Data processing method, device, medium and system based on spatio-temporal index

Abstract

The invention discloses a space-time backtracking and AI content reconstruction method, device and system based on an XYZT index. The method comprises the following steps of: retrieving a target slice in a second level from a massive discrete spatio-temporal data set by utilizing a composite spatio-temporal fingerprint index (such as an exclusive or hash structure); according to the spatial and temporal distribution discreteness of fragmented data, a generative neural network model is adopted to execute feature interpolation and scene completion in a latent space, spatial and temporal faults are eliminated, and a continuous historical scene is generated; at the same time, the eye movement tracking technology is combined to realize fixation point priority rendering, and differential privacy noise is introduced to protect individual trajectories. According to the method, the technical problem that high-fidelity historical scenes are difficult to restore by multi-source discrete data is effectively solved, and immersive backtracking and interaction of long-term historical time and space are supported.

Description

Technical Field

[0001] This invention relates to the fields of big data processing, augmented reality (AR) and artificial intelligence, and specifically to a data processing method, apparatus, medium and system based on spatiotemporal indexing. Background Technology

[0002] With the widespread adoption of the Internet of Things (IoT), mobile internet, and various portable sensing devices, location-based services (LBS) have penetrated all aspects of social life. The vast amounts of information generated by users, devices, or sensors at specific spatiotemporal coordinates constitute massive spatiotemporal datasets that record the trajectories of social activities. How to effectively trace back and reconstruct scenes from specific historical moments has become a core research topic in the fields of big data mining, digital twins, and virtual reality.

[0003] However, existing spatiotemporal retrospective technologies (such as traditional urban video surveillance systems and GPS trajectory playback systems) have significant technical bottlenecks and shortcomings when dealing with such needs, specifically as follows:

[0004] (1) Data acquisition continuity limitations and resource redundancy Traditional urban monitoring systems rely on fixed-point cameras for continuous video acquisition. This "full recording" mode not only generates massive amounts of redundant data, placing a huge burden on storage and network bandwidth, but more importantly, once the target object enters the camera's blind spot or is in an area without monitoring coverage, the continuity of information is completely interrupted, resulting in a serious "spatiotemporal discontinuity" in the backtracking scene.

[0005] (2) Lack of Processing Capability for Discrete and Heterogeneous Data In mobile social or distributed sensing scenarios, data (such as the social broadcast described in the applicant's patent application filed on the same day entitled "An Information Distribution Method, System, Device and Medium Based on Dynamic Geofencing") is often generated by random events and has highly discrete and fragmented characteristics. Existing monitoring playback systems are based on "streaming media" logic and lack the ability to semantically associate and visually stitch together discrete data points with cross-source and cross-time periods, and cannot restore the scattered records generated by different users at different moments into a logically consistent continuous scene. In this specification, "discrete spatiotemporal distribution" refers to the state in which data slices are discontinuous on the time axis (with time gaps) and non-aligned on the spatial coordinates.

[0006] (3) Limitations of the backtracking dimension: Current geographic information backtracking focuses on two-dimensional plane (XY) and time axis (T), and lacks the ability to refine perception and index the height dimension (Z). In complex urban spaces such as multi-story buildings and three-dimensional transportation, the lack of Z-axis information indexing will lead to positioning offset of the backtracking scene, which cannot meet the needs of high-precision scene heavy loads such as augmented reality (AR).

[0007] (4) The contradiction between privacy protection and data availability: Existing monitoring systems often use simple masking or desensitization methods during the backtracking process, which often destroys the integrity and realism of the scene. How to protect the privacy of individuals' precise trajectories while preserving the macro-characteristics of group activities (such as heat distribution and event popularity) is a challenge that has not yet been effectively solved in current backtracking algorithms.

[0008] (5) High-power rendering and hardware adaptation bottlenecks: Reconstructing complex 3D historical scenes places extremely high demands on the computing power of terminal devices. On head-mounted displays or handheld mobile devices, full-area high-definition rendering can lead to serious power consumption problems and latency, directly affecting the smoothness of the backtracking experience, and lacking a rendering optimization mechanism for user visual attention.

[0009] In summary, existing technologies struggle to solve the problems of accurate retrieval and high-fidelity, low-power reconstruction of multi-source discrete data in complex four-dimensional space (XYZT). Therefore, developing a spatiotemporal retrospective system capable of interpolating fragmented data using generative artificial intelligence while balancing privacy protection and rendering efficiency has significant technological and commercial value. Summary of the Invention

[0010] 3.1 Purpose of the Invention The purpose of this invention is to overcome the limitations of existing technologies in accurately retrieving and reconstructing multi-source discrete data in a complex four-dimensional space (XYZT) with high fidelity and low power consumption. Addressing the shortcomings of existing monitoring systems, such as "spatiotemporal discontinuities," lack of discrete heterogeneous data processing capabilities, insufficient Z-axis sensing, and the conflict between privacy protection and data availability, this invention provides a method, device, storage medium, and system for spatiotemporal backtracking and AI content reconstruction based on XYZT indexing. This invention aims to utilize generative artificial intelligence to perform semantic interpolation on fragmented data, restoring continuous and vivid historical spatiotemporal scenes while protecting privacy.

[0011] 3.2 Technical Solution To achieve the above objectives, this invention provides a data processing method based on spatiotemporal indexing, comprising the following steps: Step 1, Multidimensional spatiotemporal request acquisition: Receiving an access request initiated by a terminal containing a target spatiotemporal range, wherein the spatiotemporal range at least covers the geographic location dimension (XYZ) and the time dimension (T). Step 2, Composite index retrieval: Retrieving matching data slices from discrete spatiotemporal associated datasets using a pre-constructed composite spatiotemporal fingerprint index (i.e., XYZT composite index). The composite index is constructed by performing a bitwise XOR hash operation on the spatial encoding of the geographic location dimension (such as GeoHash or Hilbert curve encoding) and the stamp information of the time dimension, thereby achieving fast location of discrete heterogeneous data in four-dimensional space with O(1) complexity. Step 3, AI semantic interpolation and reconstruction: For the discreteness of the retrieved data slices in spatiotemporal distribution, performing feature interpolation in the latent space using a pre-defined generative neural network model (such as LatentDiffusion Model). The system extracts endpoint feature vectors from the slices and iteratively denoises and completes them in the latent space to eliminate spatiotemporal discontinuities between data slices, generating a logically continuous target spatiotemporal scene. The fourth step is scene output and optimization: the reconstructed target spatiotemporal scene is output to the terminal. Before output, the system can perform differential privacy processing (such as injecting Laplacian noise) based on social distance; when the terminal is a head-mounted display device, the system uses eye-tracking data to perform high-resolution rendering only on the core area of ​​the gaze point, and downsampling processing on the surrounding areas to reduce rendering power consumption.

[0012] In addition, the present invention also provides a spatiotemporal backtracking system, comprising a data acquisition module, a cloud server and a terminal display device, wherein the cloud server is equipped with the aforementioned composite index engine and AI reconstruction model.

[0013] 3.3 Beneficial Effects Compared with the prior art, the present invention has the following significant beneficial effects:

[0014] 1. Ultra-fast and accurate retrieval: By using the XYZT XOR hash index, the indexing problem of massive discrete slices in four-dimensional space is solved, the retrieval response time of the dataset of hundreds of millions is shortened to the second level, and the positioning offset problem of Z-axis height information is effectively solved.

[0015] 2. Spatiotemporal continuity reconstruction: By using generative AI models to perform semantic interpolation on fragmented data, the "spatiotemporal discontinuity" caused by traditional monitoring blind spots is filled, realizing a qualitative change from "discrete photos" to "continuous historical scenes".

[0016] 3. Balancing rendering efficiency and privacy: By using foveated priority rendering technology, the rendering power consumption of head-mounted devices is reduced by 30%-50%; by injecting Laplacian noise, the privacy of individual precise trajectories is effectively protected while restoring the thermal distribution of macroscopic scenes. Attached Figure Description

[0017] Figure 1 is a schematic diagram of the data processing flow provided in the embodiment of the present invention, which mainly shows the complete closed-loop process from the collection of multi-source heterogeneous data (including real-time social broadcast slices and existing historical data slices), the construction of XYZT composite index to AI spatiotemporal consistency reconstruction and bidirectional interaction;

[0018] Figure 2 is a schematic diagram of the XYZT composite index and retrieval process provided in the embodiment of the present invention, which focuses on the process of generating an index and locating heterogeneous slices by using the XOR-Hash algorithm to combine spatial coordinates and time features;

[0019] Figure 3 is a logical block diagram of the AI ​​spatiotemporal consistency reconstruction model provided in the embodiment of the present invention, which shows the technical path from semantic feature extraction to diffusion model iterative denoising to generate continuous scenes;

[0020] Figure 4 is a flowchart of a privacy coordinate perturbation processing method based on social distance provided in an embodiment of the present invention;

[0021] Figure 5 is a schematic diagram of the VR terminal gaze priority rendering logic provided in an embodiment of the present invention;

[0022] Figure 6 is a schematic diagram of the composition structure of a data processing device based on spatiotemporal index provided in an embodiment of the present invention. Detailed Implementation

[0023] [Explanation of Cross-Patent Associations]

[0024] The [data processing method, apparatus, medium, and system based on spatiotemporal indexing] described in this application is a component of a complete spatiotemporal information processing system. This system includes:

[0025] The distribution side (related patent / the applicant's patent application filed on the same day entitled "An Information Distribution Method, System, Device and Medium Based on Dynamic Geofencing") is responsible for the accurate push and persistent storage of discrete information based on dynamic geofencing and motion vectors;

[0026] Interactive side (related patent / the applicant's patent application filed on the same day entitled "A Remote Presence Control Method, Device and System for Heterogeneous Device Collaboration"): responsible for realizing the pose control and heterogeneous mapping of the controlled subject (including physical space entities and virtual image terminals) by the remote terminal based on social distance and permissions;

[0027] Backtracking side (this patent / a data processing method, device, medium and system based on spatiotemporal index): responsible for AI scene reconstruction and backtracking based on the discrete slices generated by the aforementioned patent based on the XYZT index.

[0028] The aforementioned patents are interconnected, forming a closed-loop spatiotemporal information processing framework that encompasses real-time distribution, interactive control, and historical reconstruction.

[0029] 5.1 Spatiotemporal Data Persistence and Discrete Slice Management

[0030] The data processing flow provided in this embodiment of the invention is shown in Figure 1. The system collects real-time social broadcast slices from audience terminals through an open interface and accesses existing historical data slices from an external database; after performing normalization preprocessing on the above-mentioned multi-source heterogeneous data and processing it using the XYZT composite index module, it is stored in a discrete spatiotemporal correlated dataset. Referring to Figure 1, when performing a backtracking task, the system receives a backtracking request containing geographic coordinates, altitude (Z), and time (T), and performs the following core steps:

[0031] 5.1.1 Sliced ​​Storage of Discrete Social Broadcasting

[0032] It is important to emphasize that the data source processed by this invention has a significant 'random event triggering' characteristic. Unlike traditional monitoring systems that perform full, continuous linear recording, the discrete data slices (such as photos, short videos, sensor poses, etc.) in the dataset of this system are generated instantaneously based on random social behaviors or user broadcast commands occurring in physical space. This randomness leads to a high degree of discretization and fragmentation of the original data in terms of both time axis and spatial distribution, thus forming the technical premise for the subsequent spatiotemporal consistency reconstruction performed by the AI ​​model in this invention.

[0033] Referring to Figure 2, this system calls the storage interface of the aforementioned spatiotemporal correlated dataset (in one embodiment, this interface can interface with the data interface of the applicant's patent application filed on the same day, entitled "An Information Distribution Method, System, Device and Medium Based on Dynamic Geofencing"). To achieve rapid alignment of the aforementioned real-time dynamic data with static stock data and other discontinuous data, the following algorithm is executed:

[0034] 1. Multidimensional dimensionality reduction coding: Map latitude, longitude and altitude to high-dimensional Z-Order curve coding.

[0035] 2. Composite hash index: Construction This index ensures that the system can accurately extract target data from hundreds of millions of discrete slices (including ancient data spanning time and space).

[0036] During the index construction process, a hash XOR operation is used. Rather than a simple weighted summation or concatenation, the technical advantage lies in the fact that the XOR operation can preserve the non-linear correlation between spatial encoding and timestamps to the greatest extent while maintaining low computational complexity, effectively reducing the hash collision rate in large-scale spatiotemporal datasets. Furthermore, the hash XOR operation achieves uniform load balancing (Sharding) of discrete data slices in a distributed storage cluster by sharding high-dimensional spatial features and timeline features at the bit level. Compared to traditional sequential storage, this approach effectively avoids the problem of data storage hotspots in specific popular geographical areas (such as city centers) or peak time periods, giving the system stronger horizontal scalability and parallel read speed when performing scenario reorganization. Especially when handling high-concurrency XYZT four-dimensional concurrent requests, this indexing method can significantly improve the response speed of multi-dimensional joint retrieval, achieving accurate location of target historical fragments within milliseconds.

[0037] 5.1.2 Preprocessing of Existing Historical Data Slices and Heterogeneous Slices The discrete spatiotemporal correlation dataset described in this invention has open data access attributes. Its data composition includes, but is not limited to:

[0038] 1. Real-time dynamic data slicing: Social broadcast, multimedia slices and sensor attitude data with high-dimensional coordinate attributes generated by the associated system (such as the associated patent "A Method, System, Device and Medium for Information Distribution Based on Dynamic Geofencing" filed by the applicant on the same day) when performing dynamic tasks;

[0039] 2. Existing historical data slices: Digital historical slices are generated by reverse engineering historical documents, ancient maps, 3D scanning models of ancient buildings and archaeological measurement data, and then performing semantic alignment and spatiotemporal indexing according to the XYZT dimensions described in this invention.

[0040] Technical effects: By converting the aforementioned existing data into a unified XYZT index format, the system can not only accurately reproduce real-time scenes, but also use AI generative algorithms to perform logical deduction, scene fitting, and visual reconstruction of long-term historical spatiotemporal data that lack real-time records.

[0041] 5.2 XYZT Composite Index and Retrieval Algorithm

[0042] 5.2.1 Implementation of XOR-Hash Algorithm

[0043] This invention uses XOR hashing to construct a composite index, and its specific retrieval process is shown in Figure 2. Specifically, the system extracts spatial coordinate codes respectively. With time characteristics A unique composite index is generated by bitwise XOR operation. This allows for the precise location of target data buckets within massive discrete datasets.

[0044] The XOR hashing mechanism used in this invention aims to solve the problem of accurately locating massive discrete fragments on a multidimensional spatiotemporal axis.

[0045] The system constructs composite indexes using the following formula.

[0046]

[0047] in:

[0048] The coordinate encoding representing the geographic location dimension is mapped to a one-dimensional integer value through Geohash or Hilbert curve;

[0049] Timestamp information representing the generation of discrete slices;

[0050] and This is a preset feature mapping function used to align spatial and temporal features to the same bit width;

[0051] This represents a bitwise XOR operation;

[0052] This represents the total capacity of the hash bucket.

[0053] By performing an XOR operation between high-dimensional spatial coordinates and discontinuous timestamps, the system can map randomly distributed fragmented data to unique hash slots. This indexing method ensures that the system can retrieve slices from massive discrete datasets at the millisecond level when performing scene reconstruction, providing the underlying technical support for real-time stitching and scene reconstruction of large-scale discrete data.

[0054] 5.2.2 Heterogeneous Data Retrieval and Retrieval Process

[0055] Based on the generated composite index The system executes the following search logic:

[0056] 1. Data Bucket Location: Utilizing hash indexes to perform operations in discrete spatiotemporal correlated datasets. High-speed addressing with low complexity to locate data buckets within the target spatiotemporal range.

[0057] 2. Slice set extraction: Retrieve all matching heterogeneous data slices in parallel from the target bucket, including real-time social broadcast slices and existing historical data slices.

[0058] 3. Output: The retrieved discrete slice set is used as input parameters and passed to the AI ​​reconstruction model described in Section 5.3.

[0059] 5.3 AI Spatiotemporal Consistency Reconstruction Model

[0060] The logical architecture of the AI ​​reconstruction model described in this invention is detailed in Figure 3. This model, through core steps such as semantic feature extraction, feature interpolation based on a diffusion model, and iterative denoising, stitches together retrieved non-visual continuous slices into a continuous scene. Specifically:

[0061] 5.3.1 Overview of Technical Principles

[0062] The core technological contribution of the data retrieval and reconstruction process described in this invention lies in processing highly fragmented, spatiotemporally distributed discrete data slices generated by multi-source heterogeneous terminals. Unlike traditional continuous spatiotemporal reproduction technologies based on streaming media or video playback, this invention uses a generative neural network model to perform generative interpolation and completion of spatiotemporal faults between discrete slices at the semantic level. This means that the system does not simply piece together existing images, but rather generates logically consistent scenes in a high-dimensional latent space based on the semantic logic of existing fragments, thus achieving a fundamental leap from 'discrete fragments' to 'continuous spatiotemporal data'.

[0063] 5.3.2 Semantic Feature Extraction from Discrete Slices

[0064] After obtaining the target slice set through the indexing mechanism described in Section 5.2, the system needs to perform normalization feature representation on these heterogeneous data. The data source described in this invention has significant "random event triggering" characteristics, and the original data exhibits high discretization and fragmentation in terms of time axis and spatial distribution.

[0065] The system uses a pre-trained deep neural network encoder (such as CLIP or visual Transformer) to extract features from the multimodal information (such as text descriptions, scene images, and spatial coordinate labels) contained in each discrete slice, mapping unstructured social fragments into semantic tensors of a high-dimensional latent space.

[0066] Input example: Retrieve time information from search results eigenvectors With time eigenvectors , which serve as the starting and ending anchor points for subsequent interpolation operations.

[0067] 5.3.3 Diffusion Model Interpolation and Iterative Denoising Logic

[0068] It should be noted that the spatiotemporal interpolation described in this embodiment has significant technical differences from traditional continuous video frame interpolation. Traditional video interpolation is based on a sequence of images with strong visual coherence acquired by the same sensor; while the interpolation object described in this application is a "data slice" from different acquisition terminals, different shooting angles, and with a highly discrete and jumpy time span.

[0069] Therefore, this algorithm introduces semantic alignment constraints of multi-source heterogeneous features through a cross-attention mechanism when performing denoising autoencoder operations. This enables the reconstruction of logically consistent dynamic scenes even in the absence of pixel-level continuous references. The specific steps are as follows:

[0070] 1. Condition-guided injection

[0071] To restore the moment and The system utilizes a Latent Diffusion Model to perform generative completion at the feature level, representing a continuous visual evolution flow between features. This involves starting with the initial features... Endpoint characteristics and the proportion of target time steps Input the cross-attention constraint into the denoising network.

[0072] 2. Iterative denoising operation

[0073] Random noise in the latent space Performing back diffusion, the system iteratively denoises Gaussian noise to recover continuous frames that conform to semantic logic. The core iterative formula is as follows:

[0074]

[0075] in:

[0076] This represents random noise in the potential space.

[0077] This is the normalized time step ratio, used to finely control the position of the reconstructed frame on the spatiotemporal axis;

[0078] This is the noise prediction function after learning;

[0079] The variance scheduling parameter during the diffusion process;

[0080] The cumulative variance factor is calculated using the following formula: .

[0081] Through the aforementioned multiple rounds of iterative calculations, the system can eliminate the original data gaps caused by random events, in discrete... and A continuous transition sequence that satisfies spatiotemporal consistency is generated between these elements.

[0082] 3. Enhanced Interaction Loop

[0083] Furthermore, the target scene reconstructed by this invention possesses significant "two-way interactivity." Since the reconstruction process is performed at the semantic level, the generated scene is not a static pixel stack, but a dynamic environment that supports secondary interaction. During the rewind process, users can again control their virtual avatar to enter the reconstructed scene using the pose mapping algorithm described in the applicant's patent application filed on the same day, entitled "A Remote Presence Control Method, Device, and System for Heterogeneous Device Collaboration," achieving asynchronous interaction and a social loop with the "past time and space."

[0084] 5.3.4 Semantic consistency smoothing, visual rendering and interaction closed loop

[0085] Before performing visual rendering, to ensure that the generated scene is geometrically consistent with physical space, the system can introduce Neural Radiation Field (NeRF) technology. Specifically, the system uses multi-angle discrete slices retrieved within the same spatiotemporal range as sparse viewpoint inputs to train an instantaneous NeRF network to construct an implicit 3D volume representation at that moment. Image features from arbitrary new viewpoints are extracted through volume rendering and used as geometric constraints for subsequent diffusion model interpolation, thereby ensuring the rigidity and realism of the reconstructed scene in 3D spatial structure.

[0086] To eliminate potential visual flickering or logical gaps in AI-generated images, the system performs the following post-processing steps:

[0087] 1. Motion vector alignment: Calculate the optical flow between adjacent reconstructed frames to force the object's motion trajectory to conform to physical continuity.

[0088] 2. Super-resolution enhancement: Generative adversarial networks (GANs) are used to repair and upsample the details of the completed intermediate frames to ensure the clarity of the output scene.

[0089] The target scene reconstructed by this invention possesses significant "two-way interactivity." Since the reconstruction process occurs at the semantic level, the reconstructed scene supports secondary interaction. During the rewind process, users can again control a virtual avatar to enter the reconstructed scene through the pose mapping algorithm described in the applicant's patent application filed on the same day, entitled "A Remote Presence Control Method, Device, and System for Heterogeneous Device Collaboration," achieving asynchronous interaction with the "past time and space." This dynamic real-time nature proves that this invention generates a "living scene" with spatiotemporal continuity, rather than a pre-rendered stack of digital slices.

[0090] 3. Reconstruction Result Output: The final output is a complete set of historical scene images that satisfy spatiotemporal continuity and have interactive responsiveness, realizing the visual closed-loop reconstruction of discrete social fragments.

[0091] 5.4 Rendering Optimization and Privacy Protection

[0092] 5.4.1 Privacy Coordinate Perturbation Processing Referring to Appendix 4, to address the privacy leakage risk associated with high-precision backtracking, the system performs differential privacy processing before outputting data:

[0093] 1. Social distance calculation: The system first determines the degree of social connection between the initiator of the backtracking request and the object being backtracked.

[0094] 2. Laplace noise injection: For "long-distance" requests from non-friends or those with low privileges, the system injects Laplace noise into the individual coordinates while preserving the macroscopic thermal distribution characteristics, generating fuzzy anonymous coordinates, thereby achieving a "usable but invisible" privacy protection effect.

[0095] 5.4.2 For gaze-priority rendering, refer to Appendix 5. For head-mounted display devices with limited computing power, the system performs rendering optimization:

[0096] 1. Gaze Point Acquisition: The device's built-in eye tracker acquires the user's gaze coordinates in the reconstructed scene in real time. .

[0097] 2. Priority Rendering Queue: The system performs 4K high-resolution rendering only on the foveated region, while downsampling is applied to the surrounding non-foveated regions. This mechanism significantly reduces the GPU's rendering load and power consumption while maintaining the subjective visual experience.

[0098] 5.5 Cross-patent collaboration, device and medium implementation

[0099] 5.5.1 Cross-Patent Collaboration Logic

[0100] Furthermore, when reconstructing a scene, the system can also receive and reload control flow data generated at historical moments by the related patent, "A Remote Presence Control Method, Device, and System for Heterogeneous Device Collaboration," filed on the same day by the applicant. For example, the system can not only reconstruct the visual image of a historical street, but also fully reproduce the pose trajectory of the "virtual user" (i.e., the asymmetric virtual presence carrier) and its interaction with pedestrians at the time, thereby achieving a full retrospective of "virtual-real combined social events." At the same time, the semantic features retrieved by the system can refer to the broadcast attributes recorded in the related patent, "An Information Distribution Method, System, Device, and Medium Based on Dynamic Geofencing," filed on the same day by the applicant, to restore the social popularity at historical moments.

[0101] 5.5.2 Devices and Media

[0102] This invention also discloses a data processing device based on spatiotemporal indexing. Its hardware structure is shown in Figure 6. The device includes a processor and a memory. The memory stores a computer program, and when the processor executes the program, it performs the following: receiving a backtracking request for target spatiotemporal coordinates; retrieving discrete data slices and matching them according to the XYZT composite index; reconstructing the scene using an AI model and performing motion vector smoothing. This invention also provides a computer-readable storage medium storing instructions for implementing the above steps.

[0103] 5.6 Typical Application Examples

[0104] 5.6.1 Example 1: Remote Immersive Time and Space Tourism and Social Interaction

[0105] In this embodiment, the system allows users to "travel through time and space" in a non-on-site environment.

[0106] User request: User D is at home and logs into the system through a VR headset. In the 3D map, he selects a famous historical square (XYZ) thousands of miles away and requests to go back to the time period (T) of "last year's New Year's Eve".

[0107] Reconstruction Logic: The system retrieves discrete social broadcast clips uploaded by hundreds or thousands of users in the square at the time. Using the AI ​​diffusion model described in Section 5.2, these fragmented photos and videos are "stitched" into a continuous 360-degree 3D environment for interactive engagement.

[0108] Interactive Experience: User D can "stroll" through last year's New Year's Eve celebration from the comfort of their home, without having to be physically present. Based on semantic features recorded in the patent application filed on the same day, "An Information Distribution Method, System, Device and Medium Based on Dynamic Geofencing," the system recreates the crowd's enthusiasm and cheers at the time, and allows User D to engage in real-time social interaction with historical figures or virtual avatars within the reconstructed scene.

[0109] 5.6.2 Example 2: Remote Retrospective Analysis and Auditing of Engineering Accidents in a Collaborative Remote Location

[0110] In this embodiment, the system is used to solve the problem of cross-regional project management.

[0111] Scenario: Headquarters experts in City A need to conduct a post-mortem analysis of a steel reinforcement collapse accident at a construction site in City B. At the time of the accident, only a few workers' mobile devices recorded scattered footage.

[0112] Backtracking operation: Experts send a backtracking request from the headquarters office, and the system uses the XYZT composite index to pinpoint the precise spatial coordinates and time of the incident.

[0113] Technical reconstruction: The system uses AI algorithms to perform depth alignment on discrete images uploaded by several workers and automatically completes the force structure evolution in the camera blind spot.

[0114] Audit results: Experts can observe the microscopic evolution of structural fractures by adjusting the time axis (T) through remote terminals, achieving a comprehensive and multi-perspective visual reproduction of the target spatiotemporal scene, thereby effectively improving the accuracy and efficiency of cross-regional engineering accident review.

[0115] 5.6.3 Example 3: Cloud-based Urban Historical Change Survey

[0116] In this embodiment, the system serves as a decision support tool for urban planning or real estate development.

[0117] Operational logic: The planner selects an old community in the city that is to be demolished on the cloud platform.

[0118] Data retrieval: The system retrieves all discrete spatial data accumulated in the community over the past ten years (including social check-ins, street view slices, user broadcasts, etc.).

[0119] Reconstruction and Comparison: The system uses foveated rendering technology to dynamically reconstruct the street scene appearance of a point in different years based on the planner's mouse click position.

[0120] Decision-making value: Planners can observe the historical evolution of population density in a community by adjusting time parameters without leaving their homes (reconstructed through the frequency distribution of social broadcasts), thereby achieving precise commercial site selection planning.

[0121] 5.6.4 Example 4: Remote Historical Spatiotemporal Immersion and Cultural Interaction

[0122] In this embodiment, the system reconstructs multi-source cultural data under specific historical spatiotemporal coordinates (XYZ, T) to achieve virtual interaction across centuries.

[0123] Scenario: Users in different locations access the coordinates of "Chang'an Street in the Tang Dynasty" through VR / AR devices.

[0124] Data retrieval and reconstruction: The system retrieves historical building restoration parameters, visual features of unearthed cultural relics, and descriptions of people's living conditions from historical records stored in a discrete spatiotemporal correlated dataset. Using the AI ​​generative algorithm described in Section 5.2 of this invention, this fragmented historical information is interpolated and reconstructed into a 3D ancient street scene with high spatial consistency.

[0125] Interactive experience: The system identifies historical figures (such as ancient literati) in the reconstructed scene and, based on the virtual personality construction technology described in claim 5 of this invention, combines a large language model to endow them with specific personality traits and language styles.

[0126] Furthermore, the virtual personality construction process includes: the system first extracts the associated historical semantic tags based on the identity identifier of the target historical figure; then it retrieves and matches the personality parameters (such as language style, historical stance, etc.) corresponding to the tags from a preset knowledge base; finally, it loads the personality parameters into the prompt word template of the Large Language Model (LLM) to generate interactive response content that conforms to the specific historical background and character setting logic.

[0127] Application Value: Users can engage in real-time dialogue and interaction with recreated 3D dynamic historical figures in a virtual space (such as co-creating poems or discussing historical events). This approach breaks through the limitations of authentically recorded data, achieving an immersive reconstruction of "deep historical spacetime" that cannot be directly collected through AI simulation.

[0128] VI. Conclusion

[0129] It should be understood that this specification is only a preferred embodiment of the present invention. For those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention (for example, replacing the diffusion model with a GAN generative adversarial network, or replacing XOR hashing with space-filling curve indexing). These improvements and modifications should also be considered within the scope of protection of the present invention.

[0130] Furthermore, the terms "terminal" and "server" mentioned in the embodiments of this invention are only used to distinguish different device objects and do not limit their specific hardware form. Those skilled in the art will understand that any computing device with data processing and communication capabilities can serve as a node in this system. Content not described in detail in this specification belongs to prior art known to those skilled in the art.

Claims

1. A data processing method based on spatiotemporal indexing, characterized in that, Includes: Acquisition Request: Receiving an access request containing a target spatiotemporal range, wherein the spatiotemporal range includes at least a geographic location dimension and a time dimension; Data retrieval: Based on the access request, a matching data slice is retrieved from the stored discrete spatiotemporal associated dataset using a composite spatiotemporal fingerprint index containing spatial and temporal features; Scene reconstruction: Feature interpolation and scene completion are performed on the retrieved spatiotemporally distributed discrete data slice using a generative neural network model; Specifically, this includes extracting the endpoint feature vectors of the data slice and performing interpolation operations in the latent space to eliminate spatiotemporal discontinuities and generate a target spatiotemporal scene; Scene output: The target spatiotemporal scene is output to the terminal.

2. The method according to claim 1, characterized in that: The target spatiotemporal range also includes a height dimension (Z-axis) for information anchoring and backtracking in three-dimensional space; the composite spatiotemporal fingerprint index is a unique identifier constructed by performing a hash XOR operation on the coordinate encoding of the geographic location dimension and the stamp information of the time dimension.

3. The method according to claim 1, characterized in that: The consistency processing includes using an artificial intelligence model to perform semantic-level feature interpolation, scene completion, or spatial feature alignment on discrete data slices from multiple heterogeneous terminals with discrete spatiotemporal distribution; at the same time, using a visual positioning system (VPS) to obtain the geometric model of the current physical space, or using a neural radiation field (NeRF) algorithm to extract multi-angle media data, and then performing coordinate registration between the generated reconstructed scene and the physical space.

4. The method according to claim 1, characterized in that: The discrete spatiotemporal association dataset includes real-time interactive slices generated by the association system when performing dynamic tasks, as well as existing slices generated by converting historical documents or archaeological image data; and before outputting the target spatiotemporal scene, the system performs differential privacy processing on the data by coordinate perturbation or Laplace noise injection according to access permissions or social distance.

5. The method according to claim 1, characterized in that, Also includes: Identify entity objects in the target spatiotemporal scene and retrieve historical data slices associated with the entity objects; Using a natural language processing model, semantic tags and behavioral feature parameters are extracted from the historical data slices to construct a virtual personality corresponding to the entity object; The virtual personality is loaded into a large language model to drive the entity object to respond to the interactive instructions of the access terminal in the target spatiotemporal scene.

6. The method according to claim 1, characterized in that: The target spatiotemporal scene is presented on a smart terminal, which includes a head-mounted display device, a handheld mobile device, or an in-vehicle display system. When the smart terminal is a head-mounted display device, the method further includes using an eye tracker to collect the coordinates of the gaze point and performing high-priority rendering processing only on the 3D spatial region covered by the gaze point.

7. A data processing apparatus based on spatiotemporal indexing, comprising a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the method as described in any one of claims 1 to 6.

8. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the method as claimed in any one of claims 1 to 6.

9. A spatiotemporal retracing system, characterized in that, include: The data acquisition module is used to acquire spatiotemporal correlated data; A cloud server is used to store discrete spatiotemporal correlated datasets and perform the scene reconstruction as described in claim 1; And terminal display devices, used to present the target spatiotemporal scene.

Images