A three-dimensional visualization interaction display method and system based on a BIM model

Abstract

This invention provides a 3D visualization interactive display method and system based on a BIM model, relating to the field of 3D display interaction. The method includes the following steps: constructing a BIM model of the target building; deconstructing and recombining the components of the BIM model of the target building; generating a contextualized focused view containing component parameters based on the recombination results; constructing an intent-view mapping table containing interactive intents and component parameters based on the contextualized focused view containing component parameters and according to preset intent identifiers; loading the corresponding contextualized focused view data from the intent-view mapping table; performing real-time reconstruction and presentation using a preset 3D rendering engine, and establishing a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of the 3D visualization interactive display. This invention significantly shortens the view loading response time, improves interaction efficiency, and ultimately achieves a highly efficient, accurate, and evolvable BIM 3D visualization interactive display effect.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional display and interaction, and more specifically, to a three-dimensional visualization and interactive display method and system based on BIM models. Background Technology

[0002] Building Information Modeling (BIM) technology, as the core of the digital transformation of the construction industry, provides 3D visualization, which is a key means of information communication and decision support in design review, construction management, and operation and maintenance analysis. Traditional BIM visualization systems mainly rely on traditional input devices such as mice and keyboards, and are operated through hierarchical menus, toolbars, and predefined views. This approach suffers from problems such as cumbersome operation, high learning costs, and unintuitive interaction. In particular, it is difficult for non-professional users or on-site workers to quickly locate the required information and perform complex queries, which seriously affects the full realization of the value of BIM data.

[0003] Currently, with the development of human-computer interaction technology, some research has begun to explore the introduction of natural interaction methods such as gesture recognition and voice control into the field of 3D visualization, aiming to lower the operational threshold. However, existing methods mostly focus on simple mapping of single interaction modalities, such as mapping specific gestures to rotation and zoom commands, or simple responses to voice commands in general scenarios, lacking a deep understanding of architectural expertise and complex user intentions. Furthermore, most systems employ a forward interaction logic, requiring users to actively navigate and filter to find targets within a massive model, a process that is inefficient. Simultaneously, existing BIM visualization interaction systems are typically static and fixed. They cannot adaptively adjust based on users' historical behavior, role preferences, or specific task scenarios, lacking the ability to continuously learn and optimize from interaction feedback.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of this, the present invention provides a three-dimensional visualization and interactive display method and system based on BIM model to solve the above-mentioned problems.

[0006] To solve the above problems, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a three-dimensional visualization and interactive display method based on a BIM model is provided, comprising the following steps: S1. Acquire the laser point cloud data of the target building and, in conjunction with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology; S2. Construct and collect interactive data from users that includes spatial behavior commands and semantic commands. Combined with the reverse recognition and parsing mechanism of interactive intent, deconstruct and reassemble the components of the BIM model of the target building. Based on the reassembly results, generate a contextualized focused view containing component parameters. S3. Based on the contextualized focused view containing component parameters and according to the preset intent identifier, construct an intent-view mapping table containing interactive intents and component parameters. S4. Collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focused view data from the intent-view mapping table; S5. Based on contextualized focused view data, the system utilizes a preset 3D rendering engine for real-time reconstruction and presentation, and establishes a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

[0007] Preferably, the data interaction channel that constructs and collects user-issued interactive data containing spatial behavior commands and semantic commands, combined with a reverse identification and parsing mechanism for interactive intent, deconstructs and reassembles the components of the target building's BIM model; and generates a contextualized focused view containing component parameters based on the reassembly results includes the following steps: S21. Based on the preset interaction method, construct a data interaction channel that includes a spatial gesture input channel and a language input channel; S22. Through multimodal fusion analysis technology, and by using spatial gesture input and language input collected through the data interaction channel for semantic parsing, we identify and generate intent feature data containing interaction intent vectors. S23. Based on intent feature data and combined with a semantic rule engine, reverse deconstruct and reorganize the components of the BIM model, and generate a contextualized focused view that matches the current interaction intent and includes component parameters.

[0008] Preferably, the step of using multimodal fusion analysis technology and performing semantic parsing on spatial gesture input and language input collected through the data interaction channel to identify and generate intent feature data containing interaction intent vectors includes the following steps: S221. Based on the preset architectural gesture semantic library, perform trajectory analysis on spatial gesture input data, and combine similarity analysis to identify gesture types and generate gesture semantic feature vectors. S222. Based on natural language processing technology, perform language semantic parsing on language input data to obtain language semantic feature vectors; S223. Through an adaptive weight fusion mechanism, the gesture semantic feature vector and the language semantic feature vector are weighted and fused to generate intent feature data containing the interaction intent vector.

[0009] Preferably, the process of reverse deconstructing and reorganizing the components of the BIM model based on intent feature data and combined with a semantic rule engine, and generating a contextualized focused view that matches the current interaction intent and includes component parameters, includes the following steps: S231. Use a semantic rule engine to parse and process the intent feature data, and extract the intent type, association range and accuracy requirements respectively; S232. Based on the accuracy requirements, match the initial set of components corresponding to the accuracy requirements from the BIM model; S233. Based on the association range, select a set of secondary components that match the association range from the initial component set, and determine the construction parameters that include geometric structure and spatial relationship; S234. Based on intent type, the corresponding view operation rules are called through the semantic rule engine to dynamically reorganize the components in the secondary component set to generate view state description data containing construction parameters. S235. Based on the view state description data, and combined with the 3D rendering engine, perform real-time rendering to generate a contextualized focused view that matches the current interaction intent and includes construction parameters.

[0010] Preferably, the step of using a semantic rule engine to parse and process the intent feature data, and extracting the intent type, association range, and accuracy requirements respectively, includes the following steps: S2311. Standardize and format the intent feature data, and convert the standardized and formatted intent feature data into a structured data object containing identifiable fields for processing by the semantic rule engine. S2312. Match the structured data object with the preset intent classification rule base, combine the rule priority and feature confidence to make a decision, determine and output the intent type; S2313. Based on structured data objects and combined with the spatial index of the BIM model, calculate and extract the associated range; S2314. Based on the determined intent type, match the corresponding accuracy level from the preset BIM accuracy rule library to generate accuracy requirements.

[0011] Preferably, the step of matching structured data objects with a preset intent classification rule base, combining rule priority and feature confidence to determine and output the intent type includes the following steps: S23121. Decompose the structured data object and extract key features for intent matching based on the decomposition results. Standardize and quantify the confidence data based on the key features. S23122. Perform parallel matching of key features with rule conditions in the preset intent classification rule base, calculate the condition satisfaction of each rule condition, and combine the standardized and quantified confidence data to generate a matching score and rule trigger identifier for each rule. S23123. Based on the rule triggering identifier, a set of candidate rules with matching scores higher than a preset threshold is selected. Based on rule priority, matching score and feature confidence, the set of candidate rules is adjudicated through a preset multi-layer adjudication logic to determine the optimal intent classification rule. S23124. Extract the intent type identifier defined by the optimal intent classification rule, and combine it with its matching score to generate an output containing the intent type.

[0012] Preferably, the step of constructing an intent-view mapping table containing interactive intents and component parameters based on a contextualized focused view containing component parameters and according to preset intent identifiers includes the following steps: S31. Extract component geometric parameters based on contextualized focused views, and calculate the topological feature vectors of components containing connection density and hierarchy depth through the BIM model topology analyzer. S32. Based on the negative sample learning mechanism, collect user records of rejection and correction operations on automatically generated views, and construct a reverse training sample set containing bias features to optimize the anti-interference ability of view generation. S33. Based on topological feature vectors and user role information, and combined with BIM semantic knowledge graph, intention-parameter semantic reasoning is performed to generate an initial mapping relationship containing implicit parameters. S34. Construct a dynamic evolution index structure with self-optimization capability, and perform adversarial training between the initial mapping relationship and the reverse training sample set to generate mapping weights with anti-interference capability. S35. Obtain user feedback and adjust mapping weights in real time to form an intent-view mapping table that can adapt to scene changes and user preferences.

[0013] Preferably, the step of performing real-time reconstruction and presentation based on contextualized focused view data using a preset 3D rendering engine, and establishing a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display includes the following steps: S51. Based on the intent-view mapping table and combined with the topological features of the BIM model, a layered rendering strategy is adopted to reconstruct the view in real time and to visualize the component parameters. S52. Collect user interaction feedback data on the real-time view, including parameter adjustment, view modification, and operation rejection records, and construct a positive and negative feedback sample set containing operation intentions and deviation characteristics. Perform intention-feedback mapping through multimodal fusion analysis technology. S53. Based on the positive and negative feedback sample set, the intention-view mapping table is dynamically weighted using an adversarial training mechanism and an adaptive optimization algorithm to achieve continuous evolution of the mapping table.

[0014] Preferably, the real-time view reconstruction based on the intent-view mapping table, combined with the topological features of the BIM model, and the visualization of component parameters using a layered rendering strategy includes the following steps: S511. Based on the view state description data associated in the intent-view mapping table, extract the topological feature vector of the target component, and calculate the rendering priority based on the connection density and hierarchy depth in the topological feature vector and the interactive intent. S512. Based on rendering priority, construct a layered rendering queue, and dynamically allocate the rendering budget and level of detail for each layer according to the current performance of the visualization terminal to obtain the dynamic allocation result. S513. An asynchronous progressive rendering pipeline is used, and combined with dynamic allocation results, rendering tasks are performed on the target components according to the layered rendering queue. S514. For the rendered target component, automatically adapt and overlay a parameter visualization layer according to its component parameter type to generate a visualization output that integrates three-dimensional spatial morphology and semantic parameters.

[0015] According to another aspect of the present invention, a three-dimensional visualization and interactive display system based on a BIM model is provided, the system comprising: The model building module is used to acquire laser point cloud data of the target building and, in combination with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology. The focused view generation module is used to build and collect data interaction channels containing user-issued interactive data including spatial behavior commands and semantic commands. Combined with the reverse recognition and parsing mechanism of interactive intent, it deconstructs and reassembles the components of the BIM model of the target building; and generates a contextualized focused view containing component parameters based on the reassembly results. The mapping table construction module is used to construct an intent-view mapping table containing interactive intents and component parameters based on the contextualized focused view containing component parameters and according to the preset intent identifiers. The data loading module is used to collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focus view data from the intent-view mapping table. The visualization module is used to reconstruct and present the data in real time using a preset 3D rendering engine based on contextualized focused view data, and to establish a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

[0016] The beneficial effects of this invention are as follows: 1. This invention constructs a high-precision BIM model by fusing laser point clouds with design parameters. Based on a multimodal interaction channel using touch gestures and language, and an intent reverse recognition and parsing mechanism, it achieves natural and accurate interaction between users and the BIM model. Furthermore, it generates contextualized focused views containing complete parameters through component deconstruction and recombination to meet information viewing needs in different scenarios. The intent-view mapping table construction and matching mechanism significantly shortens the view loading response time and improves interaction efficiency, ultimately achieving a highly efficient, accurate, and evolvable BIM 3D visualization interactive display effect.

[0017] 2. This invention constructs a dual input channel combining touch gestures and language, and combines a customized architectural gesture semantic library with a domain-specific speech recognition model to achieve single-modal semantic parsing. Through an adaptive weight fusion mechanism, it accurately extracts multimodal intent features, effectively improving the robustness and accuracy of interactive intent recognition. At the same time, with the help of the standardized processing, multi-layer adjudication logic and spatial index matching of the semantic rule engine, it achieves efficient decomposition of intent type, association range and accuracy requirements. Finally, it renders and generates a contextualized focused view that highly matches the user's intent and contains complete component geometry and spatial parameters, which not only ensures the naturalness and convenience of interaction, but also improves the accuracy and efficiency of view generation.

[0018] 3. This invention calculates component rendering priority, constructs a layered rendering queue, and dynamically allocates resources based on terminal performance. Combined with an asynchronous progressive rendering pipeline, it ensures both the high efficiency of prioritizing the presentation of core components and the adaptability of rendering effects to terminal performance. Simultaneously, by automatically overlaying parameter visualization layers, it generates a visual output that integrates 3D models and semantic parameters, satisfying users' dual needs for viewing component shapes and parameters. Furthermore, it utilizes user feedback data to construct positive and negative sample sets, and dynamically adjusts the mapping table weights using adversarial training and adaptive optimization algorithms, forming a closed loop of rendering presentation, feedback collection, and model optimization. This enables continuous iteration and precise evolution of the 3D visualization interactive display effect. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of a three-dimensional visualization and interactive display method based on a BIM model according to an embodiment of the present invention; Figure 2This is a schematic diagram of a three-dimensional visualization and interactive display system based on a BIM model according to an embodiment of the present invention.

[0020] In the picture: 1. Model building module; 2. Focused view generation module; 3. Mapping table building module; 4. Data loading module; 5. Visualization module. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0022] According to an embodiment of the present invention, a three-dimensional visualization and interactive display method and system based on a BIM model are provided.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a three-dimensional visualization and interactive display method based on a BIM model is provided, comprising the following steps: S1. Acquire the laser point cloud data of the target building and, in conjunction with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology; It should be noted that the laser point cloud data of the target building can be obtained by scanning the entire space of the target building using a ground-based 3D laser scanner or an UAV-borne laser scanner. The target building can be a building under construction, a completed building, or a building to be renovated. By collecting a set of 3D spatial coordinate points of the building entity, laser point cloud data is formed. Laser point cloud data is a point cloud matrix composed of a massive number of discrete 3D coordinate points (including X, Y, and Z spatial coordinates). Each point can also carry additional information such as reflection intensity and color. The design parameters of the target building include design drawing parameters, CAD electronic file data, and technical specifications such as building fire protection codes and structural seismic resistance requirements, for architectural, structural, and mechanical and electrical engineering disciplines.

[0024] Therefore, when constructing the BIM model of the target building, computer graphics technology is used to preprocess the collected laser point cloud data, including denoising, registration, and simplification. Then, computer vision algorithms, such as edge detection and region growing algorithms, are used to extract the features of building components from the preprocessed point cloud, such as the planar features of walls and the column features of beams / columns, and the point cloud is segmented according to component type. Next, 3D model reconstruction technology is used to perform geometric modeling on the segmented component point cloud in conjunction with design parameters. Finally, the component attributes in the design parameters are associated and bound with the reconstructed geometric model to generate a complete BIM model containing geometric information, attribute information, and topological relationships. The BIM model includes the geometric parameters, attribute parameters, topological relationships, etc. of the components.

[0025] S2. Construct and collect interactive data from users that includes spatial behavior commands and semantic commands. Combined with the reverse recognition and parsing mechanism of interactive intent, deconstruct and reassemble the components of the BIM model of the target building. Based on the reassembly results, generate a contextualized focused view containing component parameters. As a preferred implementation, the process of constructing and collecting interactive data containing spatial behavioral and semantic commands issued by users, combined with a reverse identification and parsing mechanism for interactive intent, to deconstruct and reassemble the components of the BIM model of the target building; and generating a contextualized focused view containing component parameters based on the reassembly results includes the following steps: S21. Based on the preset interaction method, construct a data interaction channel that includes a spatial gesture input channel and a language input channel; Specifically, the preset interaction methods include touch-based spatial gesture interaction and voice acquisition interaction. Touch-based spatial gesture interaction is compatible with visual terminals such as tablets and touch screen computers, and supports single / multi-finger touch, swipe, press, pinch and other actions, corresponding to BIM 3D interaction commands. The voice interaction method is near-field voice acquisition, which is compatible with the built-in microphone of the touch terminal.

[0026] The real-time acquisition of touch-based spatial gestures includes: acquiring touch data at a frame rate of 60fps through the terminal touch driver, such as the number of touch points, including single / double / multiple fingers, the two-dimensional coordinates (Xt, Yt) of each touch point, the touch pressure value, the touch duration, etc.; it also includes the coordinate sequence of continuous touch points, the length, direction, curvature, speed, etc.; the acquisition of voice input data uses the noise-canceling microphone built into the visual terminal, with a sampling rate of 16kHz and a bit depth of 16bit to acquire user voice.

[0027] S22. Through multimodal fusion analysis technology, and by using spatial gesture input and language input collected through the data interaction channel for semantic parsing, we identify and generate intent feature data containing interaction intent vectors. As a preferred embodiment, the step of using multimodal fusion analysis technology and performing semantic parsing on spatial gesture input and language input collected through the data interaction channel to identify and generate intent feature data containing interaction intent vectors includes the following steps: S221. Based on the preset architectural gesture semantic library, perform trajectory analysis on spatial gesture input data, and combine similarity analysis to identify gesture types and generate gesture semantic feature vectors. Specifically, the preset architectural gesture semantic library is a customized construction for BIM scenarios, as shown in Table 1. It includes the following three categories: (1) Gesture template library, which stores standard gesture templates adapted to touch interaction. Each template covers gesture name, trajectory features and corresponding BIM interaction semantics. For example, single-finger click corresponds to component positioning, two-finger pinch corresponds to view zooming, single-finger press and slide corresponds to area focusing, and two-finger diagonal slide corresponds to model sectioning, etc.; (2) Trajectory feature extraction rules, which specify the core parameters to be extracted, including the number of touch points, continuous coordinate sequence, touch pressure value change, trajectory length curvature and touch duration, etc.; (3) Similarity matching threshold, which presets the matching threshold for various gestures. Usually, the value is 0.8. If it is lower than this threshold, it is judged as an invalid gesture.

[0028] Table 1. Architectural Gesture Semantic Library

[0029] Specifically, when performing trajectory analysis on spatial gesture input data, it is necessary to extract continuous core feature parameters of the touch trajectory from the preprocessed touch-based spatial gesture data. This includes determining the number of touch points to distinguish between single-finger, two-finger, or multi-finger operations; extracting the temporal coordinate sequence of each touch point to form complete trajectory data; calculating the length curvature of the trajectory and the trend of touch pressure value changes at each point; and recording the total duration of the touch action and the pause time of key actions. All extracted feature parameters are standardized, uniformly normalizing the value range to between 0 and 1 to eliminate the differences in parameter magnitudes between different gesture actions, generating a low-dimensional gesture trajectory feature vector, typically 32-dimensional. Then, a cosine similarity algorithm is used to compare the generated gesture trajectory feature vector with the feature vectors of all standard gesture templates in the architectural gesture semantic library, calculating the similarity value between the two. The standard gesture template with the highest similarity value and not lower than the similarity matching threshold is selected, and the gesture name and interaction semantics corresponding to this template are used as the recognition result of the current touch gesture. Based on the recognition results, the corresponding semantic codes are retrieved from the architectural gesture semantic database. For example, a single-finger click is encoded as 1000, and a two-finger pinch is encoded as 0100. The semantic codes of the gesture trajectory feature vector and the recognition confidence score are fused, where the recognition confidence score is the highest similarity value obtained from the similarity analysis. After fusion, a high-dimensional gesture semantic feature vector is generated.

[0030] In addition, if the similarity value of all templates is below the threshold, it is marked as an invalid gesture and the prompting mechanism of the visualization terminal is triggered to remind the user to make a standard BIM interactive touch gesture.

[0031] S222. Based on natural language processing technology, perform language semantic parsing on language input data to obtain language semantic feature vectors; Specifically, when performing language semantic parsing on language input data based on natural language processing technology, a speech recognition model customized for the construction field is adopted. This speech recognition model is built on a converter architecture and has a built-in professional vocabulary in the construction field, covering professional terms such as beams, columns, shear walls, fire hydrants, HVAC pipelines, and accuracy levels.

[0032] Among them, the speech recognition model has been adapted for the architectural field in the input layer, encoder and output layer. The specific structure is as follows: (1) Input layer and feature extraction module, which is used to receive the preprocessed digital speech signal, the single channel audio with a sampling rate of 16kHz and a bit depth of 16bit, and convert the time domain speech signal into Mel spectrum features through short time Fourier transform. For example, the number of Mel filter banks is set to 80, the frame length is 25ms and the frame shift is 10ms, forming a two-dimensional feature matrix as the input of the speech recognition model. (2) Encoder module, which is used to stack 6 encoders, each layer containing a multi-head self-attention mechanism. For example, the number of heads is set to 8 and the dimension of the speech recognition model is 512. 1. Feedforward neural network with a hidden layer dimension of 2048, layer normalization and residual connection; (3) Decoder module, using a stack of 6 decoders, each layer containing a multi-head self-attention mechanism, encoder-decoder attention mechanism, and feedforward neural network, with the same dimension as the encoder; (4) Output layer and decoding module, used to map the decoder output to the vocabulary dimension of the building domain vocabulary using a fully connected layer, and output the probability distribution of each word through the softmax function; By using beam search decoding with rescoring of the language recognition model, the beam width is set to 4, and the decoding result is corrected twice by combining the customized language recognition model of the building domain, the accuracy of professional term recognition is further improved. At the same time, when the speech recognition model is trained, it collects the voice commands of different roles in the building industry, such as designers, construction workers and maintenance personnel, and then performs speech-to-text annotation, and then performs semantic entity annotation of the transcribed text in the building domain to form a training set. Finally, the speech recognition model is trained and adjusted through the training set to form a speech recognition model that meets the preset standards.

[0033] Therefore, the preprocessed original language data can be converted into Chinese text data through this semantic recognition model, and at the same time, it supports the regularization processing of colloquial instructions. For example, "Help me check the fire hydrant on the 3rd floor" is regularized to "Query the parameters of the fire hydrant on the 3rd floor". When outputting the text data, a speech recognition confidence score is generated, with a value range of 0 to 1, which is used to mark the validity of the text data. Then, the format of the output document is regularized, punctuation marks, modal particles, and invalid repeated words in the text are removed, and then spelling mistakes are corrected. Based on the professional vocabulary list in the construction field, the edit distance algorithm is used to correct the misrecognized words. Finally, based on the dictionary-based word segmentation algorithm and combined with the custom dictionary in the construction field, the text data is segmented into independent lexical units. For example, "Query the parameters of the fire hydrant in the east wing on the 3rd floor" is segmented into query, 3rd floor, east wing, fire hydrant, and parameter. After obtaining the lexical units, they are converted into low-dimensional word vectors, usually with a dimension of 64. At the same time, weights are assigned according to the importance of the words in the semantics. The weight of the operation entity is set to 0.4, the weight of the component entity is set to 0.3, and the weight of the spatial entity is set to 0.3. The weighted sum of all word vectors is calculated to generate a language semantic feature vector with a dimension of 128.

[0034] S223. Through the adaptive weight fusion mechanism, the gesture semantic feature vector and the language semantic feature vector are weighted and fused to generate intention feature data containing the interaction intention vector.

[0035] Specifically, when performing weighted fusion, the initial weights of the gesture modality and the language modality are preset to form an initial weight vector. For example, both are set to 0.5. The initial weights of the gesture modality and the language modality are combined in a fixed order to form a one-dimensional initial weight vector, which is pre-stored in the weight configuration library of the semantic rule engine. When starting the multi-modal fusion process each time, this vector is automatically retrieved from the configuration library as the initial reference value for weight adjustment. Then, the gesture recognition confidence and the language recognition confidence are extracted for dual-modal validity verification. If both dual-modal confidences meet the requirements of the valid threshold, the dynamic weights are calculated based on the confidence positive correlation formula, and then fine-tuned in combination with the characteristics of the BIM interaction scenario. For example, the weight of the gesture modality is increased in the spatial positioning scenario, and the weight of the language modality is increased in the precise query scenario, to obtain the optimal weight vector that takes into account both data quality and scenario requirements. Furthermore, if only a single modality of data is valid, the weight of that modality is directly adjusted to the dominant proportion, while the weight of the other modality is reduced to a low proportion threshold. Then, based on the optimal weight vector, a weighted summation operation is performed on the gesture semantic feature vector and the language semantic feature vector with the same dimension to generate a fusion intermediate vector. This intermediate vector is then concatenated with the dual-modal confidence scores to obtain a high-dimensional fusion feature vector. Finally, a lightweight fully connected neural network is used to reduce the dimensionality of the high-dimensional fusion feature vector and extract core features to generate an interaction intent vector containing key content such as operation intent, target information, and accuracy requirements. This is then combined with the optimal weight vector, dual-modal confidence scores, and original data association identifiers to encapsulate standardized intent feature data.

[0036] S23. Based on intent feature data and combined with a semantic rule engine, reverse deconstruct and reorganize the components of the BIM model, and generate a contextualized focused view that matches the current interaction intent and includes component parameters.

[0037] As a preferred implementation, the process of reverse deconstructing and reorganizing the components of the BIM model based on intent feature data and combined with a semantic rule engine, and generating a contextualized focused view that matches the current interaction intent and includes component parameters, includes the following steps: S231. Use a semantic rule engine to parse and process the intent feature data, and extract the intent type, association range and accuracy requirements respectively; In a preferred embodiment, the step of using a semantic rule engine to parse and process the intent feature data and extract the intent type, association range, and accuracy requirements includes the following steps: S2311. Standardize and format the intent feature data, and convert the standardized and formatted intent feature data into a structured data object containing identifiable fields for processing by the semantic rule engine. It should be noted that the standardized formatting process includes methods such as field regularization, data type unification, and outlier filtering. Based on the information processed by the standardized format, a fixed set of fields is defined for the structured data object. The specific fields are as follows: Basic information: including data ID, parsing timestamp, and total confidence score; Semantic features: including interaction intent vector; a list of semantic entities containing entity type, entity value, and confidence score; Modal information; weights of each modality and confidence scores of each modality; Structured data objects are constructed using the JSON-LD format, making the field semantics clear and directly indexable by the rule engine.

[0038] S2312. Match the structured data object with the preset intent classification rule base, combine the rule priority and feature confidence to make a decision, determine and output the intent type; In a preferred embodiment, the step of matching structured data objects with a preset intent classification rule base, combining rule priority and feature confidence to determine and output the intent type includes the following steps: S23121. Decompose the structured data object and extract key features for intent matching based on the decomposition results. Standardize and quantify the confidence data based on the key features. It should be noted that the structured data object is decomposed into a hierarchy of basic information layer, semantic feature layer, and modal information layer, and redundant intermediate calculation fields are removed. Key features strongly related to BIM interaction intent are extracted from the semantic feature layer, including operation features, target features, confidence features, and core dimension features of the interaction intent vector. Among them, operation features correspond to BIM operation commands such as query, section, and focus, and target features correspond to component entities such as beams, columns, and fire hydrants, or spatial entities such as the 3rd floor and the east wing. Then, the min-max normalization algorithm is used to quantize the confidence data of key features from different sources into the 0-1 range, eliminate the difference in units, and generate a key-value pair set of feature type, feature value, and quantized confidence.

[0039] S23122. Perform parallel matching of key features with rule conditions in the preset intent classification rule base, calculate the condition satisfaction of each rule condition, and combine the standardized and quantified confidence data to generate a matching score and rule trigger identifier for each rule. Specifically, the pre-defined architectural domain-customized intent classification rule library contains multiple rules adapted to BIM interaction scenarios. Each rule clearly defines the rule ID, rule conditions, corresponding intent type, rule priority, and trigger threshold. A multi-threaded parallel matching mechanism is used to compare the extracted key feature key-value pairs with the conditions of each rule in the rule library one by one. The condition satisfaction is calculated based on the completeness of feature matching. A complete match results in a condition satisfaction score of 1. A partial match results in a condition satisfaction score calculated as the proportion of the number of matched features to the total number of rule conditions. A complete mismatch results in a condition satisfaction score of 0. The condition satisfaction score is then multiplied by the average quantified confidence score of the key features to obtain the matching score score for each rule. Finally, the matching score score is compared with the rule trigger threshold. If the score is greater than or equal to the threshold, the rule trigger flag is set to 1; otherwise, it is set to 0, forming a complete matching result list containing the rule ID, matching score score, and trigger flag.

[0040] S23123. Based on the rule trigger identifier, a set of candidate rules with matching scores higher than a preset threshold is selected. Based on rule priority, matching score and feature confidence, the set of candidate rules is adjudicated through a preset multi-layer adjudication logic to determine the optimal intent classification rule.

[0041] It should be noted that, based on the rule trigger identifier, rules with an identifier of 1 and a matching score higher than a preset threshold are selected to form a candidate rule set. If the candidate set is empty, the matching is marked as failed and the process is returned to S23121 for reprocessing. If there are multiple candidate rules, the preset multi-level adjudication logic is activated. First, the rules are sorted from high to low priority, and the rule with the highest priority is retained. If there are multiple rules with the same priority, the second level of adjudication is entered, and the rules are sorted from high to low matching scores, and the rule with the highest score is retained. If there are still multiple rules with the same score, the third level of adjudication is entered, and the total confidence of the key features corresponding to these rules is calculated. The rule with the highest total confidence is retained. Finally, it is checked whether the matching score of the rule is higher than the strict threshold of 0.8. If it meets the threshold, it is determined as the optimal intent classification rule. If it does not meet the threshold, the trigger threshold range is expanded and the rules are re-selected and adjudicated.

[0042] S23124. Extract the intent type identifier defined by the optimal intent classification rule, and combine it with its matching score to generate an output containing the intent type.

[0043] It should be noted that a preset intent type identifier is extracted from the finally determined optimal intent classification rule. This identifier is related to the core intent of the BIM interaction scenario. Figure 1 One-to-one correspondence, such as component query corresponding to IT001 and region focus corresponding to IT002. Then, the intent type identifier, intent type name, optimal rule ID, matching score and key feature matching details are integrated and encapsulated in a standardized JSON format that the semantic rule engine can recognize to generate output data containing intent type.

[0044] S2313. Based on structured data objects and combined with the spatial index of the BIM model, calculate and extract the associated range; Specifically, a pre-defined hierarchical spatial index is retrieved from the BIM model database. This index is a multi-level index system built according to the characteristics of the building, divided into three levels from macro to micro: the first level index is the building zoning, such as the east wing, west wing, and underground parking garage; the second level index is the floor, such as the 1st floor, 3rd floor, and equipment floor; and the third level index is the component topology unit, such as beam unit, column unit, and pipeline cluster. Each index node is bound to the corresponding three-dimensional spatial coordinate boundary and component ID list. At the same time, structured data objects are received, and the core fields related to space are extracted, including spatial entities, target component entities, and corresponding confidence scores. Semantic parsing is performed on spatial entities in structured data to transform spatial information described in natural language into index nodes recognizable by the BIM spatial index. Based on the matched spatial index nodes, the three-dimensional spatial coordinate boundaries of the corresponding nodes are retrieved, and the minimum bounding box algorithm is used to calculate the minimum three-dimensional bounding box covering the target spatial area as the basic boundary of the association range. If the interaction intent includes topological association requirements, such as fire hydrants and connected pipelines, the association range is extended to the directly associated components of the target component by combining the component topological relationships of the BIM model, generating a composite association range of basic spatial boundary and topological association extension.

[0045] S2314. Based on the determined intent type, match the corresponding accuracy level from the preset BIM accuracy rule library to generate accuracy requirements.

[0046] Specifically, a pre-defined BIM accuracy rule library for the architectural field is built based on the LOD standard (LOD100-LOD500) of the architectural industry. The core content is the mapping relationship between intent type, accuracy level, and parameter granularity, while also binding terminal performance adaptation conditions. By precisely matching the intent type identifier with the intent types in the accuracy rule library, fully matching rule entries are filtered out. If multiple matching rules exist, the optimal matching rule is further selected based on the performance test results of the current visualization terminal. For example, a component query intent might match LOD300 on a mobile terminal and LOD400 on a workstation terminal. Based on the optimal matching rule, the corresponding accuracy level and parameter granularity are extracted and refined into executable accuracy requirement parameters: first, BIM model accuracy parameters; second, parameter display parameters; and third, rendering adaptation parameters. These refined accuracy requirements are encapsulated into a standardized data structure containing core fields: accuracy level (LOD value), parameter display list, rendering adaptation parameters, and terminal performance adaptation identifier. Simultaneously, the matching confidence of the accuracy requirements is marked as an accuracy filtering condition for selecting the initial component set.

[0047] S232. Based on the accuracy requirements, match the initial set of components corresponding to the accuracy requirements from the BIM model; It should be noted that, based on the accuracy tiered retrieval module of the BIM model, a preset LOD accuracy tiered component dataset is loaded. This dataset classifies and stores all components of the BIM model according to accuracy levels from LOD100 to LOD500. Each accuracy level corresponds to a different component detail granularity. Then, the accuracy requirement parameters are read, and the accuracy level is used as the core filtering condition to select components with an accuracy level equal to or higher than the required accuracy from all components in the BIM model. Simplified components with insufficient accuracy are removed. At the same time, the consistency between the accuracy attributes of the components and the requirements is verified, and finally, an initial set of components that meets the accuracy requirements is generated.

[0048] S233. Based on the association range, select a set of secondary components that match the association range from the initial component set, and determine the construction parameters that include geometric structure and spatial relationship; Specifically, based on the associated range parameters, which include a 3D spatial coordinate bounding box and a list of associated index nodes, with spatial coordinate boundaries as the primary filtering condition, the initial component set is spatially verified. It is determined whether the geometric center of each component is within the associated range boundary. Then, combined with the topological units corresponding to the associated index nodes, associated components with direct topological connections to the target component are added, forming a secondary component set. From this set, the core parameters of all components are extracted, including geometric structural parameters and spatial relationship parameters. Geometric structural parameters include dimensions, spatial coordinates, and cross-sectional shape; spatial relationship parameters include the location of connecting nodes between components, hierarchical affiliation, and topological connection density. These parameters are uniquely bound to the component ID to ensure that the parameter information of each component is complete and traceable, providing accurate data support for subsequent dynamic reorganization of components.

[0049] S234. Based on intent type, the corresponding view operation rules are called through the semantic rule engine to dynamically reorganize the components in the secondary component set to generate view state description data containing construction parameters. Specifically, based on intent type identifiers, the semantic rule engine automatically retrieves the corresponding rule entries from the preset view operation rule library. Different intent types match different component reorganization logic. For example, the component query intent matches the "component highlighting and associated component simplification" rule, and the model sectioning intent matches the "section component detail display and non-section component semi-transparency" rule. Then, using the secondary component set as the processing object, the engine performs dynamic component reorganization according to the matched view operation rules. During the reorganization process, the topological spatial relationship between components is strictly preserved to avoid component misalignment or connection breakage. At the same time, view state parameters are added, including the target view's viewing angle, scaling ratio, component display mode, parameter visualization method, etc. The reorganized component topology, component parameter set, and view state parameters are then integrated to generate standardized view state description data. This data is stored in a structured format and can be directly parsed by the 3D rendering engine.

[0050] S235. Based on the view state description data, and combined with the 3D rendering engine, perform real-time rendering to generate a contextualized focused view that matches the current interaction intent and includes construction parameters.

[0051] It should be noted that the 3D rendering engine reads the view state description data and parses the component topology and geometric parameters. A layered rendering strategy is used to load the secondary component set, prioritizing the rendering of core components and then overlaying related components. This ensures that the rendering efficiency is adapted to the terminal performance. At the same time, based on the parameter visualization configuration in the description data, the component parameters are bound to the BIM model, and different parameter display methods are configured. For example, geometric dimensions are marked on the component surface, and attribute parameters are set to hover display. The rendering perspective, lighting effects, and component display transparency are adjusted in combination with the view state parameters to generate a 3D visualization view in real time. The rendering results are then optimized to eliminate visual ambiguity caused by overlapping and occlusion of components, so that the view clearly presents the core information that matches the current interaction intent. Finally, a contextualized focused view containing complete component parameters is output for users to view and interact with.

[0052] S3. Based on the contextualized focused view containing component parameters and according to the preset intent identifier, construct an intent-view mapping table containing interactive intents and component parameters. In a preferred embodiment, the step of constructing an intent-view mapping table containing interactive intents and component parameters based on a contextualized focused view including component parameters and according to preset intent identifiers includes the following steps: S31. Extract component geometric parameters based on contextualized focused views, and calculate the topological feature vectors of components containing connection density and hierarchy depth through the BIM model topology analyzer. It should be noted that, from the generated contextualized focused view containing component parameters, the core geometric parameters of all target components are extracted in batches, including key data such as the component's three-dimensional spatial coordinates, dimensions, cross-sectional shape, and geometric contour accuracy. At the same time, the topology analyzer of the BIM model is called, using the component ID as the association index, to traverse the topological connection relationships of all components in the view, calculate the connection density and hierarchy depth between components, and then perform feature fusion and dimensionality reduction processing on the extracted geometric parameters and the calculated topological features to generate a topological structure feature vector of a unified dimension. This vector will serve as the core structural basis for subsequent intention and view association matching.

[0053] S32. Based on the negative sample learning mechanism, collect user records of rejection and correction operations on automatically generated views, and construct a reverse training sample set containing bias features to optimize the anti-interference ability of view generation. It should be noted that a negative sample collection mechanism can be established through the user interaction log system to record in real time the user's rejection and manual correction operations on the contextualized focused view automatically generated by the system. Rejection operations include directly closing the view or switching to other views, while manual correction operations include adjusting the display precision of components, hiding non-core components, and modifying the parameter visualization method. Furthermore, for each operation record, deviation features are extracted by comparing the differences between the automatically generated view and the view after user correction. These deviation features include deviations in component selection range, accuracy level adaptation, and parameter display granularity. These deviation features are then bound to the corresponding interaction intent identifiers and view IDs, and packaged into reverse training samples in a unified format. After a large amount of accumulation, a structured reverse training sample set is formed. The core function of this sample set is to provide negative reference for the optimization of mapping relationships, helping the system learn to identify and avoid view generation logic that users do not approve of, thereby improving the anti-interference ability and user adaptability of view generation.

[0054] S33. Based on topological feature vectors and user role information, and combined with BIM semantic knowledge graph, intention-parameter semantic reasoning is performed to generate an initial mapping relationship containing implicit parameters. It should be noted that the user role information is pre-stored, which covers dimensions such as the user's occupation type, common interaction intents, and preferred view display modes. Simultaneously, a pre-defined BIM semantic knowledge graph is loaded. This graph contains core content such as semantic association rules for building components, matching rules between interaction intents and component parameters, and view requirement rules corresponding to different roles. Then, based on semantic reasoning algorithms, the topological feature vector, user role information, and BIM semantic knowledge graph are deeply matched to infer the correspondence between interaction intent identifiers and contextualized focused views and component parameters. Hidden parameters are also mined, such as the building scenarios applicable to the view, the maintenance priority of components, and the display priority of parameters. This information is then integrated to generate an initial mapping relationship containing intent identifiers, view IDs, component parameter lists, topological feature vectors, and hidden parameters, laying the foundation for the construction of the mapping table.

[0055] S34. Construct a dynamic evolution index structure with self-optimization capability, and perform adversarial training between the initial mapping relationship and the reverse training sample set to generate mapping weights with anti-interference capability. It should be noted that the constructed self-optimizing dynamic evolution index structure adopts a hierarchical index design. The upper layer is the intent identifier index, the middle layer is the view ID index, and the lower layer is the component parameter and topological feature vector index. It supports incremental updates and fast retrieval. At the same time, the generated initial mapping relationship is used as a positive sample, and the constructed reverse training sample set is used as a negative sample. These are input into the adversarial training model. The adversarial training model adopts a variant architecture of generative adversarial network (GAN) and consists of two core modules: a generator and a discriminator. The generator generates candidate intent-view mapping weights based on the initial mapping relationship. The discriminator takes positive and negative samples as input and judges the effectiveness of the generator's output weights by comparison. Then, through multiple rounds of iterative training, the adversarial training model learns to distinguish between high-quality and low-quality mapping relationships. During the training process, the weight values ​​of different feature dimensions are dynamically adjusted. Among them, the matching degree weight of the topological structure feature vector and the user demand adaptation weight will be optimized in particular. Finally, a mapping weight with anti-interference ability is generated. This weight will serve as the core indicator for measuring the matching degree between intent and view.

[0056] S35. Obtain user feedback and adjust mapping weights in real time to form an intent-view mapping table that can adapt to scene changes and user preferences.

[0057] It should be noted that by establishing a multi-dimensional user feedback collection mechanism, user feedback information on the mapping table matching results can be obtained in real time, including view call frequency, view satisfaction rating, manual adjustment operation records, etc. Based on a preset weight adjustment algorithm, the mapping weights are dynamically optimized. For example, the weight value of frequently called intent-view mapping relationships is increased, while the weight value of mapping relationships that users have repeatedly rejected is decreased or even marked as invalid mappings. At the same time, when a new architectural scene or user preference change is detected, the incremental update process of the mapping table is automatically triggered to include the new intent-view association data into the mapping table. Through continuous feedback and weight adjustment, an intent-view mapping table that can adapt to changes in different architectural interaction scenarios and user personalized preferences is finally formed, enabling rapid and accurate matching of subsequent user interaction intents and contextualized focused views.

[0058] S4. Collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focused view data from the intent-view mapping table; Specifically, by establishing touch gesture input channels and voice input channels, user interaction data is collected in real time. Combined with multimodal fusion analysis technology, intent feature data containing interaction intent vectors is generated. Then, the semantic rule engine parses and extracts core information such as intent type and related scope to generate a unique intent identifier. The intent identifier matching mechanism is then activated. First, a precise match is performed in the intent-view mapping table. If a completely matching intent identifier exists, the corresponding contextualized focused view data is directly retrieved. If not, the closest view data is selected according to the mapping weight from high to low. At the same time, information such as the component parameters and topological feature vectors associated with the view are loaded. The entire process achieves fast retrieval through a dynamically evolving index structure, which significantly shortens the response time and improves the user interaction experience compared to regenerating the view.

[0059] S5. Based on contextualized focused view data, the system utilizes a preset 3D rendering engine for real-time reconstruction and presentation, and establishes a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

[0060] As a preferred implementation, the step of real-time reconstruction and presentation based on contextualized focused view data using a preset 3D rendering engine, and establishing a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display includes the following steps: S51. Based on the intent-view mapping table and combined with the topological features of the BIM model, a layered rendering strategy is adopted to reconstruct the view in real time and to visualize the component parameters. As a preferred embodiment, the real-time view reconstruction based on the intent-view mapping table, combined with the topological features of the BIM model, and the visualization processing of component parameters using a layered rendering strategy includes the following steps: S511. Based on the view state description data associated in the intent-view mapping table, extract the topological feature vector of the target component, and calculate the rendering priority based on the connection density and hierarchy depth in the topological feature vector and the interactive intent. Specifically, the view state description data bound to the current interaction intent is retrieved from the intent-view mapping table. The topological feature vector of the target component is extracted from it, focusing on the connection density representing the closeness of the component association and the hierarchical depth representing the importance of the component. The priority weight is quantified in combination with the interaction intent type. For example, the weight coefficient of the core component under the component query intent is set to 0.6, the connection density weight is set to 0.25, and the hierarchical depth weight is set to 0.15. The rendering priority score of each target component is calculated by weighted summation formula. The higher the score, the higher the rendering priority, so that the core component is rendered first.

[0061] S512. Based on rendering priority, construct a layered rendering queue, and dynamically allocate the rendering budget and level of detail for each layer according to the current performance of the visualization terminal to obtain the dynamic allocation result. Specifically, based on the calculated rendering priority score, the target components need to be sorted from high to low to construct a layered rendering queue with three levels: high, medium, and low. The high-priority queue corresponds to core functional components, such as target components in query intents and cross-sectional components in sectioning intents. The medium-priority queue corresponds to directly related components, such as connecting components of target components. The low-priority queue corresponds to indirectly related components, such as peripheral auxiliary components. At the same time, the terminal performance detection interface is called to obtain the core performance parameters of the current visualization terminal, such as computing power, memory, and video memory. The rendering budget (unit: milliseconds) and level of detail (LOD level) of each level are dynamically allocated according to the performance level. High-performance terminals can be allocated more rendering budget in high-priority queues and have their level of detail increased, while low-performance terminals can have their budget in low-priority queues appropriately compressed and their level of detail reduced, ultimately forming a dynamic allocation result adapted to the terminal performance.

[0062] S513. An asynchronous progressive rendering pipeline is used, and combined with dynamic allocation results, rendering tasks are performed on the target components according to the layered rendering queue. Specifically, the asynchronous progressive rendering pipeline employs a parallel processing and staged rendering architecture. While rendering high-priority components, it preprocesses data for medium and low-priority components in the background, preventing the rendering of a single component from blocking the overall process. Combined with dynamic allocation results, rendering tasks are executed sequentially according to the high, medium, and low-priority rendering queues. During the rendering process, the terminal performance load is monitored in real time. If the load is too high, such as CPU utilization exceeding 85%, the rendering strategy is automatically adjusted, temporarily reducing the level of detail of the current queue or pausing the rendering of low-priority queues to prioritize the rendering quality and speed of high-priority components. After the high-priority components are rendered and displayed stably, the rendering of medium and low-priority queues is gradually resumed to ensure a smooth and lag-free overall rendering process.

[0063] S514. For the rendered target component, automatically adapt and overlay a parameter visualization layer according to its component parameter type to generate a visualization output that integrates three-dimensional spatial morphology and semantic parameters.

[0064] Specifically, for the target component that has been rendered, its associated component parameters are first classified into two main categories: geometric parameters and attribute parameters. Then, the corresponding visualization display method is automatically adapted according to the parameter type. Geometric parameters are displayed by directly overlaying a 3D annotation layer on the component surface, such as annotating the length dimensions at both ends of a beam component. Attribute parameters are displayed by setting a hover information layer to trigger display, such as displaying material and maintenance records when the mouse hovers over a fire hydrant. Finally, the two types of visualization layers are precisely aligned and integrated with the 3D rendering, so that the parameter annotation position matches the component geometry and does not obscure the core visual information. The final result is a visualization output that combines 3D spatial form and complete semantic parameters, meeting the user's dual needs for viewing component form and obtaining parameters.

[0065] S52. Collect user interaction feedback data on the real-time view, including parameter adjustment, view modification, and operation rejection records, and construct a positive and negative feedback sample set containing operation intentions and deviation characteristics. Perform intention-feedback mapping through multimodal fusion analysis technology. It should be noted that the interaction log collection module based on the visual terminal captures all user operation feedback data on the rendered contextualized focused view in real time, such as modifying the display position of component parameters, switching the display format of parameter annotations; adjusting the view perspective, zoom ratio, component display precision, and hiding non-core components; directly closing the view, switching to other views, and repeatedly triggering the same intent to generate a new view. This feedback data is then categorized and labeled, with user-saved, frequently invoked, and unmodified view operations labeled as positive feedback, and user-rejected and repeatedly modified view operations labeled as negative feedback. Simultaneously, deviation features in the feedback data are extracted, including precision adaptation deviation, parameter display deviation, and spatial range deviation. Combined with multimodal fusion analysis technology, the user's operational intent is precisely bound to the feedback deviation features, generating structured data containing intent identifiers, view IDs, feedback types, deviation feature vectors, and confidence scores. Finally, this data is integrated to form a positive and negative feedback sample set that has both positive reference value and reverse optimization basis.

[0066] S53. Based on the positive and negative feedback sample set, the intention-view mapping table is dynamically weighted using an adversarial training mechanism and an adaptive optimization algorithm to achieve continuous evolution of the mapping table.

[0067] Specifically, by inputting the constructed positive and negative feedback sample sets into a pre-defined adversarial training model, positive feedback samples are used as positive references for training to strengthen the weight ratio of high-matching intent-view mapping relationships, while negative feedback samples are used as reverse constraints to weaken the weight of low-matching mapping relationships. Through multiple rounds of iterative training, the core features that distinguish between high-quality and low-quality mappings are learned. At the same time, an adaptive optimization algorithm is introduced to dynamically adjust the weight optimization strategy based on the frequency and timeliness of user feedback. For example, the weight increment is increased for mapping relationships with recent high-frequency positive feedback, while the weight of mapping relationships with multiple negative feedbacks is directly reduced or even marked as a mapping to be eliminated. The trained and optimized weight parameters are then synchronously updated to the dynamic evolution index structure of the intent-view mapping table to achieve real-time iteration of the mapping table weights. Through continuous feedback collection, training, and weight adjustment closed loop, the intent-view mapping table has the ability to adapt to changes in user preferences and updates in scene requirements, gradually improving the accuracy of subsequent intent matching and the effectiveness of view loading.

[0068] like Figure 2 As shown, according to an embodiment of the present invention, a three-dimensional visualization and interactive display system based on a BIM model is provided, the system comprising: Model building module 1 is used to acquire laser point cloud data of the target building and, in combination with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology. The Focus View Generation Module 2 is used to construct and collect interactive data from user-issued data containing spatial behavior commands and semantic commands. Combined with the reverse recognition and parsing mechanism of interactive intent, it deconstructs and reassembles the components of the BIM model of the target building; and generates a contextualized focus view containing component parameters based on the reassembly results. The mapping table construction module 3 is used to construct an intent-view mapping table containing interactive intents and component parameters based on the contextualized focused view containing component parameters and according to the preset intent identifiers. Data loading module 4 is used to collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focus view data from the intent-view mapping table; The visualization module 5 is used to reconstruct and present the data in real time using a preset 3D rendering engine based on contextualized focused view data, and to establish a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

[0069] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0070] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for three-dimensional visual interactive presentation based on a BIM model, characterized in that, Includes the following steps: S1. Acquire the laser point cloud data of the target building and, in conjunction with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology; S2. Construct and collect interactive data channels containing spatial behavior commands and semantic commands issued by users, and combine them with the reverse recognition and parsing mechanism of interactive intent to deconstruct and reassemble the components of the BIM model of the target building. Based on the reorganization results, a contextualized focused view containing component parameters is generated; S3. Based on the contextualized focused view containing component parameters and according to the preset intent identifier, construct an intent-view mapping table containing interactive intents and component parameters. S4. Collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focused view data from the intent-view mapping table; S5. Based on contextualized focused view data, the system utilizes a preset 3D rendering engine for real-time reconstruction and presentation, and establishes a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

2. The three-dimensional visualization and interactive display method based on a BIM model according to claim 1, characterized in that, The data interaction channel that constructs and collects interactive data containing spatial behavior commands and semantic commands issued by users, combined with the reverse recognition and parsing mechanism of interactive intent, deconstructs and reassembles the components of the BIM model of the target building. Based on the reorganization results, generating a contextualized focused view containing component parameters includes the following steps: S21. Based on the preset interaction method, construct a data interaction channel that includes a spatial gesture input channel and a language input channel; S22. Through multimodal fusion analysis technology, and by using spatial gesture input and language input collected through the data interaction channel for semantic parsing, we identify and generate intent feature data containing interaction intent vectors. S23. Based on intent feature data and combined with a semantic rule engine, reverse deconstruct and reorganize the components of the BIM model, and generate a contextualized focused view that matches the current interaction intent and includes component parameters.

3. The three-dimensional visualization and interactive display method based on a BIM model according to claim 2, characterized in that, The process of using multimodal fusion analysis technology and semantic parsing of spatial gesture input and language input collected through data interaction channels to identify and generate intent feature data containing interaction intent vectors includes the following steps: S221. Based on the preset architectural gesture semantic library, perform trajectory analysis on spatial gesture input data, and combine similarity analysis to identify gesture types and generate gesture semantic feature vectors. S222. Based on natural language processing technology, perform language semantic parsing on language input data to obtain language semantic feature vectors; S223. Through an adaptive weight fusion mechanism, the gesture semantic feature vector and the language semantic feature vector are weighted and fused to generate intent feature data containing the interaction intent vector.

4. The three-dimensional visualization and interactive display method based on a BIM model according to claim 3, characterized in that, The process of reverse deconstructing and reconstructing the components of the BIM model based on intent feature data and combined with a semantic rule engine to generate a contextualized focused view that matches the current interaction intent and includes component parameters includes the following steps: S231. Use a semantic rule engine to parse and process the intent feature data, and extract the intent type, association range and accuracy requirements respectively; S232. Based on the accuracy requirements, match the initial set of components corresponding to the accuracy requirements from the BIM model; S233. Based on the association range, select a set of secondary components that match the association range from the initial component set, and determine the construction parameters that include geometric structure and spatial relationship; S234. Based on intent type, the corresponding view operation rules are called through the semantic rule engine to dynamically reorganize the components in the secondary component set to generate view state description data containing construction parameters. S235. Based on the view state description data, and combined with the 3D rendering engine, perform real-time rendering to generate a contextualized focused view that matches the current interaction intent and includes construction parameters.

5. The three-dimensional visualization and interactive display method based on a BIM model according to claim 4, characterized in that, The process of parsing and processing intent feature data using a semantic rule engine, and extracting intent type, association range, and accuracy requirements, includes the following steps: S2311. Standardize and format the intent feature data, and convert the standardized and formatted intent feature data into a structured data object containing identifiable fields for processing by the semantic rule engine. S2312. Match the structured data object with the preset intent classification rule base, combine the rule priority and feature confidence to make a decision, determine and output the intent type; S2313. Based on structured data objects and combined with the spatial index of the BIM model, calculate and extract the associated range; S2314. Based on the determined intent type, match the corresponding accuracy level from the preset BIM accuracy rule library to generate accuracy requirements.

6. The three-dimensional visualization and interactive display method based on a BIM model according to claim 5, characterized in that, The process of matching structured data objects with a preset intent classification rule base, combining rule priority and feature confidence to determine and output the intent type includes the following steps: S23121. Decompose the structured data object and extract key features for intent matching based on the decomposition results. Standardize and quantify the confidence data based on the key features. S23122. Perform parallel matching of key features with rule conditions in the preset intent classification rule base, calculate the condition satisfaction of each rule condition, and combine the standardized and quantified confidence data to generate a matching score and rule trigger identifier for each rule. S23123. Based on the rule triggering identifier, a set of candidate rules with matching scores higher than a preset threshold is selected. Based on rule priority, matching score and feature confidence, the set of candidate rules is adjudicated through a preset multi-layer adjudication logic to determine the optimal intent classification rule. S23124. Extract the intent type identifier defined by the optimal intent classification rule, and combine it with its matching score to generate an output containing the intent type.

7. The three-dimensional visualization and interactive display method based on a BIM model according to claim 1, characterized in that, The step of constructing an intent-view mapping table containing interactive intents and component parameters based on a contextualized focused view containing component parameters and according to preset intent identifiers includes the following steps: S31. Extract component geometric parameters based on contextualized focused views, and calculate the topological feature vectors of components containing connection density and hierarchy depth through the BIM model topology analyzer. S32. Based on the negative sample learning mechanism, collect user records of rejection and correction operations on automatically generated views, and construct a reverse training sample set containing bias features to optimize the anti-interference ability of view generation. S33. Based on topological feature vectors and user role information, and combined with BIM semantic knowledge graph, intention-parameter semantic reasoning is performed to generate an initial mapping relationship containing implicit parameters. S34. Construct a dynamic evolution index structure with self-optimization capability, and perform adversarial training between the initial mapping relationship and the reverse training sample set to generate mapping weights with anti-interference capability. S35. Obtain user feedback and adjust mapping weights in real time to form an intent-view mapping table that can adapt to scene changes and user preferences.

8. The three-dimensional visualization and interactive display method based on a BIM model according to claim 1, characterized in that, The process of reconstructing and presenting the 3D visualization interactive display in real time using a preset 3D rendering engine based on contextualized focused view data, and establishing a closed-loop feedback and adaptive optimization mechanism, includes the following steps: S51. Based on the intent-view mapping table and combined with the topological features of the BIM model, a layered rendering strategy is adopted to reconstruct the view in real time and to visualize the component parameters. S52. Collect user interaction feedback data on the real-time view, including parameter adjustment, view modification, and operation rejection records, and construct a positive and negative feedback sample set containing operation intentions and deviation characteristics. Perform intention-feedback mapping through multimodal fusion analysis technology. S53. Based on the positive and negative feedback sample set, the intention-view mapping table is dynamically weighted using an adversarial training mechanism and an adaptive optimization algorithm to achieve continuous evolution of the mapping table.

9. The three-dimensional visualization and interactive display method based on a BIM model according to claim 8, characterized in that, The process of real-time view reconstruction based on the intent-view mapping table, combined with the topological features of the BIM model, and using a layered rendering strategy to visualize component parameters includes the following steps: S511. Based on the view state description data associated in the intent-view mapping table, extract the topological feature vector of the target component, and calculate the rendering priority based on the connection density and hierarchy depth in the topological feature vector and the interactive intent. S512. Based on rendering priority, construct a layered rendering queue, and dynamically allocate the rendering budget and level of detail for each layer according to the current performance of the visualization terminal to obtain the dynamic allocation result. S513. An asynchronous progressive rendering pipeline is used, and combined with dynamic allocation results, rendering tasks are performed on the target components according to the layered rendering queue. S514. For the rendered target component, automatically adapt and overlay a parameter visualization layer according to its component parameter type to generate a visualization output that integrates three-dimensional spatial morphology and semantic parameters.

10. A BIM model-based 3D visualization interactive display system, used to implement the BIM model-based 3D visualization interactive display method according to any one of claims 1-9, characterized in that, The system includes: The model building module is used to acquire laser point cloud data of the target building and, in combination with the design parameters of the target building, construct a BIM model of the target building using preset computer graphics technology. The focused view generation module is used to build and collect data interaction channels containing user-issued interactive data including spatial behavior commands and semantic commands. Combined with the reverse recognition and parsing mechanism of interactive intent, it deconstructs and reassembles the components of the BIM model of the target building; and generates a contextualized focused view containing component parameters based on the reassembly results. The mapping table construction module is used to construct an intent-view mapping table containing interactive intents and component parameters based on the contextualized focused view containing component parameters and according to the preset intent identifiers. The data loading module is used to collect users' real-time interaction intents and use the intent identifier matching mechanism to load the corresponding contextualized focus view data from the intent-view mapping table. The visualization module is used to reconstruct and present the data in real time using a preset 3D rendering engine based on contextualized focused view data, and to establish a closed-loop feedback and adaptive optimization mechanism to achieve the evolution and optimization of 3D visualization interactive display.

Images