranked higher in the sorted set. For the interactive response volume evaluation, the paths are sorted from large to small by the total interactive traffic that each node can carry per unit time, with paths with larger total traffic being ranked higher. For the operational complexity evaluation, the paths are sorted from small to large by the number of terminal configurations and node parameters that require adjustment, with paths with fewer parameters being ranked higher.

[0074] Based on the evaluation value, several initial interaction paths are probabilistically selected from each of the three sorted sets, forming three subpopulations with different optimization directions. The probability of selection is proportional to the evaluation value; that is, paths with higher evaluation values ​​have a greater probability of being selected from the corresponding sorted set. For example, in the path fluency sorted set, paths with higher rankings have a higher probability of being selected than paths with lower rankings due to their lower cumulative switching delays and higher evaluation values. This probabilistic selection mechanism allows paths that excel in different evaluation dimensions to be selected into their corresponding subpopulations, thus forming three subpopulations with different optimization directions.

[0075] Within each subpopulation, a genetic optimization algorithm undergoes genetic recombination, mutation adjustment, and individual learning, eliminating unreachable path solutions. During genetic recombination, individual paths within the subpopulation are cross-combined and their genetic segments (i.e., the sequence of scene nodes within the path) are exchanged, generating new path individuals. Mutation adjustment randomly replaces or adjusts certain scene nodes within an individual's path to increase population diversity and prevent the algorithm from falling into a local optimum. Individual learning allows each individual path to optimize its node sequence by referencing the path structures of outstanding individuals within the subpopulation. During this process, path accessibility is continuously checked. If a path contains a scene node that is unreachable from the starting point or the destination is inaccessible from the end point, the path solution is eliminated. Each subpopulation undergoes a preset number of iterations (for example, 100 or 200, which can be set based on actual needs) to form three subpopulations.

[0076] Individuals are randomly selected from the three subpopulations based on their evaluation scores to form the base gene pool. This random selection is not completely equiprobable, but rather a probabilistic selection based on the evaluation scores, with individuals with higher evaluation scores being more likely to be selected. After the base gene pool is formed, the three subpopulations are crossed two by two to form three crossover subpopulations. For example, subpopulation A crosses with subpopulation B to form the first crossover subpopulation, subpopulation B crosses with subpopulation C to form the second, and subpopulation C crosses with subpopulation A to form the third.

[0077] For each set of crossover subpopulations, the three evaluation values ​​for each individual are calculated, and then a multi-objective ranking is performed on the two evaluation items involved in the crossover subpopulations. For example, the first crossover subpopulation is formed by the intersection of subpopulations A and subpopulation B. Subpopulation A mainly optimizes path fluency, and subpopulation B mainly optimizes interactive response. Therefore, this set of crossover subpopulations involves the path fluency evaluation item and the interactive response evaluation item, and these two evaluation items need to be sorted by multiple objectives. When sorting by multiple objectives, it is necessary to comprehensively consider the values ​​of the two evaluation items to determine the relative merits of each individual, rather than sorting based solely on a single evaluation item.

[0078] Before crossover subpopulations undergo crossover, a dynamic adjustment mechanism is employed to select crossover candidates. Specifically, the probability of selection is assigned based on the order of the individuals to be crossed. The later the individual is ranked, the greater the probability of selection. In other words, the later the individual is ranked, the higher the probability of being selected as a crossover candidate. For example, in a multi-target ranking of a crossover subpopulation, the probability of selection for the 10th-ranked individual is higher than that for the 1st-ranked individual. This mechanism prevents the algorithm from prematurely converging to a local optimum. By increasing the crossover probability for later-ranked individuals, it introduces more diverse genetic combinations, thereby enhancing the algorithm's global search capabilities.

[0079] Based on the sorting results, the crossover subpopulations undergo genetic recombination, mutation adjustment, and individual learning, similarly eliminating unreachable path solutions. The genetic recombination and mutation adjustment procedures are similar to those within the subpopulations. However, because the genes of two different subpopulations are combined, the individuals in the crossover paths may produce new path structures. Individual learning optimizes its own structure by referencing the path characteristics of outstanding individuals in the crossover subpopulations. After the three crossover subpopulations have been iterated a preset number of times, the three subpopulations are combined.

[0080] After combining, the three evaluation items are weighted and summed to obtain a comprehensive evaluation value. The weight of each evaluation item can be set based on the needs of the actual application scenario. For example, the weight of the path fluency evaluation item is 0.4, the interactive response evaluation item is 0.3, and the operation complexity evaluation item is 0.3. This weighted summation transforms the evaluation values ​​of the three different dimensions into a comprehensive evaluation value, allowing for a unified comparison of all individuals.

[0081] Based on the comprehensive evaluation value, the combined population undergoes genetic recombination, mutation adjustment, and individual learning, and iterates a preset number of times. During this iterative process, the individual path structure is continuously optimized, ultimately resulting in the individual with the highest evaluation value, which is the optimal interaction path. This iterative optimization process, through multi-objective sorting, dynamic crossover selection, and genetic manipulation, gradually improves the overall performance of the interaction path, ensuring that the planned path achieves an optimal balance between path smoothness, interactive responsiveness, and operational complexity. Example 3:

[0082] This embodiment illustrates that when constructing an interaction model, matched interaction pairs need to be screened to eliminate mismatched interaction pairs, thereby ensuring the accuracy of subsequent interaction model construction.

[0083] Matched interaction pairs are obtained by acquiring user interaction data collected by multiple terminals, and then going through steps such as behavioral feature extraction, feature descriptor extraction, timing label allocation, and matching node search. However, in this process, some interaction pairs may be mismatched due to factors such as terminal data collection errors, behavioral feature extraction bias, or matching algorithm limitations. Therefore, they need to be checked and verified through a screening mechanism.

[0084] The screening operation adopts the method of two-way verification and uniqueness verification. Two-way verification verifies information from both directions of the interaction pair to ensure the consistency and accuracy of the interaction relationship. For example, for a set of interaction pairs obtained by matching the behavioral features of the touch terminal and the voice terminal, it is necessary to verify whether the feature descriptors corresponding to the touch behavior and the feature descriptors corresponding to the voice behavior match each other in terms of timing labels and attribute sequences. Specifically, starting from the behavioral features of the touch terminal, according to its feature descriptors and timing labels, check whether there is a corresponding matching node in the voice terminal to form a forward verification; then starting from the behavioral features of the voice terminal, also according to its feature descriptors and timing labels, check whether there is a corresponding matching node in the touch terminal to form a reverse verification. Only when both the forward verification and the reverse verification are passed, the interaction pair is considered to be consistent in the two-way information, otherwise it is determined to be a mismatched interaction pair and is eliminated.

[0085] Uniqueness verification is used to eliminate duplicate or conflicting interaction pairs, ensuring their uniqueness and validity. In multi-terminal interaction scenarios, the same behavioral feature may be matched repeatedly by multiple terminals, or behavioral features from different terminals may be mistakenly matched, resulting in duplicate interaction pairs. For example, a motion-sensing interaction terminal and a touch-screen interaction terminal may simultaneously respond to a user action. If the matching algorithm is not precise, two interaction pairs with similar content but slightly different timing labels may be generated. These interaction pairs are considered duplicate interaction pairs. Uniqueness verification requires a traversal check of all matched interaction pairs, determining whether there are any duplications or conflicts based on feature descriptors and timing labels. Specifically, for each interaction pair, its feature descriptor (i.e., the sequence of attributes describing the behavioral features) and timing labels are compared with those of all other interaction pairs. If there are interaction pairs with highly similar feature descriptors and timing labels within the allowable error range, they are considered duplicate interaction pairs. One of these pairs (e.g., the one with the more accurate timing label or the more complete feature descriptor) is retained, while the remaining duplicate pairs are discarded. Conflicting interaction pairs (i.e., those with conflicting feature descriptors or obvious logical errors in the timing labels) are directly discarded.

[0086] During the bidirectional verification process, detailed verification rules need to be established. Taking the interaction between a touch terminal and a voice terminal as an example, the characteristic descriptor of touch behavior may include a sequence of attributes such as touch position, touch force, and touch duration. The characteristic descriptor of voice behavior may include a sequence of attributes such as voice keywords, volume, intonation, and voice duration. During bidirectional verification, not only must the semantic and logical correlation between the attribute sequences of touch behavior and voice behavior be checked—for example, whether the action of touching a certain exhibit model corresponds to the voice action of saying the exhibit's name—but also the rationality of the timing labels of both on the timeline must be ensured. That is, the time interval between the touch behavior and the voice action is within a preset reasonable range (e.g., between 0.5 seconds and 2 seconds). If the time interval is too long or too short, exceeding the reasonable range, the bidirectional verification is considered to have failed.

[0087] Uniqueness verification also needs to consider the differences in behavioral characteristics of different terminals and matching tolerances. Due to the different methods and accuracy of data collection by different terminals, for example, the accuracy of motion data collected by somatosensory interaction terminals is different from that of touch data collected by touch interaction terminals. When comparing feature descriptors, complete consistency cannot be required, but a reasonable matching tolerance should be set. For example, for the position coordinates of somatosensory actions and touch position coordinates, similar features are allowed within a certain coordinate range (such as a 5% error in the X, Y, and Z axes); for the recognition of voice keywords, features are allowed to be considered identical within a certain voice recognition error range (such as the phoneme matching degree of the keyword reaches more than 80%). By setting a matching tolerance, completely duplicate or conflicting interaction pairs can be effectively excluded, and reasonable interaction pairs caused by differences in terminal data can be avoided from being mistakenly eliminated.

[0088] After completing bidirectional validation and uniqueness verification, all filtered interaction pairs are used to construct the interaction model. Using the filtered interaction pairs as anchor points, the interaction model is obtained through continuous matching. Specifically, starting from a specific interaction pair, based on its feature descriptors and timing labels, other adjacent or related interaction pairs are searched for, gradually expanding the interaction network to form an interaction model covering specific interaction scenarios. During this expansion process, the feature information of the filtered interaction pairs is also referenced to ensure that the newly included interaction pairs are consistent with the existing interaction pairs in terms of logic and timing.

[0089] When constructing an interaction model using a scenario-based segmentation approach, the sparse scenarios obtained by matching behavioral features between multiple terminals (i.e., the scenarios covered by the interaction pairs obtained after initial screening may be relatively scattered) are expanded and filtered to obtain a complete or partial interaction path model covering the target area. During the expansion process, based on the filtered interaction pairs, possible interaction extension points are identified based on their feature descriptors and the scenario information of the target area, and new interaction pairs are added. The screening process involves a simple re-verification of the expanded interaction pairs to ensure the accuracy of the newly added interaction pairs and avoid the introduction of new mismatched pairs.

[0090] By performing bidirectional verification and uniqueness validation on matching interaction pairs, the accuracy of interaction pairs can be effectively improved, thereby ensuring the reliability and effectiveness of the interaction model. An accurate interaction model provides a solid foundation for subsequent scene encapsulation and generation of standard interaction packages, enabling the virtual exhibition hall's interactive system to more accurately respond to user interactions and enhance the user experience. This entire screening process, which runs through the early stages of interaction model construction, is a key step in ensuring the proper functioning of the interactive system. Through rigorous verification and validation mechanisms, the impact of mismatched interaction pairs is minimized, laying the foundation for the smooth operation of subsequent interaction processes. Example 4:

[0091] This embodiment relates to an interactive system for a virtual exhibition hall. This system applies the aforementioned interactive method, with the interactive controller as its core component. The interactive controller is responsible for key tasks within the system, including data acquisition, feature extraction, interaction pair matching, interaction model construction, and scenario encapsulation. The following describes its implementation in detail using specific examples.

[0092] For example, a virtual art exhibition hall allows users to interact with the hall through touchscreen terminals (such as tablets), voice-activated terminals (such as smart voice assistants), and motion-activated terminals (such as Kinect devices). The interaction controller first needs to acquire user interaction data collected by various terminals. When a user browses a painting in the virtual exhibition hall, the touchscreen terminal collects touch data from the user clicking on the painting to zoom in, including touch location and duration. The voice-activated terminal collects voice data from the user saying "introduce this painting," including voice waveforms and keywords. The motion-activated terminal collects motion data from the user approaching the painting, including movement trajectory and body movements. The interaction controller receives this interaction data from various terminals in real time, providing a foundation for subsequent processing.

[0093] Next, the interaction controller extracts behavioral features from the acquired user interaction data. For touch data, behavioral features such as the virtual exhibition hall coordinates corresponding to the touch position and the touch operation type (such as click, drag) are extracted. For voice data, behavioral features such as keywords (such as "introduction" and "painting") and voice intonation are extracted through voice recognition technology. For somatosensory data, behavioral features such as the speed, direction, and amplitude of the user's movement are extracted. For example, in a scenario where the user clicks on a painting and says "introduction", the behavioral feature of the touch terminal is the click operation at the virtual coordinates (100, 200), the behavioral feature of the voice terminal is a voice command containing the keyword "introduction" with an upward tone, and the behavioral feature of the somatosensory terminal is the user's action of moving 0.5 meters toward the painting.

[0094] After extracting the behavior features, the interaction controller extracts feature descriptors from the behavior features. These feature descriptors are sequences of attributes that describe the behavior features. For a touch behavior feature, the feature descriptor might be "{coordinates: 100, 200; action type: click; timestamp: 2025-06-2010:00:01}"; for a voice behavior feature, the feature descriptor might be "{keywords: introduction, painting; tone: rising; timestamp: 2025-06-2010:00:02}"; and for a somatosensory behavior feature, the feature descriptor might be "{movement direction: forward; movement distance: 0.5 meters; timestamp: 2025-06-2010:00:00.5}." At the same time, a time sequence label is assigned to the behavioral features. The time sequence label is determined based on the time sequence of data collection and is used to identify the time sequence of the behavioral features. For example, the timestamp of the above-mentioned somatosensory data is 10:00:00.5, the touch data is 10:00:01, and the voice data is 10:00:02, forming a time sequence.

[0095] The interaction controller searches for matching nodes based on feature descriptors to obtain matching interaction pairs. For example, the "coordinates 100, 200" in the touch behavior's feature descriptor corresponds to the location of a painting in the virtual exhibition hall, and the "introduce the painting" in the voice behavior's feature descriptor corresponds to an instruction introducing the painting. The timestamps of the two are within a reasonable time range (e.g., within 1 second) and are semantically related. The interaction controller uses a preset matching algorithm to search the virtual exhibition hall's interaction node library for nodes that match these feature descriptors. It finds that the interaction node for the painting contains response logic for both touch clicks and voice introductions, and therefore matches the touch and voice behaviors into a pair of interactions. Similarly, in the somatosensory behavior, the user's movement toward the painting and the touch click are continuous in time, and the movement direction points to the painting's location. The interaction controller can match the somatosensory and touch behaviors into another pair of interactions, forming a combination of multi-terminal interaction pairs.

[0096] After obtaining the interaction pairs, the interaction controller uses these matching interaction pairs as anchor points and continuously matches them to obtain the interaction model. Still using the scenario of a user browsing a painting as an example, the initial matching interaction pairs are (touch click, voice introduction) and (body movement, touch click). The interaction controller uses these interaction pairs as a starting point to search for other related interaction behaviors. For example, after hearing the introduction of the painting, the user may issue a "zoom in details" command through the voice terminal while simultaneously performing a drag operation on the touch terminal. The interaction controller continuously matches this new set of voice and touch behaviors with the previous interaction pairs to confirm that they belong to extended behaviors within the same interaction scenario. It then incorporates the new interaction pairs into the interaction model, gradually expanding the interaction network.

[0097] When building the interaction model, a scenario-based segmentation approach is employed. Based on the sparse scenarios generated by matching behavioral features across multiple terminals, a complete or partial interaction path model covering the target area is expanded and filtered. For example, the initially matched interaction pairs only cover the user's actions of approaching a painting, clicking on it, and requesting an introduction, forming a sparse scene fragment. Based on the virtual exhibition hall's topology and the target area (i.e., the painting's display area), the interaction controller expands possible interaction behaviors, such as the user zooming in on details or rotating the painting, to filter out reasonable interaction paths, forming a complete interaction path model from the user entering the area to the completion of the interaction.

[0098] The interaction controller employs layered scene encapsulation, generating a standard interaction package through two levels of scene encapsulation. The first level encapsulates the original interaction space into an intermediate scene expressible by a protocol. The original interaction space consists of the various interactive behavior data generated by users across multiple terminals. The interaction controller converts this data into intermediate scene data that conforms to a specific protocol (such as JSON format). For example, actions such as touch clicks and voice commands are converted into a protocol-specified format, containing fields such as action type, parameters, and timestamps. The second level encapsulates the intermediate scene data into the target space for interaction. The target space is the specific interactive environment of the virtual exhibition hall. The interaction controller maps the intermediate scene data to the target space model (such as a 3D model of a painting or a display stand) to generate a standard interaction package. The standard interaction package contains information about all interaction pairs, the interaction path model, and the mapping relationship to the target space. This package can be used by the virtual exhibition hall's rendering engine and interaction logic module to respond to user interactions, such as displaying a description of the painting or zooming in on details.

[0099] Throughout the entire process, the interaction controller coordinates and processes interaction data from multiple terminals, ensuring that the behavioral characteristics of different terminals are accurately extracted and matched, and that the constructed interaction model truly reflects the user's interaction intent. The layered encapsulation generates a standard interaction package that provides a unified interface and data format for virtual exhibition hall interactions, enabling the system to efficiently and accurately respond to user operations and enhance the interactive experience of the virtual exhibition hall. For example, when a user simultaneously approaches a painting through motion sensing, selects it through touch and click, and requests an introduction through voice, the interaction controller can quickly process these three types of data, match interaction pairs, build an interaction model, and generate a standard interaction package. This drives the virtual exhibition hall to simultaneously execute operations such as moving the viewing angle, highlighting the painting, and playing an introduction voice, achieving seamless multimodal interaction. Example 5:

[0100] In this embodiment, the interactive controller in the virtual exhibition hall interactive system, in addition to the functions described in Example 4, is also used to obtain exhibition hall topology data and generate interaction paths. Multiple terminals collect user interaction data based on these paths. The following describes a specific application scenario of a virtual exhibition hall for technology products in detail.

[0101] Suppose a user enters a virtual exhibition hall and wishes to browse a brand's latest smart device display area. The interaction controller obtains the exhibition hall topology data for the desired interaction, which includes the target scene, node status, and terminal load data for the desired interaction area. The target scene is the three-dimensional model and layout of the smart device display area, including the display location of each exhibit and interaction nodes (such as clicking to view details, dragging and rotating models, etc.); the node status indicates the operating status of each interaction node, such as whether the 3D model loading node for exhibit A is currently available, or whether the video playback node for exhibit B is overloaded; the terminal load data records the real-time load of touch interaction terminals, voice interaction terminals, and somatosensory interaction terminals, such as the CPU occupancy rate and memory usage of a touch terminal.

[0102] The interactive controller performs boundary recognition based on the target scene to construct an interactive area. The target scene of the smart device exhibition area is processed through three-dimensional modeling software, and the scope of the interactive area is determined using a boundary recognition algorithm (such as a recognition method based on geometric shapes). For example, the walls around the exhibition area and the edges of the booths are used as boundaries to form a closed interactive area. Next, a two-dimensional exhibition hall topology map is constructed in the target scene to be interacted with, and the three-dimensional exhibition area model is projected onto a two-dimensional plane, marking the location of each exhibit, the coordinates of the interactive nodes, and the direction of the channel. According to the two-dimensional exhibition hall topology map, the target scene to be interacted with is divided into several grid scenes with a preset interval as the side length. The preset interval is set to 1 meter. In this way, the entire exhibition area is divided into multiple 1 meter × 1 meter grids, and each grid corresponds to a specific area in the virtual exhibition hall.

[0103] The interaction controller categorizes the two-dimensional grid scenarios based on the target scenario, node status, and terminal load data into three types: core interaction zone, general scenario zone, and blocked interaction zone. The core interaction zones are prioritized. In the smart device exhibition area, the area showcasing the latest smartwatches is classified as the core interaction zone due to high user interaction demand, good interaction node status, and low terminal load. The area displaying accessories is classified as the general scenario zone due to low interaction demand. Finally, an exhibit near the entrance is classified as the blocked interaction zone due to a faulty interaction node and high corresponding terminal load. The core interaction zone is assigned the highest priority for prioritization in interaction path planning.

[0104] After classification, the interaction controller randomly generates several initial interaction paths to form an initial interaction path set. For example, initial path 1 is: entry → smartwatch core interaction area → headset general scene area → exit; initial path 2 is: entry → smart speaker general scene area → smartwatch core interaction area → exit; initial path 3 is: entry → blocked interaction area → smartwatch core interaction area → exit (this path contains a blocked interaction area and will be optimized in subsequent iterations). Through random generation, multiple different initial paths are obtained, providing a foundation for subsequent optimization.

[0105] Next, the interaction controller constructs a fitness evaluation system, which includes evaluation items for path fluency, interactive response, and operation complexity. Taking initial path 1 as an example, the path fluency evaluation item is the cumulative value of the switching delay between each scene. Assuming the switching delay from the entrance to the smartwatch core interactive area is 0.3 seconds, the switching delay from the core interactive area to the headset's general scene area is 0.2 seconds, and the switching delay from the general scene area to the exit is 0.1 seconds, the cumulative value is 0.6 seconds. The interactive response evaluation item is the sum of the interactive traffic that each node can carry per unit time. The nodes in the smartwatch core interactive area can carry 50 traffic per second, and the nodes in the headset's general scene area can carry 20 traffic per second, for a total of 70 traffic per second. The operation complexity evaluation item is the number of terminal configurations and node parameters that need to be adjusted. In this path, no terminal configuration or node parameter adjustments are required, so the number is 0.

[0106] The interactive controller uses the initial interaction path set as the initial population and performs iterative optimization according to the fitness evaluation system. First, the evaluation value of each initial interaction path is calculated and sorted according to the three evaluation items. For example, in the path fluency evaluation item, the cumulative switching delay of the initial path 2 is 0.5 seconds, which is better than the 0.6 seconds of path 1. Because path 3 contains a blocked interaction area, the cumulative switching delay is 1.2 seconds, and the ranking is path 2>path 1>path 3; in the interaction response evaluation item, the total of path 1 is 70 times / second, and the total traffic that can be carried by the nodes in the general scene area and the core interaction area of ​​the smart speaker in path 2 is 60 times / second, and the ranking is path 1>path 2>path 3; in the operation complexity evaluation item, the number of paths 1 and 2 is both 0. Path 3 may need to adjust the node parameters because it needs to bypass the blocked interaction area. The number is 1, and the ranking is path 1=path 2>path 3.

[0107] Based on the evaluation values, several initial interaction paths are probabilistically selected from the three sorted sets to form three subpopulations. Assume that in the path fluency sorted set, path 2 has the highest evaluation value and the highest probability of selection, followed by path 1, and path 3 has the lowest. In the interaction response sorted set, path 1 has the highest probability of selection. Within each subpopulation, genetic recombination, mutation adjustment, and individual learning are performed using a genetic optimization algorithm. Unreachable path solutions are eliminated. For example, path 3 is eliminated because it contains a blocked interaction zone, making some nodes unreachable. After a preset number of iterations, the three subpopulations are formed.

[0108] Individuals are randomly selected from the three subpopulations to form the base gene pool. Crossover is then performed between the subpopulations to form three crossover subpopulations. For each crossover subpopulation, three evaluation values ​​are calculated for the individuals, followed by a multi-objective ranking. A dynamic adjustment mechanism is used to select crossover candidates. Individuals ranked lower in the ranking have a higher probability of being selected. For example, within a crossover subpopulation, individuals ranked lower in the ranking are more likely to be selected during crossover, thereby increasing population diversity. After crossover, genetic recombination, mutation adjustment, and individual learning are performed to eliminate unreachable solutions. The process is repeated a preset number of times.

[0109] The three subpopulations were combined, and a weighted sum of the three evaluation items was calculated to obtain a comprehensive evaluation value, assuming weights of 0.4 for path fluency, 0.4 for interactive responsiveness, and 0.2 for operational complexity. The individual with the highest evaluation value was determined as the optimal interaction path. For example, the optimized path could be: entrance → smartwatch core interaction area → smart headset general scenario area → smart speaker general scenario area → exit. This path strikes a balance between fluency, responsiveness, and operational complexity.

[0110] Multiple terminals collect and acquire user interaction data based on this interaction path. When a user browses along the optimal path, the touch terminal collects touch data from the user clicking on the smartwatch model to view details, the voice terminal collects voice data from the user saying "play headphone function introduction", and the somatosensory terminal collects trajectory data of the user moving within the exhibition area. At the same time, the interaction controller establishes an interaction benchmark based on node status and terminal load data. For example, it sets the interaction frequency limit for the core interaction zone to 30 times / minute and the terminal data transmission rate to no less than 10Mbps. This ensures that the collected user interaction data meets the benchmark requirements and provides accurate data for subsequent behavioral feature extraction and interaction pair matching.

[0111] By processing exhibition hall topology data and generating interactive paths through the interactive controller, combined with path-based data collection by multiple terminals, the virtual exhibition hall can guide users efficiently through core interactive areas, rationally allocate terminal resources, and enhance the overall interactive experience. For example, in the aforementioned smart device exhibition area, the optimal interactive path avoids congested areas and prioritizes users to core exhibits. Terminals collect data based on the path and benchmarks, enabling the system to accurately respond to user actions and achieve smooth multimodal interaction.

[0112] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0113] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for interacting with a virtual exhibition hall, characterized in that: Includes the following: Obtain user interaction data collected by multiple terminals, extract behavioral features based on the user interaction data, and match them to obtain interaction pairs based on the extracted behavioral features; The various terminals include touch interactive terminals, voice interactive terminals and somatosensory interactive terminals; Matching is performed based on the extracted behavioral features to obtain interaction pairs, including: extracting feature descriptors at the behavioral features, assigning time series labels to the behavioral features, and finding matching nodes based on the feature descriptors to obtain matching interaction pairs; wherein the feature descriptors are attribute sequences that describe the behavioral features; Build an interaction model based on user interaction data and interaction pairs, and encapsulate the interaction model into scenarios based on the user interaction data to generate a standard interaction package, including: The interaction model is obtained through continuous matching using the interaction pairs obtained by mutual matching as anchor points. A scenario segmentation method is used to expand and filter the sparse scenarios obtained by matching the behavioral characteristics of multiple terminals to obtain a complete or partial interaction path model covering the target area. Layered scene encapsulation is adopted, and scene encapsulation is performed through two levels. The first level encapsulates the original interaction space into the intermediate scene that can be expressed by the protocol; the second level encapsulates the data of the intermediate scene into the target space that requires interaction to generate a standard interaction package.

2. The method for interacting with a virtual exhibition hall according to claim 1, characterized in that: Also included: Obtain the exhibition hall topology data required for interaction and generate the interaction path based on the exhibition hall topology data; multiple terminals collect user interaction data based on the interaction path; The exhibition hall topology data includes the target scenario, node status, and terminal load data of the required interaction area. The interaction path is generated based on the exhibition hall topology data, including the following: Identify boundaries based on the target scenario to build an interaction area, and plan interaction paths based on the interaction area. And establish an interaction benchmark based on node status and terminal load data, and collect user interaction data based on the interaction benchmark.

3. The method for interacting with a virtual exhibition hall according to claim 2, characterized in that: Planning the interaction path based on the interaction area includes: Construct a two-dimensional exhibition hall topology map in the target scene to be interacted with, and divide the target scene to be interacted with into a number of grid scenes with preset intervals as the side length according to the two-dimensional exhibition hall topology map; Classify the two-dimensional grid scene according to the target scene, node status and terminal load data into three types: core interaction area, general scene area and blocked interaction area, and mark the core interaction area with priority; Randomly generate several initial interaction paths to form an initial interaction path set; Constructing a fitness evaluation system, wherein the fitness evaluation system includes a path fluency evaluation item, an interactive response quantity evaluation item, and an operation complexity evaluation item; The path fluency evaluation item is the cumulative value of the switching delay between scenes in the initial interaction path; The interactive response evaluation item is the sum of the interactive traffic that each node in the initial interactive path can carry per unit time; The operational complexity evaluation item is the number of terminal configurations and node parameters that need to be adjusted in the initial interaction path; Taking the initial interaction path set as the initial population, iterative optimization is performed according to the fitness evaluation system to obtain the optimal interaction path and complete the interaction path planning.

4. The method for interacting with a virtual exhibition hall according to claim 3, characterized in that: Using the initial interaction path set as the initial population, iterative optimization based on the fitness evaluation system includes: According to the fitness evaluation system, the evaluation value of each initial interaction path is calculated, and the initial interaction paths are sorted from high to low according to the three evaluation items to obtain a sorted set of three initial interaction paths; According to the evaluation value, several initial interaction paths are selected from the three sorted sets by probability to form three sub-populations in different directions. The probability of being selected is proportional to the evaluation value. Within each subpopulation, genetic recombination, mutation adjustment, and individual learning are performed according to the genetic optimization algorithm, and unreachable path solutions are eliminated. The pre-set number of iterations are repeated to form three subpopulations after iteration. Individuals are randomly selected from the three subpopulations based on their evaluation values ​​to form a basic gene pool. Three subpopulations are then cross-pollinated to form three groups of crossover subpopulations. For each crossover subpopulation, the three evaluation values ​​of each individual are calculated. A multi-objective ranking is performed on the two evaluation values ​​involved in the crossover subpopulations. Based on the ranking results, gene recombination, mutation adjustment, and individual learning are performed, and unreachable path solutions are eliminated. After the three groups of crossover sub-populations have iterated a preset number of times, the three sub-populations are combined together, and the weighted sum of the three evaluation items is taken to obtain a comprehensive evaluation value. Gene recombination, mutation adjustment and individual learning are performed based on the comprehensive evaluation value. After iterating a preset number of times, the individual with the largest evaluation value is obtained as the optimal interaction path.

5. The method for interacting with a virtual exhibition hall according to claim 4, characterized in that: The two evaluation values ​​involved in the crossover sub-population are sorted by multiple objectives, and gene recombination, variation adjustment and individual learning are performed according to the sorting results. Before the crossover sub-population crosses, a dynamic adjustment mechanism is used to select the crossover object, wherein the selection probability of the crossover object is related to the position of the individual in the sorting. The later the position of the individual in the sorting, the higher the probability of being selected.

6. The method for interacting with a virtual exhibition hall according to claim 5, characterized in that: Also includes: The selection probability is assigned according to the sorting order of the individuals to be crossed. The later the individual is sorted, the greater the selection probability value.

7. The method for interacting with a virtual exhibition hall according to claim 1, characterized in that: Also included: The matched interaction pairs are screened, and the mismatched interaction pairs are screened. When building the interaction model, it is constructed based on the screened interaction pairs.

8. The method for interacting with a virtual exhibition hall according to claim 7, characterized in that: The matching interaction pairs are screened by using two-way verification and uniqueness verification to eliminate incorrectly matched interaction pairs.

9. An interactive system for a virtual exhibition hall, applied to the interactive method for a virtual exhibition hall according to any one of claims 1 to 8, comprising an interactive controller, characterized in that: The interaction controller is used to obtain user interaction data collected by various terminals, extract behavioral features based on the user interaction data, and match the extracted behavioral features to obtain interaction pairs; The interaction controller is also used to build an interaction model based on user interaction data and interaction pairs, and to perform scenario encapsulation on the interaction model based on the user interaction data to generate a standard interaction package.

10. The interactive system of a virtual exhibition hall according to claim 9, characterized in that: The interaction controller is also used to generate an interaction path based on the exhibition hall topology data, and various terminals collect and obtain user interaction data based on the interaction path.

Images