Design method for generating concrete interactive reading interface by converting plane information into three dimensions

Abstract

The application discloses a design method for converting plane information into a concrete interactive reading interface, and particularly relates to the field of human-computer interaction and geographic information presentation, and is used for solving the problem of information fragmentation and rigid interactive experience caused by the lack of direct mapping between planar text and spatial features. The method is characterized in that: text and images are bound in the same coordinate space, core entities that can be three-dimensionally depicted are selected by a representative decision maker, geometric scales and subdivision levels are planned by a representative coefficient, semantics and morphology are synchronously presented in the same view, a focus of attention is instantaneously converted into interpretation through line-of-sight interaction, a browsing track is reversely corrected to modify a model density and a marker position, an adaptive closed loop is formed by evolution with use, multi-source evidence is aggregated in the representative coefficient to avoid neglecting spatial reproducibility by only relying on heat to make a selection, and the aggregation promotes the combination of content selection and three-dimensional expression, and the reading process is converted from plane interpretation into stereoscopic perception, and topography and building features are directly and externally shown, so that the overall immersion and consistency are improved.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction and geographic information presentation, and more specifically, to a design method for transforming planar information into a three-dimensional representation of a concrete interactive reading interface. Background Technology

[0002] In the fields of human-computer interaction and geographic information presentation, current digital platforms generally use flat text or simple vector annotations to convey city and place name information: map sidebars display single-line introductions, web encyclopedias list administrative history, population, and economic data by chapter, and travel guides rely on paragraph-style narration and static illustrations. When readers browse two-dimensional characters on the screen, they need to imagine the terrain differences, street textures, and landmark facades in their minds, and there is a lack of direct mapping between text and spatial features; if they want to obtain a three-dimensional sense, they usually need to jump to a third-party 3D globe or architectural model platform, resulting in a significant sense of temporal and spatial discontinuity.

[0003] The existing two-dimensional information presentation mode exposes three key shortcomings: First, text, images, and coordinate data are isolated from each other, lacking real-time linkage from semantics to geometry, resulting in the phenomenon of "reading words but not shapes," making it difficult to capture the undulations of the city skyline or the layers of landforms in the instant of reading; second, the flat interface cannot respond to changes in the user's gaze and focus, lacks immersive feedback, and the interaction remains at the level of clicking and turning pages, resulting in a rigid overall experience; third, any attempt to showcase architectural features or landform textures relies on manual modeling or pre-rendered videos, which are costly to produce and maintain and have slow updates, causing the urban form to be disconnected from the textual description for a long time.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a design method for generating a concrete interactive reading interface through the three-dimensional transformation of planar information. This method binds text and images within the same coordinate space, uses a representative decision-maker to select core entities that can be depicted in three dimensions, and coordinates geometric scale and subdivision levels using a representativeness coefficient, allowing semantics and form to be presented synchronously within the same visual field. Eye-tracking interaction instantly transforms focus into interpretation, and the browsing trajectory reversely corrects model density and marker positions, forming an adaptive closed loop that evolves with usage. Multi-source evidence is aggregated in the representativeness coefficient, avoiding selection based solely on popularity while ignoring spatial reproducibility, thus promoting a fit between content selection and three-dimensional expression. The reading process transforms from planar interpretation to three-dimensional perception, with terrain and architectural features becoming intuitively apparent, improving overall immersion and consistency, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] S1: Receive the region name text and satellite imagery, complete semantic segmentation and spatial registration, generate a text pixel mapping table and write it into the data pool;

[0008] S2: Candidate entities are formed based on the mapping table. The representative decision-maker calculates the main axis fit rate and the piecewise reproducibility, outputs the representativeness coefficient, and selects entities to write into the feature library.

[0009] S3: Read the feature library and representativeness coefficients, set geometric parameters for the selected entities to generate a 3D model, and bind the corresponding text tags;

[0010] S4: Loads a 3D model and binds eye tracking. Text descriptions are displayed synchronously when the user looks at the target. Rotation, scaling and transparency adjustment are supported.

[0011] S5: Collect browsing trajectory and compare it with text markers and representativeness coefficients. When occlusion or abnormal stops occur, adjust the model and marker positions, write back to the feature library and iteratively optimize.

[0012] In a preferred embodiment, step S1: receive the region name text and satellite image, extract the place name set and descriptive phrase through word segmentation and named entity recognition, use geocoding to query the geographic coordinates of the place name center and convert them into image pixel coordinates through coordinate transformation, extract a set of pixel coordinates with a fixed radius centered on the pixel coordinates, generate a text pixel mapping table containing place name identifiers, pixel coordinate sets, descriptive phrases and source indexes, and then write it into the original data pool.

[0013] In a preferred embodiment, step S2: read the text pixel mapping table from the original data pool to generate a candidate entity set, extract the boundary pixel coordinates through edge detection and region segmentation algorithms, calculate the principal axis fit rate and the polyline reproducibility, construct a small sample positive and negative pair set, input the principal axis fit rate and the polyline reproducibility, generate a discriminative embedding vector through pairwise comparison learning, and calculate the representativeness coefficient based on the boundary margin maximization criterion; set an entry threshold to determine the selected entities, sort them in descending order of representativeness coefficient, and write the selected entities, along with the boundary pixel coordinates, descriptive phrases, and entity numbers, into the feature library.

[0014] In a preferred embodiment, for each candidate entity, the morphological principal axis description is extracted from the descriptive phrase and transformed into the expected principal axis direction and expected curvature direction. The skeleton is extracted and refined based on the boundary pixel coordinates to calculate the total arc length of the skeleton. The angular deviation between the local tangential direction and the expected principal axis direction is calculated point by point along the skeleton to determine the principal axis fitting rate. The Douglas-Puk algorithm is used to simplify the boundary with a fixed error tolerance to obtain the minimum number of line segments. The polyline reproducibility is calculated by linear normalization through empirical upper and lower bounds.

[0015] In a preferred embodiment, step S3: The modeling module is invoked to read the selected entities and related data from the feature library, and it is determined whether to use stretching modeling or surface modeling based on the descriptive phrase. For stretching modeling, the stretching height is set based on the boundary pixel coordinates. For surface modeling, a non-uniform rational B-spline surface is fitted and the number of control points is set. The subdivision level and texture resolution are allocated according to the representativeness coefficient to generate a 3D model containing geometric data and texture data. The text markers extracted from the descriptive phrase are bound to the geometric center of the model and the source index and entity number are embedded to provide visualization data support for interactive rendering.

[0016] In a preferred embodiment, step S4: load the 3D model to the rendering interface through the graphics rendering engine, bind the eye tracking device to capture the coordinates of the user's gaze point, and display text labels on the target facade when the threshold is exceeded, and respond to rotation, scaling and transparency adjustments to ensure viewpoint synchronization.

[0017] In a preferred embodiment, step S5 involves: collecting user browsing trajectory data from interaction logs, including gaze coordinates, gaze duration, operation type, and entity number; detecting occlusion or abnormal dwell time through ray casting; adjusting the height of the 3D model or the number of surface control points based on the number of occlusions and the duration of abnormal dwell time; updating the text marker position to the non-occluded area offset from the gaze point by a fixed distance; writing the representativeness coefficient of the interaction evidence and the entity number into the memory sample; triggering small sample comparison learning to rank and update the feature library; and completing the screening and modeling iteration.

[0018] The technical effects and advantages of the design method for generating a concrete interactive reading interface by transforming planar information into three dimensions according to the present invention are as follows:

[0019] This invention binds text and images in the same coordinate space, uses a representative decision-maker to select core entities that can be described in three dimensions, and uses a representative coefficient to coordinate geometric scale and subdivision level, so that semantics and form are presented synchronously in the same field of view; eye-tracking interaction instantly transforms the focus into interpretation, and the browsing trajectory reversely corrects the model density and marker position, forming an adaptive closed loop that evolves with use; multi-source evidence is aggregated in the representative coefficient, avoiding selection based solely on popularity while ignoring spatial reproducibility, thus promoting the fit between content selection and three-dimensional expression; the reading process is transformed from planar interpretation to three-dimensional perception, with terrain and architectural features becoming intuitively visible, while maintenance costs and update cycles are shortened, and the overall immersion and consistency are improved. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a design method for generating a concrete interactive reading interface from planar information in three dimensions, according to the present invention.

[0021] Figure 2 This is a perspective view of the effect obtained by the design method of generating a concrete interactive reading interface from planar information in three dimensions according to the present invention;

[0022] Figure 3 This is a top view of the design method for generating a concrete interactive reading interface by transforming planar information into three dimensions, according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1: Figure 1 This invention presents a design method for generating a concrete interactive reading interface from planar information in three dimensions, comprising:

[0025] S1: Receives region name text and satellite imagery, performs semantic segmentation and spatial registration, generates a text-pixel mapping table, and writes it to the data pool.

[0026] S2: Candidate entities are formed based on the mapping table. The representative decision-maker calculates the main axis fit rate and the piecewise reproducibility, outputs the representativeness coefficient, and selects entities to write into the feature library.

[0027] S3: Read the feature library and representativeness coefficients, set geometric parameters for the selected entities to generate a 3D model, and bind the corresponding text tags.

[0028] S4: Loads 3D models and binds eye tracking. Text descriptions are displayed synchronously when the user looks at the target. It supports rotation, scaling, and transparency adjustment.

[0029] S5: Collect browsing trajectory and compare it with text markers and representativeness coefficients. When occlusion or abnormal stops occur, adjust the model and marker positions, write back to the feature library and iteratively optimize.

[0030] This invention presents a design method for generating a concrete interactive reading interface from planar information into three dimensions, addressing the shortcomings of existing digital platforms in presenting city and place name information. Traditional two-dimensional interfaces, such as map sidebars, web encyclopedias, or travel guides, rely on planar text, static images, or simple vector annotations to convey information. Readers need to imagine terrain differences, street patterns, or building facades themselves. The lack of direct connection between text and spatial features leads to low information comprehension efficiency and an inability to intuitively experience geographical and cultural atmosphere. Furthermore, in existing methods, text, images, and coordinate data are isolated, making it difficult to achieve real-time semantic and geometric linkage. Additionally, 3D presentation typically relies on costly manual modeling or pre-rendered videos, resulting in delayed updates and difficult maintenance.

[0031] This invention integrates regional name text with satellite imagery to generate a three-dimensional figurative model. By combining eye tracking and user interaction feedback, it constructs a dynamic and immersive reading interface, which intuitively presents the terrain and architectural features, improves information absorption efficiency and interaction fluency, and reduces maintenance costs.

[0032] Step S1, as the starting point of this process, is responsible for receiving the region name text and satellite imagery, establishing a precise mapping relationship between the text and image coordinates, generating a text-pixel mapping table and writing it into the raw data pool, providing a unified data entry point for entity screening, 3D modeling and interactive rendering.

[0033] Step S1 is the primary step in data initialization, aiming to structurally associate the user-input region name text with satellite imagery to form a text-pixel mapping table. The region name text includes the place name and its description, such as the name of a city, street, or building, and its topographical or architectural features; the satellite imagery provides high-resolution geographic information, covering terrain and building outlines. Step S1 establishes a mapping relationship between place names and image pixel coordinates, ensuring precise binding of semantic and spatial information, and generating a data structure that can be directly used in subsequent steps.

[0034] First, the system receives user-input region name text and satellite imagery. The region name text includes the place name and its description, such as "Beijing Palace Museum, Forbidden City, red walls and yellow tiles, symmetrical layout along a central axis," stored as a text set in natural language. The satellite imagery is a high-resolution two-dimensional pixel matrix containing geographic coordinate information (latitude and longitude) and color values. In the preprocessing stage, the text set is segmented and named entity recognition is performed to extract place name entities and related descriptive phrases (such as "red walls and yellow tiles"), forming a place name set. The satellite imagery undergoes geometric correction and coordinate system standardization to ensure a one-to-one correspondence between the geographic coordinates and pixel coordinates of each pixel, providing an accurate spatial reference for mapping.

[0035] For each place name in the set, its central geographic coordinates are retrieved using geocoding. Utilizing coordinate transformation parameters from satellite imagery, the geographic coordinates of the place names are converted into pixel coordinates within the imagery. This is achieved through linear transformation calculations, ensuring that the place name location precisely corresponds to a pixel in the imagery. Using the converted pixel coordinates as the center, an area with a fixed radius (e.g., 100 pixels) is extracted from the satellite imagery, forming the spatial extent corresponding to each place name. This is denoted as the pixel coordinate set and used to represent the spatial coverage of the place name within the imagery.

[0036] For each place name in the set, a record is generated, forming a text-to-pixel mapping table. Each record contains the following fields: place name identifier, pixel coordinate set (the image area corresponding to the place name), descriptive phrase (the text description associated with the place name), and source index (a number pointing to the original text and image). The generation process iterates through the place name set, recording the pixel coordinate set, associated descriptive phrase, and source index for each place name in turn, forming a structured text-to-pixel mapping table. If a geocoding query for a place name fails or the pixel coordinate set is empty, it is marked as an invalid record and logged.

[0037] Step S1 uses geocoding and coordinate transformation to precisely bind the place names in the text with the pixel locations in the satellite imagery, generating a text-pixel mapping table to ensure the consistency of semantic and spatial information. The processing involves using the place name as the semantic core, obtaining its spatial location through geocoding, expanding the spatial range with a set of pixel coordinates, and combining it with descriptive phrases to form a structured record, providing a traceable and unified data entry point for subsequent analysis and processing. This method avoids the problem of text and image separation in traditional two-dimensional presentations, and supports efficient entity filtering and 3D modeling through standardized coordinate mapping and data organization.

[0038] Step S1 establishes a text-to-pixel mapping table by receiving the region name text and satellite imagery, providing a unified data entry point for binding semantic and spatial information. However, to achieve the transformation from planar information to a 3D model, the coordinate mapping of text and imagery alone is insufficient to select core entities suitable for 3D modeling. It is necessary to further extract representative entities from the mapping table and evaluate their semantic and geometric representationability.

[0039] Step S2, based on the text-pixel mapping table in the original data pool generated in Step S1, extracts and filters the candidate entity set to prepare high-quality input data for 3D modeling. The text-pixel mapping table contains place name identifiers, pixel coordinate sets, descriptive phrases, and source indexes, providing image regions and semantic information corresponding to place names for Step S2. The task of Step S2 is to calculate the principal axis fit rate and polyline reproducibility using a representative decision maker, combine small-sample comparative learning ranking, generate representativeness coefficients and selection criteria, and store the filtered entities and their representativeness coefficients in the feature library, providing accurate geometric and semantic basis for 3D modeling in Step S3.

[0040] Candidate entity set generation: The text pixel mapping table, containing place name identifiers, pixel coordinate sets, descriptive phrases, and source indexes, is read from the raw data pool. For each place name identifier, the corresponding image region is extracted from the satellite imagery based on its pixel coordinate set. Independent entities within the image region, such as buildings, rivers, or roads, are identified using edge detection and region segmentation algorithms, forming a candidate entity set. Each candidate entity records its boundary pixel coordinates (a subset of the pixel coordinate set) and associated descriptive phrase, and is assigned a unique entity number.

[0041] Principal axis fit calculation: For each candidate entity, the morphological principal axis description is extracted from its associated descriptive phrase, such as river-direction or central axis symmetry, and converted into the expected principal axis direction (expressed in angle) and the expected curvature direction (a dimensionless quantity describing the curvature change trend). In satellite imagery, based on the boundary pixel coordinates of the candidate entity, a skeletonization algorithm (such as Zhang's thinning algorithm) is applied to extract the thinned skeleton, and the total arc length of the skeleton is calculated. The local tangential direction (expressed in angle) is calculated point-by-point along the skeleton at fixed intervals, and compared with the expected principal axis direction to calculate the angular deviation.

[0042] The spindle fit rate is defined as the proportion of the skeleton arc length with an angular deviation less than the limit tolerance (e.g., 30 degrees) to the total arc length. It ranges from 0 to 1, is dimensionless, and represents the degree of similarity between the descriptive phrase and the entity shape.

[0043] Polyline reproducibility calculation: For each candidate entity, a polyline approximation is performed in the satellite imagery based on its boundary pixel coordinates. The Douglas-Peucker algorithm is used to simplify the boundaries with a fixed error tolerance (e.g., 2 pixels, matching the rendering resolution) to obtain the minimum number of line segments. The number of line segments is mapped to the polyline reproducibility, and linear normalization is performed using empirical upper and lower bounds (e.g., minimum 5, maximum 50): when the number of line segments is equal to or less than the minimum value, the score is 1; when it is equal to or more than the maximum value, the score is 0; when it is between the two, the score is the maximum value minus the number of line segments divided by the difference between the maximum and minimum values. The polyline reproducibility value ranges from 0 to 1, is dimensionless, and represents the geometric compressibility of the entity contour. A higher value indicates that 3D reconstruction is easier to achieve.

[0044] Small-sample contrastive learning ranking: Construct a small set of positive and negative pairs, containing 10 to 20 pairs of positive samples (entities suitable for modeling) and negative samples (entities unsuitable for modeling). Each pair includes principal axis fit rate, piecewise reproducibility, and source index. Generate discriminative embedding vectors through pairwise contrastive learning. Calculate the representativeness coefficient based on the boundary margin maximization criterion; the value ranges from 0 to 1, is dimensionless, and represents the overall representativeness of an entity's suitability for 3D modeling. Set an inclusion threshold (e.g., 0.7); when the representativeness coefficient is greater than or equal to the threshold, the candidate entity is selected.

[0045] For example, the calculation process can be as follows:

[0046] Small-sample comparison learning and ranking to construct a small-sample positive-negative pair set ,in These are positive samples (entities known to be suitable for modeling). Negative samples (entities unsuitable for modeling) should be used, with a sample size of 10 to 20 pairs. Input principal axis fit rate. Polyline reproducibility and source index Discriminative embedding vectors are generated through pairwise contrastive learning. The representativeness coefficient is calculated using the boundary margin maximization criterion. :

[0047]

[0048] in The learned discriminant vector, and The parameters are scalars (initial values ​​are 1 and 0 respectively), and optimization is achieved through iterative methods. The value ranges from 0 to 1, is dimensionless, and represents the comprehensive representativeness of the entity suitable for 3D modeling. Based on Determine the selection threshold (e.g., 0.7), if ,but For selected entities.

[0049] Feature library storage: Sort the selected entities in descending order of representativeness coefficient to generate a feature library. Record the boundary pixel coordinates, descriptive phrase, representativeness coefficient and entity number of each selected entity and store them as a structured table.

[0050] Step S2's processing logic generates a candidate entity set through edge detection and region segmentation, calculates the principal axis fit rate to quantify the degree of alignment between semantic description and entity shape, calculates the polyline reproducibility to assess the geometric compressibility of the contour, and integrates the two parameters using small-sample comparative learning to generate a representative coefficient, thus filtering entities suitable for 3D modeling. This approach, centered on complementary semantic and geometric evaluation, ensures semantic fit through the principal axis fit rate, guarantees modeling feasibility through polyline reproducibility, and uses the representative coefficient to drive filtering and ranking, providing high-quality input data for 3D modeling.

[0051] Step S2 generates a candidate entity set based on the text-pixel mapping table, selects suitable entities for 3D modeling, and stores them in the feature library. However, to achieve the transformation from planar information to a 3D concrete model, the selected entities need to be converted into 3D models, ensuring that the geometric shape of the model is consistent with the semantic description, while optimizing the allocation of rendering resources.

[0052] Step S3 utilizes the feature library generated in Step S2 to read the selected entities, boundary pixel coordinates, descriptive phrases, representativeness coefficients, and entity numbers, and performs a 3D modeling task. The feature library provides filtered entity data, containing semantic and geometric information, and the representativeness coefficient reflects the modeling priority of the entities. The task of Step S3 is to set the extrusion height or surface parameters for each selected entity through the modeling module, allocate subdivision levels and texture resolutions according to the representativeness coefficient, generate a 3D model consistent with the entity shape, and bind text tags and source indexes inside the model to provide visualization data for the interactive rendering in Step S4.

[0053] Set the stretching height or surface parameters: For each selected entity, determine the modeling type based on the semantic information in its descriptive phrase. If the descriptive phrase contains height-related terms, such as "high-rise building," use stretching modeling; if it contains curvature-related terms, such as "winding river," use surface modeling. For stretching modeling, extract the entity boundary from the boundary pixel coordinates, calculate the area of ​​the bounding rectangle (in pixels squared), and set the stretching height to the base height (e.g., 50 meters) multiplied by (1 plus the height adjustment coefficient multiplied by the representativeness coefficient), where the height adjustment coefficient is a dimensionless constant (e.g., 0.5). For surface modeling, fit a non-uniform rational B-spline surface based on the boundary pixel coordinates. The number of surface control points is the base control point number (e.g., 10) plus the representativeness coefficient multiplied by the difference between the upper and lower bounds of the control point number (e.g., 100 minus 10) and the adjustment coefficient (e.g., 0.8), and take the integer part.

[0054] Assigning subdivision levels and texture resolution: Subdivision levels and texture resolutions are assigned to selected entities based on a representativeness coefficient. The subdivision level controls the geometric details of the 3D model and is calculated as the base subdivision level (e.g., 2), plus the representativeness coefficient multiplied by the difference between the upper and lower bounds of the subdivision level (e.g., 10 minus 2), and an adjustment coefficient (e.g., 1.0), taking the integer part.

[0055] Texture resolution controls the surface detail of the model and is calculated by multiplying the base resolution (e.g., 256 pixels / meter) by (1 plus the resolution adjustment factor multiplied by the representativeness factor), where the resolution adjustment factor is a dimensionless constant (e.g., 0.5). Entities with high representativeness factors receive higher subdivision levels and texture resolution, ensuring the visual representation of important entities.

[0056] Generating 3D Models: For extrusion modeling, the boundary pixel coordinates are converted into a 3D mesh, which is then extruded vertically to a set height to generate the 3D geometry. For surface modeling, a non-uniform rational B-spline surface is used to fit the boundary pixel coordinates to generate a smooth surface model. The geometry is then subdivided into meshes using a subdivision level. Texture maps are extracted from satellite imagery and applied with texture resolution to generate the 3D model. The 3D model is stored in a standard format, containing both geometric and texture data.

[0057] Binding Text Tags and Source Indexes: Text tags and source indexes are bound within the 3D model. Text tags extract key descriptions from descriptive phrases, such as "red walls and yellow tiles," are located at the model's geometric center, and stored as annotation point coordinates (in meters). Source indexes and entity numbers are embedded together in the model's metadata, ensuring the model can be traced back to the text-pixel mapping table in the original data pool.

[0058] Step S3 determines the modeling type through semantic analysis, generates stretched or curved models based on boundary pixel coordinates, dynamically allocates subdivision levels and texture resolution according to representativeness coefficients, generates a 3D model, and binds text tags and source indexes. It optimizes modeling resource allocation with representativeness coefficients as the core, ensures geometric consistency with semantics by controlling the number of points through stretching height or surface control, and balances visual expressiveness and computational efficiency through subdivision levels and texture resolution, providing high-quality 3D models for interactive rendering.

[0059] Step S3 generates a 3D model and binds text markers and a source index. To achieve immersive interactive reading, the 3D model needs to be loaded into the rendering interface, responding to user gaze and actions, and enabling synchronous updates of text descriptions and the model. Step S4 loads the 3D model into the rendering interface, binds a gaze tracker, and synchronously presents text descriptions when the user gazes at the target, responding to rotation, scaling, and transparency adjustments, providing dynamic presentation support for subsequent interactive optimization.

[0060] Step S4 utilizes the 3D model generated in Step S3 to read the 3D model, bound text markers, source index, and entity number, and performs rendering and interaction tasks. The 3D model contains geometric and texture data, the text markers provide semantic descriptions, and the source index ensures traceability to the text pixel mapping table. The task of Step S4 is to load the 3D model into the rendering interface, bind a gaze tracker, respond to user gaze and operations, realize the synchronous presentation of text descriptions and model adjustments, and provide interactive data for the user browsing trajectory collection in Step S5.

[0061] The 3D model, including geometric data, texture data, text markers, and source index, is read from the storage in step S3 and imported into the rendering interface via the graphics rendering engine. The rendering interface is initialized as a 3D viewport with a fixed top-down angle (e.g., 45 degrees), and the model coordinate system is aligned with the boundary pixel coordinates of the satellite image. During loading, entities with high representativeness are rendered first to ensure the visual prominence of important entities.

[0062] An integrated gaze tracking device is attached to the rendering interface to capture the user's gaze coordinates in real time (in pixels at the rendering interface resolution). A gaze time threshold is set for each 3D model. When the user's gaze time exceeds the threshold and the gaze point is projected onto the model's target facade (calculated by ray projection from the gaze point to the camera position), synchronized text descriptions are triggered.

[0063] When the gaze tracker detects that a user is looking at a target facade, descriptive text is extracted from the text markers bound to the 3D model and overlaid on the rendering interface. The text description is positioned at a fixed distance offset from the gaze point coordinates to avoid occluding the model, and a semi-transparent overlay (initial transparency is 0.8) is used. The text content is directly derived from descriptive phrases, and a source index is associated to support data traceability.

[0064] Responding to user actions: Rotation updates the model's view angle via mouse dragging or gestures; scaling adjusts the model's proportions via the scroll wheel; transparency adjustment is set via keyboard shortcuts to the initial transparency multiplied by one minus the product of the adjustment coefficient and the representative coefficient, where the adjustment coefficient is a dimensionless constant. All operations update the rendering interface in real time, ensuring that the text descriptions are synchronized with the model's viewpoint.

[0065] An event listener mechanism monitors the gaze tracker and user input, updating the rendered interface at a fixed frame rate per second. Synchronous updates utilize a buffer mechanism, first calculating the text description position and model transformation under the new viewpoint, then rendering uniformly. Interaction logs are recorded, including gaze duration and operation type, associated with entity numbers, and stored as a temporary data structure for use in step S5.

[0066] Step S4 loads the 3D model into the rendering interface and binds a gaze tracker to synchronize text descriptions with the user's gaze, responding to rotation, scaling, and transparency adjustments to ensure consistency between the descriptions and the model's perspective. This step addresses the lack of immersive feedback in flat interfaces, improving information absorption efficiency and user experience through dynamic interaction, and providing real-time data support for closed-loop optimization of the browsing trajectory.

[0067] Step S4 loads the 3D model into the rendering interface and binds a gaze tracker to achieve synchronized presentation of text descriptions and user interaction responses. However, to achieve adaptive closed-loop optimization, it is necessary to continuously collect user interaction data, analyze browsing trajectories to detect problems, and adjust the model and marker positions.

[0068] Step S5 utilizes the rendered interface and interaction logs generated in Step S4 to collect user browsing trajectories, analyze the 3D model, text markers, representativeness coefficients, entity numbers, and source indexes, and achieve adaptive optimization. The interaction logs record the user's gaze coordinates, gaze time, and operation type, providing data for trajectory analysis. The task of Step S5 is to continuously collect user browsing trajectories, detect occlusion or abnormal dwell times, trigger fine-tuning of the 3D model height and text marker positions, write interaction evidence into the memory samples of the representative decision-maker, complete the iteration of screening and modeling, and enhance the dynamic adaptability of the interactive interface.

[0069] Collect user browsing trajectory: Read the user browsing trajectory from the interaction log in step S4, including gaze coordinates (in pixels of the rendered interface resolution), gaze duration, operation type (such as rotation, zoom), and entity number. The browsing trajectory is stored as a time series, collected at a fixed frame rate per second to ensure high-frequency data capture. The time series records the gaze coordinates, gaze duration, operation type, and associated entity number for each trajectory point.

[0070] Occlusion and Abnormal Dwelling Detection: For each trajectory point, the intersection point with the 3D model is calculated from the gaze point coordinates and camera position using ray projection. If the intersection point is empty or located on a non-target facade, it is considered an occlusion event, and the number of occlusions is recorded. Abnormal dwelling is defined as gaze duration exceeding a fixed threshold without any operation. The duration of abnormal dwelling is calculated as the sum of gaze durations exceeding the threshold, and an entity number is associated to identify the problematic entity.

[0071] Triggering Model Height Fine-Tuning: If the number of occlusions exceeds a fixed threshold or the duration of abnormal dwell times exceeds a fixed threshold, fine-tuning of the 3D model height is triggered. For stretched models, the adjusted height is calculated as the product of the original height multiplied by one plus a height adjustment coefficient, the sum of the number of occlusions and the duration of abnormal dwell times divided by its threshold, and then divided by the sum of the occlusion threshold plus one, where the height adjustment coefficient is a dimensionless constant. For curved models, the adjusted number of control points is calculated as the original number of control points plus the product of the control point adjustment coefficient, the sum of the number of occlusions and the duration of abnormal dwell times divided by its threshold, and the integer part, where the control point adjustment coefficient is a dimensionless constant. The adjusted model is regenerated and updated to the rendering interface.

[0072] If occlusion or abnormal dwell time is detected, the text marker position is adjusted. The new position is calculated as a fixed distance offset from the current gaze point coordinates, prioritizing unobstructed areas (verified via raycasting). The adjusted text marker is then reattached to the 3D model, and the rendering interface is updated.

[0073] The representative decision maker memory sample is updated by incorporating interactive evidence, including the number of occlusions, the duration of abnormal dwell times, the adjusted height, the number of control points, and the location of text markers, along with the entity ID and representativeness coefficient, and writing it into the representative decision maker's memory sample. The memory sample updates the small sample positive-negative pair set, triggering the contrastive learning ranking in step S2, recalculating the representativeness coefficient, and updating the feature library, completing the iteration of screening and modeling.

[0074] Step S5 continuously collects user browsing trajectories, detects occlusion and abnormal pauses, triggers fine-tuning of the 3D model height and text marker positions, and writes the interaction evidence into the memory samples of the representative decision-maker, completing the screening and modeling iteration. This step achieves adaptive optimization of the interactive interface, solves the problems of occlusion and poor information presentation, improves the immersive experience and information transmission efficiency of the user experience, and enhances the dynamic adaptability of the overall process.

[0075] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0076] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0077] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0079] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A design method for transforming planar information into a three-dimensional representation of an interactive reading interface, characterized in that, Including the following steps: S1: Receive the region name text and satellite imagery, complete semantic segmentation and spatial registration, generate a text pixel mapping table and write it into the data pool; S2: Candidate entities are formed based on the mapping table. The representative decision-maker calculates the main axis fit rate and the piecewise reproducibility, outputs the representativeness coefficient, and selects entities to write into the feature library. S3: Read the feature library and representativeness coefficients, set geometric parameters for the selected entities to generate a 3D model, and bind the corresponding text tags; S4: Loads a 3D model and binds eye tracking. Text descriptions are displayed synchronously when the user looks at the target. Rotation, scaling and transparency adjustment are supported. S5: Collect browsing trajectory and compare it with text markers and representativeness coefficients. When occlusion or abnormal stops occur, adjust the model and marker positions, write back to the feature library and iteratively optimize.

2. The design method for generating a concrete interactive reading interface from planar information in three dimensions according to claim 1, characterized in that, Step S1: The system receives region name text and satellite imagery, extracts place name sets and descriptive phrases through word segmentation and named entity recognition, queries the geographic coordinates of the place name center using geocoding and converts them into image pixel coordinates through coordinate transformation, extracts a set of pixel coordinates with a fixed radius centered on the pixel coordinates, and generates a text-pixel mapping table containing place name identifiers, pixel coordinate sets, descriptive phrases and source indexes, which is then written into the raw data pool.

3. The design method for generating a concrete interactive reading interface by transforming planar information into three dimensions according to claim 2, characterized in that, Step S2: The candidate entity set is generated by reading the text pixel mapping table from the original data pool, extracting the boundary pixel coordinates through edge detection and region segmentation algorithms, calculating the principal axis fit rate and the piecewise reproducibility, constructing a small sample positive and negative pair set, inputting the principal axis fit rate and the piecewise reproducibility, generating the discriminative embedding vector through pairwise comparison learning, and calculating the representativeness coefficient based on the boundary margin maximization criterion.

4. The design method for generating a concrete interactive reading interface by converting planar information into three dimensions according to claim 3, characterized in that: Set an inclusion threshold to determine the entities to be included, sort them in descending order of representativeness coefficient, and write the included entities, along with their boundary pixel coordinates, descriptive phrases, and entity numbers, into the feature library.

5. The design method for generating a concrete interactive reading interface by converting planar information into three dimensions according to claim 3, characterized in that: For each candidate entity, the morphological principal axis description is extracted from the description phrase and transformed into the expected principal axis direction and expected curvature direction. The skeleton is extracted and refined based on the boundary pixel coordinates to calculate the total arc length of the skeleton. The angle deviation between the local tangential direction and the expected principal axis direction is calculated point by point along the skeleton to determine the principal axis fitting rate. The Douglas-Puk algorithm is used to simplify the boundary with a fixed error tolerance to obtain the minimum number of line segments, and the piecewise reproducibility is calculated by linear normalization using empirical upper and lower bounds.

6. The design method for generating a concrete interactive reading interface by transforming planar information into three dimensions according to claim 4, characterized in that, Step S3: The modeling module is invoked to read the selected entities and related data from the feature library. Based on the descriptive phrase, it determines whether to use stretching modeling or surface modeling. For stretching modeling, the stretching height is set based on the boundary pixel coordinates. For surface modeling, a non-uniform rational B-spline surface is fitted and the number of control points is set. The subdivision level and texture resolution are allocated according to the representativeness coefficient to generate a 3D model containing geometric data and texture data.

7. The design method for generating a concrete interactive reading interface by converting planar information into three dimensions according to claim 6, characterized in that: By binding text tags extracted from descriptive phrases to the geometric center of the model and embedding source indexes and entity numbers, visual data support is provided for interactive rendering.

8. The design method for generating a concrete interactive reading interface by transforming planar information into three dimensions according to claim 7, characterized in that, Step S4: The 3D model is loaded into the rendering interface through the graphics rendering engine, and the eye-tracking device is bound to capture the coordinates of the user's gaze point. When the threshold is exceeded, text labels are displayed on the target facade. The system responds to rotation, scaling and transparency adjustments to ensure that the viewing angle is synchronized.

9. The design method for generating a concrete interactive reading interface by transforming planar information into three dimensions according to claim 8, characterized in that, Step S5: The user browsing trajectory is collected from the interaction log, including gaze coordinates, gaze duration, operation type, and entity number. Occlusion or abnormal dwell time is detected by ray casting. The height of the 3D model or the number of surface control points is adjusted based on the number of occlusions and the duration of abnormal dwell time. The text marker position is updated to the non-occluded area with a fixed distance offset from the gaze point. The representativeness coefficient of the interaction evidence is associated with the entity number and written into the memory sample. Small sample comparison learning is triggered to rank and update the feature library, completing the screening and modeling iteration.

Images